Public Accounts Committee

Oral evidence: DWP Annual Report & Accounts 2022-23, HC 1339

Monday 18 September 2023

Ordered by the House of Commons to be published on 18 September 2023.

Watch the meeting

Members present: Dame Meg Hillier (Chair); Mr Jonathan Djanogly; Mrs Flick Drummond; Peter Grant; Ben Lake.

Work and Pensions Committee member present: Sir Stephen Timms, Chair.

Gareth Davies, Comptroller and Auditor General, Joshua Reddaway, Director, National Audit Office, and Marius Gallaher, Alternate Treasury Officer of Accounts, were in attendance.

Questions 1-108

Witnesses

I: Bozena Hillyer, Director for Counter Fraud Compliance and Debt, Department for Work and Pensions, Peter Schofield CB, Permanent Secretary, DWP, Neil Couling, Director General Change and Resilience, and Universal Credit SRO, DWP, Catherine Vaughan, Finance Director General, DWP, and Richard Hawthorn, Director, Operational Excellence, Customer Service Group, HMRC.

Report by the Comptroller and Auditor General

Report on Accounts: Department for Work & Pensions (HC 1455)

Examination of witnesses

Witnesses: Bozena Hillyer, Peter Schofield, Neil Couling, Catherine Vaughan and Richard Hawthorn.

Chair: Welcome to the Public Accounts Committee on Monday 18 September 2023. Today we are examining the Department for Work and Pensions annual report and accounts. In 2022-23, the Department spent £234.8 billion on benefits and state pension payments to claimants and pensioners. But this is the 35th year in which DWP’s accounts have been qualified by the Comptroller and Auditor General due to material issues around fraud and error, which we will probe today. We also intend to probe some of the other issues around overpayment and underpayment of benefits, as well as other issues.

I would like to welcome our witnesses today. We have Peter Schofield, the permanent secretary at DWP, and Neil Couling, the director general for change and resilience, and universal credit SRO. He has been with universal credit since its inception, so he is the world expert on it, I think it’s fair to say.

Neil Couling: That is very kind.

Chair: I’m not sure you will think that was kind when we start asking you questions about it, but anyway.

Then we have Bozena Hillyer, who is the director for counter-fraud compliance and debt, and also the head of function in the civil service for fraud. Is that the right title? Across the civil service, you are the—

Bozena Hillyer: No, for the Department.  

Q1                Chair: Forgive me, for the Department. Nevertheless, you are one of the leading people dealing with fraud and compliance across Whitehall.

We also have Catherine Vaughan, finance director general at DWP. From His Majesty’s Revenue and Customs, we have Richard Hawthorn, director for operational excellence, Customer Services Group—that is because there are interactions between HMRC and DWP, particularly as people move from tax credits into certain benefits. I also want to warmly welcome Sir Stephen Timms, Chair of the Work and Pensions Committee, who is here as a guest. It is always a pleasure to have a brother Chair with us, so thank you for that.

Before we go into the main session, Mr Schofield, I wanted to ask whether you have any RAAC in your buildings and, if so, what you are doing about it.

Peter Schofield: The story from our estates is that back in 2018, as the Committee will probably remember, our private finance initiative deal came to an end and we took control of our estates and built our own in-house estates team. I am glad to say that we have good expertise within the Department. Since about 2021, when I think the first concerns were raised, we have been working our way through to understand the nature of the risk within DWP. Our Secretary of State himself said quite recently that less than a handful of our 900 or so buildings are affected. We are monitoring the condition in line with the guidance from the Institution of Structural Engineers. We always keep the safety of our staff and customers paramount.

Q2                Chair: Is there any particular region of the country or age of property that is affected?

Peter Schofield: As I say, there are a few isolated examples, and they are very much under management.

Q3                Chair: Brilliant. Thank you for that.

To move into the main session, we are now seeing a little litany of issues—we will come to the detail of the underpayments on pensions later—in quite substantial areas, like pensions, where everyone was pretty trusting that the system was just going to deliver. We see this around older women—times two—and we have the issue of child benefit credits being applied to national insurance contributions. Are you doing any audits to look at whether there are any further problems in the Department? It seems almost out of nowhere. We have suddenly seen these problems arise over the last few years.

Peter Schofield: Yes. I can talk about two initiatives. The first is that we are doing more work in the sampling each year of the state pension. We changed the methodology in 2021-22. We have done full case reviews and, rather than relying on the documentation we have within the Department, we have contacted the claimant or pensioner and talked to them about their claim to try to understand the issues they have, the nature of their claim and whether it is fully being paid. That is what identified the HRP issue, which I am sure we will talk more about later, but you are right—

Q4                Chair: So it came up through manual sampling, effectively?

Peter Schofield: It came up through the normal sampling that we do. The official statistics that we produce for overpayments and underpayments each May are based on a sample of around 13,000 cases. The annual report and accounts go into quite a lot of detail about how we do the sample, but in the case of those that were sampled from the state pension, from about 2021-22 onwards we actually did a lot more work around understanding those claims.

Where we selected a state pension case for sampling, we then actually not only looked at the detail of what we had, but we then contacted the claimant—the pensioner—to say, “Can we understand the nature of the claim and go through that?” As part of that we asked, “Should you have been claiming HRP because you were claiming child benefit at some time in the past?” It is that work that has identified these underpayments and enabled us to do the work that I am sure we will talk more about. One element is better sampling and being more thorough about the state pension. That has identified some of the underpayments.

If I might answer the other part of your question about internal audit, to understand the nature of the controls we have and to give assurance about whether we are getting this right across the board, Jim Harra and I—

Q5                Chair: Jim Harra, the head of HMRC—

Peter Schofield: Jim Harra, the accounting officer at HMRC—absolutely. We have commissioned our respective internal audit teams to do some collective work to look at how the national insurance record is produced and whether there are any issues in terms of the quality and assurance we should have, particularly from a DWP perspective. Catherine may want to say more, but I think that the timing of that is that we will get that for the end of this calendar year and it will go to our departmental audit and risk committee.

Q6                Chair: Are you using any algorithms to pick up any of these problems, or is it this manual sampling that is the main driver of you finding these cases?

Peter Schofield: The algorithms sort of follow in terms of how you then do scans after the event to work out the nature of the challenge—for example, how many cases there are likely to be—and then we do sampling to understand the amount of money per case that we are likely to find. All of that work then feeds into the provisions that we make in the annual report and accounts, as well as the delivery plans for the LEAP exercises. But the starting point is the thorough sampling, and there is no substitute, really, for just methodically going through the case. It is the same methodology that we are using now for the targeted case review, where you actually contact the claimant, or contact the pensioner, and say, “We want to make sure that we have got this right.” Sometimes—obviously, a lot of the time—that identifies overpayments. But we found a lot of underpayments—well, a small proportion of a very large number in the state pension as well.

Neil Couling: An algorithm wouldn’t pick it up, would it Peter, because—

Peter Schofield: We wouldn’t know what to ask it.

Neil Couling: It wouldn’t know what to ask because, here, it was the samplers saying, “Were you receiving child benefit?” and the pensioner would say, “Yes, I was between 1988 and 2000”, say. Then we would go back and check the national insurance record and there was no HRP credit on there, and that is what flagged the problem. The claimant didn’t know, we didn’t know and HMRC didn’t know; it was the act of doing the inquiry that found it. It was quite interesting; we got a bit of flak about doing this. Some groups said, “Why are you questioning pensioners in this detail?” It was a small sample, but it was that act that allowed us to find this problem.

Peter Schofield: But, crucially, once you know what you are looking for, you can then do the scan.

Q7                Chair: When you start doing that exercise, do you find that any examples of previous fraud are thrown up through these historical exercises as well—through this sampling, Bozena? Is there anything that you feed in?

Bozena Hillyer: Peter was talking about the estimate that we do every year, which is what we talk about—

Q8                Chair: In the annual report.

Bozena Hillyer: Yes, absolutely, and there is the learning that we get from that. For example, when we look at what comes out of the sample, if the officer doing that sampling thinks it might be fraud, it comes across to our fraud teams—

Q9                Chair: Even if it is historical?

Bozena Hillyer: Yes, even if it is historical, we look at it because we need to learn from it and we want to ascertain exactly what we have found—“What  does this look like?”—because then that allows us to, as Neil says, go back into the data to say, “Find some more that look like this.”

Q10            Chair: Okay, so basically if someone has committed fraud against the DWP, you have a long arm to reach back quite a long way?

Bozena Hillyer: Absolutely, we will look, yes.

Q11            Chair: Another thing we have looked at, before I pass this to Stephen on more specific points, is issues in other Departments around data retention. That has come up with the pensions paperwork with you. Mr Schofield, are you looking at influencing how long the Government hold on to records?  It has been very difficult, with a number of those older women pensioners, especially if they have died, for people to go back. If you don’t keep the records, it is very hard to know that any one individual will have kept the records. What is the plan? I tag on to that, given that we have digital transformation washing through Whitehall, is it being planned in that people will be able to interrogate their records if they are kept longer, and can they be kept longer?

Peter Schofield: It is a difficult balance here, because, of course, we should only hold on to records—personal data—for as long as we have a business reason to do so.

Chair: But I think these examples have proven the business reasons for quite a long time.

Peter Schofield: Exactly. So in the case of the state pension underpayment exercise that we have talked about in this Committee a number of times, we immediately realised that we had a business reason to hold on to data that we previously did not think that we had. I suspect that that may be something that comes out of the internal audit work that I have just described, in terms of how we work together with HMRC. That may identify things that we need to do there.

Of course, one of the challenges that we have with the HRP exercise, although Richard may want to say more, is that HMRC, for understandable reasons, has not hung on—

Q12            Chair: We will come on to the HRP exercise in more detail, I think, but there is no legal barrier to keeping it longer, although I suppose there might be a test case, at some point, with the Information Commissioner.

Peter Schofield: We have to be able to demonstrate that we have an operational reason to do so, and the challenge is that we do not always know at this stage where the operational reason is.

Neil Couling: But the Government has recently announced that there is going to be a new credit for people who are subject to the high income charge on child benefit—people who are not claiming child benefit. That new credit will credit them with class 1 contributions towards the state pension. As we are developing that with our colleagues in HMRC, we are looking at how we will hold this information, because we are going to need it potentially over 30 or 40 years, until the pension is ultimately claimed. We are trying to learn lessons from this, as we develop new policies as well as enact the existing ones.

Chair: Thank you. I will now turn to Sir Stephen Timms MP.

Q13            Sir Stephen Timms: You said in the report that the figure for total overpaid benefits last year was £8.2 billion. How did that outturn figure compare with what you were expecting?

Peter Schofield: This is an area of considerable uncertainty. This will probably play into the conversation you might want to have about setting a target. There are two things that you have to look at when we are looking ahead at what is likely to happen on fraud and error: on the one hand is the effectiveness of our work in dealing with fraud by reducing, preventing, detecting and recovering fraud; on the other hand is the question of what is happening to the baseline and what is the underlying position.

The National Audit Office published the forecast that was underpinning the OBR’s spring numbers in their report and accounts; you will see it here in the paper that they included in our annual report and accounts. That showed that the outturn in universal credit of overpayments was expected to be 12.1% and it turned out to be 12.8%, I think, in 2022-23. That was just between March and May, when the statistics were published.

There is considerable uncertainty. It is not because we did not hit our targets. I mean, the target that I set Bozena of £1.1 billion savings in her work was absolutely met. It is about what was going on in the baseline; that is the uncertainty and it plays into what we are seeing in terms of the uncertainty around growth in the propensity of fraud, and how that affects the baseline position on the monetary value of fraud and error.

Q14            Sir Stephen Timms: The level of fraud and error is much higher than it was before the pandemic. Are you expecting the pandemic effect to gradually wash out, or has something substantially changed during the pandemic that means we can expect these high levels to last for some time?

Peter Schofield: I suppose yes on both. One of the things that we have done—you see it in the annual report and accounts this year, and we did it last year as well—is to set out some analysis that we have done of the claims that were made in three different cohorts: immediately before the pandemic, during the height of the pandemic and after the pandemic. If you compare where those cohorts are now with where they were last year, you will see that both the cohort in the middle of the pandemic, at the height of the pandemic, and that immediately after the pandemic, have come down, but certainly neither the middle one nor the final one has come down to the level of the cohort pre pandemic.

I think that is telling us two things. First, it is telling us that the impact of the pandemic is falling away over time. You will see in the numbers the impact as we reintroduce the easements—this year we have particularly seen it in self-employed earnings. You will see that that has driven a reduction in fraud and error in universal credit from where it was last year to where it has been this year.

Secondly, there is also something going on that is driving up the propensity to fraud and error, which I think will mean it will be difficult to get back to where we were pre pandemic, but we have a lot of initiatives to make a difference there, of which the work of the compliance and counter-fraud teams that Bozena leads, and in particular the investment in the targeted case review, should help to drive that down.

The OBR forecast makes assumptions both in terms of the effectiveness of our interventions and the impact of the baseline propensity for fraud increasing, and you will see in that forecast that fraud and error are set to come down. But there’s a lot of uncertainty there, and in particular a lot of uncertainty about the baseline position.

Q15            Sir Stephen Timms: Chart G on page 108 shows that the percentage of DWP expenditure overpaid each year, broken down by benefit, came to about 2% typically. It went up a little bit in the year before the pandemic, then it shot up to 4%, and last year, it came down a bit. From what you are saying, we can expect it to carry on coming down, but we are not quite sure how far it will get back towards where we started. Is that right?

Peter Schofield: The National Audit Office published the forecast, so I can turn to that as well. It is in the Report—Joshua, whereabouts is it?

Joshua Reddaway: It’s on page 37, figure 8.

Peter Schofield: That goes a bit beyond the chart in the annual report and accounts that you described. There are two things to bear in mind in all this. First, remember that our statistics are somewhat lagged, because they are based on a sample carried out in the previous calendar year. The statistics we published in May this year were based on a sample that we carried out between September 2021 and October 2022. That is the first thing: the effect of reintroducing controls is lagged.

How does that play out, in particular in the self-employed? In August 2021, we reintroduced the gainful self-employment test. It took us a year to roll it out to all self-employment claims. Once we had rolled it out, we then introduced a minimum income floor 12 months later. If you apply that to the sampling period we are talking about—from September 2021 to October 2022—towards the end of that sample period you will see the minimum income floor starting to cut in. That was making a difference to the self-employed number, but, as the NAO itself points out, that will then play out more strongly next year and the year after; we will only see the full effect in 2025.

That is one of the things going on, but beyond that you will see the interaction of our interventions. We are starting to see the targeted case review building up over a period of time as we not only recruit the people but get through the cases. That then plays out in work in other ways, such as the growth in the propensity for fraud, which we estimate at 5% per annum.

Q16            Sir Stephen Timms: At the moment, one in three universal credit claims is incorrect. Can we conclude from that that the part of the universal credit business case that was about reducing fraud and error has not really materialised?

Peter Schofield: I would answer that in two ways—I think you asked a similar question this time last year, Sir Stephen, and I would answer it in a similar way. You can look at the impact that a lot of the measures carried out as part of universal credit, in the move from the legacy benefits, are having in the statistics—for example, in the way that the automated nature of being able to understand earnings from employment played out. We have seen that the earnings from employment fraud and overpayment have come down from 2.5% in legacy benefits to virtually nothing. The overpayments from earnings have all come from the self-employed, where we do not have the same data feed, but we can talk about that another time as well. The second element to that is the way we have designed out fraud, for example, on childcare. The fraud from childcare costs has come down to virtually zero. These are things that were designed out either through the design or the implementation of universal credit.

The second thing, on your point that there’s all this stock of fraud in the system, is that so much of that relates to what came in during the pandemic. It also comes in from quite low-level fraud and overpayments. You see that in the analysis that the NAO has done, which shows the distribution of that fraud in terms of the amounts. The best way to address that is through our targeted case review, which will really get to the heart of this. I think that by the end of the whole period, we are aiming for the investment to cover something like 8 million claims or to go through 8 million claims one at a time. We are aiming to get through those claims to understand the nature of where there is overpayment and get it stopped.

Q17            Sir Stephen Timms: To what extent do you think that the very high proportion of claims being wrong is about complexity? Is it because it is just really difficult for people to work out what the correct payment ought to be?

Peter Schofield: Quite a lot of this is down to how you make it easy for claimants to help us pay the right amount by declaring their change of circumstance at the right time. The annual report and accounts—Catherine may turn to this at some point as well—has a page about the number of continuous improvement changes we are making to universal credit, including things we are doing to make it easier for people to declare their change of circumstance, so, for example, on capital. We have seen a reduction in fraud and error from capital over the last couple of years. It went up a little bit last year, but there was a reduction the year before that, in part due to the fact that we make it easier for people to declare capital.

There is another element of the fraud and error situation going the other way, in terms of underpayments. A key part of the universal credit business case was to help address underclaiming. By connecting a whole load of legacy benefits together, it is possible to deal with a situation where someone might have been claiming JSA but not housing benefit, because it is now together in one claim. About £1 billion going the other way, in terms of additional costs through the universal credit business case, was helping to address underpayments, and I think underpayments are relatively low compared to other benefits.

Q18            Sir Stephen Timms: You have indicated that most of the fall that we saw last year was accounted for by tightening up the self-employed compliance. When do you think the other measures in the strategy will start to show in these figures, in terms of improvements?

Neil Couling: We set out on page 99 of the departmental report when we think that the effect of the measures we have taken so far will start to appear in the statistics. Each year I come here and explain the time lag point to you—you are probably all sick of me doing that, so I have put it in the report this year, to try to set that out.

On the targeted case review, in our forecast, we have made quite a conservative assessment of its yield. As we get into more volume, we will get a sense of just how successful that is. No doubt that will inform downward pressure on the forecast, but we are seeing increasing propensity towards fraud as well. That might zero out or one may exceed the other, but we just don’t know yet. This is quite a hard area to forecast in.

Catherine Vaughan: I want to add that we have taken quite a tough line on deciding that overpayments might be fraudulent. That is picked out in figure 4 of the NAO Report. Where we don’t get responses from claimants, we are assuming that they are fraudulent. That accounts for about 15% of the total. Many will be, and that might be the reason they have not engaged with us, but others may not get back in touch with us for reasons of vulnerability. We have set ourselves a particularly tough test on assuming that all of those claims and all of their value are fraudulent.

Q19            Sir Stephen Timms: Just to pick up Neil’s point, page 99 of the report tells us whether there is an impact on last year’s outcome, but I don’t think it tells us when we can expect to see improvements.

Neil Couling: If you look down at the final column, so the habitual residence test—

Peter Schofield: Look at the bottom row, for example. The MIF is the critical one.

Q20            Sir Stephen Timms: It says full impact in 2025, but that is for self-employment. It’s the other things you are doing in the strategy that I wondered about.

Neil Couling: The targeted case review is having an effect today— it is stopping fraud today—but it won’t be at a volume big enough to be really picked up in the sampling, unless you were really lucky in the cases you sampled. We take 3,000-odd universal credit cases through the sampling. That will take two years before it starts to have a measurable effect.

Sir Stephen Timms: Two years.

Peter Schofield: Figure 10 on page 41 of the NAO Report shows you all the different elements that we are talking about—groupings of the different elements of the plan. That shows when the savings will come in. They then need to show up in the statistics, so then you have got the annual lag. That is one way of looking at that.

I just want to answer your previous question a bit better, in terms of helping claimants understand the nature of the information that they need to provide. If you turn to page 105 of the annual report and accounts—sorry, there are two documents to navigate around—you will see a whole series of initiatives that we are putting in place. For example, on capital, they will help customers understand what they need to do to update us about their claim. Living together is another big—

Q21            Chair: I think people following might not all know what you mean by capital. We do—it is savings, basically, isn’t it?

Peter Schofield: Yes, exactly—sorry. Your universal credit claim is reduced if you have savings over a certain amount, and it is important that claimants tell us about savings so that we get their universal credit correct.

Equally, if you have two people who form a couple and start living together, they become a single household and that has an impact on their entitlement as well. If you look down that page, you will see “Living Together”. We are saying more online to help people to understand when they need to declare a change in relationship to get their claim right.

There are a number of different things like this in our continuous improvement initiatives, which are factored into the OBR forecast only in so far as we are able to say to the OBR up front, “These are the things that we are working on.” The whole point of continuous improvement is obviously, as you go along, you develop new ideas, but they only get scored when we can say to the OBR, “We know exactly what we are going to do.”

Neil Couling: There is one other thing I might add, which I have been exploring with Josh and the NAO a bit—we will probably go a bit further. In some sense, the policy choices here lead you to some levels of fraud, too. Take living together fraud, for example: £900 million—a serious amount of public money—is being lost to people not declaring that they are part of a couple.

I could, in one sense, get rid of all that fraud tomorrow by persuading Ministers that we don’t have a household test. That would cost £40 billion. Am I saving £39.1 billion or am I losing £900 million? The truth is that I want to reduce that £900 million, but the choice in terms of protecting the Exchequer is a policy question as well. If you target benefits on those with the most need, you will have a higher level of fraud than if you pay benefits to absolutely everyone.

Peter Schofield: Which, of course, is why we have the difference between fraud and error in the state pension, compared with universal credit.

Q22            Peter Grant: I want to come back to the questions the Chair was asking about the pension underpayments—in particular, what happened was you did not have the information you needed about whether someone had previously received child benefit payments. I know that you said that data protection legislation means that you can only hold information for as long as you need, but legislation always says the information has to be accurate, complete and sufficient. You are not allowed to hold information if you do not have enough to perform the tasks that you have collected it for.

Have you considered, in particular, the example that I mentioned, where you can say either that the information you held was inaccurate because your system believed that that person had never received child benefit payment when they had, or, alternatively, that the information was incomplete because you did not know whether or not they had built up that entitlement? Does that not mean that, for that length of time, in each one of those cases, you were failing to comply with data protection legislation?

Peter Schofield: The national insurance record is the responsibility of HMRC, but in the case of the HRP situation, back in 2009, there was understood to be an issue and we worked together with HMRC to try to put that right. We put a number of records correct and put them into payment correctly. I guess at the time it was then assumed that the situation had been addressed and solved, and, in that situation, no one would have been able to make an operational case to hold on to that information for any longer. You have to make a judgment based on what you know at the time and what you believed to be the situation.

Q23            Peter Grant: When you said it had been assumed, who made that assumption?

Peter Schofield: The teams that were working on it between 2009 and 2011 went through the cases that they had identified. What we think happened is that we looked through the cases—Richard may want to add to this—where the individual was already receiving the state pension and put in place changes to put right their national insurance record and then pay the right state pension. For those who had not yet reached state pension age we put in place a new stage in the process, which we thought would pick up those cases where there had been child benefit that was paid historically, but, as we go through this new set of cases, that might well be where the problem arose and we did not actually get that bit of the process correct.

Q24            Peter Grant: Can I just check that DWP and HMRC are separately registered as data controllers and data holders with the Information Commissioner and that there isn’t just a single registration for the whole of Government?

Peter Schofield: We are all responsible to the Information Commissioner for the data that we hold. The national insurance record is the responsibility of HMRC, but we feed information into HMRC on certain things, such as—

Q25            Peter Grant: Does DWP have its own registration and its own registered data protection officer notified to the Information Commissioner? Yes? In that case, it was the responsibility of the DWP to make sure it had all the information it needed. You did not have that information and you carried on not having that information.

Peter Schofield: We thought we had the right information because we were relying on the national insurance record to make the right payments for the state pension. That was the process that we followed. As the Chair said at the beginning, there is now a question about whether we had the right assurance over the accuracy of the national insurance record. That is why my opposite number at HMRC and I have commissioned our internal audit teams to do the review that you have described.

Q26            Ben Lake: Can I ask about the way in which fraud is classified for the purposes of the Report? I note that it is estimated that some 76% of overpayments were the result of fraud. I appreciate that it is quite difficult to assess whether or not there was an intent behind some of these claims that were later judged to be fraudulent. There has been quite a broad range of circumstances, and the Comptroller and Auditor General’s Report suggests that some decisions are very difficult to ascertain whether they are fraudulent or not. May I ask Mr Couling this first? Do you think there is a benefit, in terms of transparency and better understanding for the public, in moving to or perhaps adopting the classification system that HMRC operates in terms of innocent, careless and deliberate behaviour?

Peter Schofield: Shall I come in first, and then Neil might want to add to this? I think there are two things going on here. One is the statistics that we have produced, and the statistics are based on the sample that we do of all benefit recipients. It is described in a lot of detail in the annual report and accounts. As part of that we have to assess whether something is an official error, a claimant error, or fraud. Those are the categories, which I think are helpful in terms of the understanding. The NAO points out that the data that we produce and the information we publish is well recognised internationally for its quality. That is the first thing.

The second element is how we treat any individual case. I am making a distinction between a sample so that we can produce high-quality statistics every year and how we handle any individual case that might come out of the sample. It might be a case that we sample and we think, “That looks like an intention to defraud. We will classify it as fraud for the statistics,” or it might come out of the wider compliance and investigations that Bozena and other colleagues across DWP lead for us.

You then have to make a decision about how to deal with that case—whether it is a case where you just need to put it right and ask for an overpayment; whether there is a threshold that is passed in terms of the criminality; and whether there is a case for prosecution, in which case is it of a scale that requires a criminal prosecution, or is it a case where an administrative penalty is more appropriate? In all of that, we always, always keep close regard to whether the individual is vulnerable and needs special support—we could talk more about how we do that.

That is our approach. I just wanted to make a distinction between the statistics, which I think are well recognised, and how we handle a case in real time. Bozena, do you want to say a bit about how we deal with the question of how we proportionally follow up on a fraud case?

Bozena Hillyer: Absolutely. The really important thing to understand about the statistics is that, if the officer who is doing it thinks it is fraud, it does come across to my teams to have a look at. We have talked already about the fact that, with a failure to reply, we make an assumption that that is fraud, but for the other ones it is because we have looked at it and we do take a view as to whether it was fraud or not, and therefore what the root cause of it was. I am really comfortable in terms of the statistics that we have there.

It is very important, when we look at everything that we do, that we do need to make sure that in the right cases we prosecute, because it is part of our deterrence strategy. For the vast majority of people, who are not fraudulent, it is really important for them to know that it is a level playing field, and that we do do something about the people we find. But we are also really clear when we are applying any of our processes that we look very carefully for any markers of vulnerability, and if we find markers of vulnerability, then we take that investigation through a different route, working with other colleagues in the Department—our advanced customer support teams, for example—and we will put in the things that that person might need. We might do a face to face, we might do a visit, or whatever it is. So it is both sides of the coin: we need to make sure that we do everything we can to prosecute where that is the right thing to do.

Q27            Chair: But it will still be recorded as fraud whether you prosecute or not?

Bozena Hillyer: Yes, absolutely.

Q28            Ben Lake: Just on that, the information you have just given is very useful, but I still want to understand if there has been a case of what HMRC might characterise as carelessness on the part of the claimant when there is no denying that it is incorrect and they have been overpaid incorrectly. I still think there is a distinction—an important and valuable one—between fraudulent intent and carelessness. In that case, am I right to understand from what you have just told me that you are able to say, “Of the 76% of overpayments that we have characterised as fraud, we can tell you how many are really carelessness, and we need to better support individuals”—vulnerable individuals, as you mention—“and those that are intentional fraudulent behaviour by those with ill intent and, indeed, criminal organisations”?

Neil Couling: I think Bozena’s people, when they do the fraud scrub of the claims, will look and see, and if it was careless, they would say it was customer error, not fraud.

Peter Schofield: If you turn to page 93 of the annual report and accounts, what we do there is we set out the different definitions. This would, I am sure, be regarded as claimant error and, as Neil says, recorded in the statistics as such.

Q29            Ben Lake: So to all intents and purposes, we do not need to adopt the formal terminology that HMRC use, because in practice you are happy that those cases that are defined as fraud would be, for the purposes of the HMRC categorisation, deliberate, and claimant error would almost coincide with HMRC’s careless categorisation.

Bozena Hillyer: Peter, could I just add—maybe you can indulge me, as an ex-tax inspector, slightly—that that partly stems from the different legislation and the things actually defined in legislation, which is done differently. So there are reasons why we define things in a particular way, and why HMRC—I will not speak for HMRC, obviously—go down the root that they do.

Q30            Chair: Mr Hawthorn, do you want to add anything?

Richard Hawthorn: I think Bozena has summed it up absolutely rightly. The legislation is framed in different ways, and I know Jim Harra has been to this Committee before and explained the basis for carelessness and other definitions in our compliance activity.

Q31            Chair: Of course, you will be dealing more with tax payments, because tax credits are effectively moving out of your Department soon and they will be less of an issue.

Richard Hawthorn: Exactly, yes.

Q32            Peter Grant: Mr Schofield, I want to switch topic for a minute and come back to the questions I asked you a few months ago. They were specifically about the staff involved in assessments of benefit claimants—capability assessments, health assessments—and I highlighted the fact that, among the parts of the service that have transferred to the Scottish Government, to Social Security Scotland, right across the board staff surveys and claimants surveys are showing that Social Security Scotland is better regarded by staff and by claimants than people doing the equivalent work in the DWP. What have you done since then to understand the reasons for that?

Peter Schofield: As I said to you when we met back in July, I work closely with my opposite number in Scotland—in fact, he is a former colleague of mine in DWP—and we work very closely with the head of the Social Security Scotland group and are talking to them. We talk very regularly with them about the progress that they are making in terms of taking on and setting up the adult disability payment, and we talk very regularly about what we can learn from each other. I think one of your specific questions was around what we can learn from the in-sourcing of health assessments—the fact that they do them in-house. That is something that we keep in touch on and talk regularly about.

In terms of the people survey, we are about to start the process across the civil service for the annual people survey. The best thing, in terms of your point about the people survey, will be to see what we see coming out of the current year and to learn from that. The data by then will be the most up to date.

Q33            Peter Grant: So basically you haven’t done anything specifically in relation to the question I asked 10 weeks ago, because the answer you have given me is almost identical to the one you gave me 10 weeks ago.

Peter Schofield: I think it is pretty identical. As I say—

Q34            Peter Grant: For example, I flagged up that 90% of people who have a capability assessment carried out by Social Security Scotland think they have been fairly treated and treated with dignity during the process. The figure for your own staff, your own Department, is nowhere near as high as that. Ten weeks later, have you done anything to try to find out why claimants in Scotland are so much more likely to feel as if they have been fairly treated than claimants elsewhere?

Peter Schofield: We regularly talk to them about the process, but, as we were talking about back in July, we are currently awarding the contracts with a new functional assessment service. This is something that we will look at in the context of that.

Q35            Peter Grant: So there is a possibility that in-sourcing is a significant factor in giving people a better service, but in spite of that, you are still going to carry on outsourcing, or re-outsourcing, the work of the DWP.

Peter Schofield: We talked in July about the decision we made not to in-source, and there has obviously been no change in that decision since July, Mr Grant. We will need to see how the process goes. As I say, I regard this as a process going forward where we need to learn from each other, but we are awarding contracts for the next five years, and we will see how Social Security Scotland do in terms of their process. In 2027—I think we were talking about the timescale back in July—we will need to make the decision at that point about what we do beyond 2029 in terms of the health assessment service. It is those sorts of timescales. I am afraid that it is not a strategy that we would change on a week-by-week basis, Mr Grant; it is something that we need to look at in the timescales that I talked you through back in July.

Q36            Chair: We touched earlier on the underpayment of the state pension process—the LEAP exercise and so on that you are going through. Could you give us an update on where you are on those numbers?

Peter Schofield: Yes. It is set out in quite a lot of detail in the annual report and accounts.

Chair: Can you amplify it?

Peter Schofield: We are making good progress in terms of ploughing through the cases that we need to look at. As the annual report and accounts says, we are accelerating the progress at which we get through those cases. In the last quarter that we reported on, I think we got through about 108,000 of those cases. What we are seeing as we go through those cases is we are getting a better understanding of the likely number of people who will be eligible for payment and how many will be able to trace and pay, and then where we end up in terms of the overall amount of money that we will need to spend. Catherine will probably remind you of the provision—what provision are we making now?

Catherine Vaughan: I was just going to give an update on the monthly processing rates. The latest data we published in July shows that we have reviewed an average of over 30,000 cases a month in the last quarter, compared with an average of 5,000 a month in the first 22 months. Actually, I think it is a graph that the NAO also produced in its Report on accounts to show the increasing pace. We are on track to complete the exercise for category BL and category D cases by the end of 2023, which is a commitment that we have had in place for a while, and we are on track to deliver that.

Peter Schofield: If you turn to page 124 of the annual report and accounts, you will see chart Q. I want to be quite open and transparent about this, because I know this is a question that you regularly ask.

Q37            Chair: Obviously those caseworkers will be coming off this as time goes on, but you are picking up other problems. Will they be using the same cohort? Will you roll them on to other problems?

Peter Schofield: Yes. As Catherine says, we are looking to complete the category BL and category D work by the end of this calendar year. Happily—I’m not sure whether to use that word, but I suppose it is the right one—they have the right skillset to take forward the work in terms of remediating the state pension for those where an HRP is missing. Richard may want to say a bit about the timetable for HMRC working through their scans of people who might be missing an HRP.

Chair: That brings us to Sir Stephen Timms, who is going to pick up on that.

Q38            Sir Stephen Timms: Yes, I want to pick up on a point that Neil touched on a few minutes ago, around the question of people whose income is too high to receive child benefit, and the worry that they will not do what they are supposed to in order to maintain their national insurance credits for their future state pension claim. I think you were saying that the Government have taken an initiative on that. Is that right? Could you tell us what that is?

Neil Couling: I think Ministers announced in April, if I remember correctly, that there would be a new credit here.

Q39            Chair: This is where one partner in a couple is staying at home with caring responsibilities, but the other’s income means they cannot get child benefit.

Neil Couling: Yes. They are not claiming child benefit, so they do not qualify for the credit that comes with claiming child benefit. A new credit is being created for people in that category, where they have a working partner who is earning above £50,000, which I think is the high-income charge, and I think it tapers away by the time you get to £60,000—above that level child benefit is not being claimed. Your choice is: do you claim the child benefit and pay it back through your tax return, or do you not claim the child benefit? We decided in our family not to claim the child benefit.

Q40            Sir Stephen Timms: Is the new credit going to be awarded automatically, or do you have to apply for it?

Neil Couling: That is what we are working through with our colleagues in HMRC. I raised this because it potentially brings up issues about how you keep records over a period of time.

Q41            Sir Stephen Timms: It would not be at all surprising if in a few years’ time there are quite a few people who get to state pension age and have a nasty shock when they discover they are not entitled to a full state pension, if the credit is not awarded automatically. When will you decide whether it will be awarded automatically?

Neil Couling: It is not really for us to say, although we are feeding into HMRC about what a good outcome from our perspective would be. I cannot speak for HMRC as to what it will do here. If in 20 years—I don’t think this will be for me because even I may have retired by then—this were to manifest as a problem similar to how the current HRP problem is manifesting now, the question will be: what steps did you take? We want to be certain, and I know that HMRC does too, that we take the maximum possible steps to ensure that an analogous problem to what we have with HRP now does not occur with this new credit.

Q42            Sir Stephen Timms: Richard, do you want to come in?

Richard Hawthorn: The whole point of that piece of legislation, which we are talking to Ministers about literally as we speak, is to avoid the very scenario where people end up in a few years’ time without the full credits you are describing because they have not claimed child benefits. We always aim to make it as easy and as simple as possible for our customers to get the benefit of new legislation. The detail we will need to work through with Ministers, but that is our aim.

Q43            Sir Stephen Timms: Will it be automatic? Being easy is one thing but a lot of people will—

Richard Hawthorn: I cannot commit us to being completely automatic because that conversation is ongoing. Our intention will be to make it as seamless and easy as possible, and that implies a level of automation.

Q44            Sir Stephen Timms: A level of automation?

Richard Hawthorn: It depends on the framing of the legislation and how we work it through. We have got no more desire than anyone else to make this hard for customers or for ourselves. The more automatic it is, the easier it will be for everyone. The policy objective is to avoid the circumstance where someone inadvertently arrives in a situation where they have not had the full credits.

Q45            Sir Stephen Timms: What is the difficulty with making it automatic?

Richard Hawthorn: There is a question about whether it is something that should be claimed or given, and that will be part of the framing of the legislation that will need to be put in place.

Peter Schofield: This will be something about whether there is data. What you are trying to prove is whether someone would have been entitled to this claim. Does HMRC have the data that enables you to understand the nature of it?

Q46            Chair: But going forward, on the points that Sir Stephen has just raised—I will come back to the historical stuff in a moment—surely when the child benefit rules were changed, it was foreseen that it is a gateway benefit. When my daughter went to primary school, I was supposed to provide child benefit proof to prove that I was the parent. I cannot remember what the reasons were, but it was the document that was required to prove that I was the parent. I know sometimes that things are announced by politicians without always the care and attention to detail, but do you now think that maybe there should have been some warning that there were going to be issues with people withdrawing from claiming child benefit completely?

Peter Schofield: This is territory well beyond the Department for Work and Pensions. It is probably something to pick up with Jim when you have your hearing with him. Our interest here, and the reason why we are interested in this, is that we want to make sure that we are paying the state pension accurately, and this goes to the heart of the work that we are doing jointly with Jim in terms of the integrity of the system.

Q47            Chair: Mr Hawthorn, this is causing people complications. They have either had to do a tax return when they had not before, or had to not forget to add into an existing tax return. It seems like we have a very complex picture now, and it is having more ramifications than were first envisaged.

Richard Hawthorn: It is hard to deny that it is a more complex picture because of the—

Chair: Universality is always going to be more complex, obviously.

Richard Hawthorn: Yes. I recall when HICBC was first introduced and the limitation, if you like, on people claiming benefit and it effectively being means-tested. That was brought in very quickly at around the time of the introduction of the austerity agenda. We have been learning since. In the context of the HRP issues and others, we are now trying to make sure that we fix that particular gap and make sure the number of customers who are affected by that risk is minimised as much as possible. That is our intent.

Q48            Chair: Who is going to be responsible for keeping accurate records? Which Department wants to take that on?

Peter Schofield: Child benefit is the responsibility of HMRC. The national insurance contribution record is HMRC. Our interest is making sure that we have an accurate national insurance record to pay the right state pension.

Q49            Chair: Is assuring yourself of that part of your joint work with HMRC?

Peter Schofield: This is a thing for us to get to the bottom of—how exactly does that work? We obviously knew that we had an interdependency here, but the HRP process has demonstrated that. How do we make sure that we have joint assurance on the overall accuracy of the record?

Q50            Chair: So, practically, can you just talk us through the HRP issue? I suppose this question is for you, Mr Hawthorn. Someone had a child and was entitled to home responsibilities protection. The child has grown up. It has only become apparent at the point of their claiming their pension that the records were not complete. The reverse engineering of that is even more complicated than the reverse engineering of the mistakes with the underpayment of the state pension for women of certain ages. What are you doing? What are the problems? We know a little bit of this from what the NAO has told us, but do you want to give it to us from the horse’s mouth?

Peter Schofield: Richard will come in, but HRP was something that began in 1978, I think, and lasted through till 2010. We think it is related to people who were claiming child benefit up until 2000, where there seems to be an issue about the fact that the organisation that was responsible at the time did not ask the parent to put in their national insurance number as part of the claim, and therefore it was difficult to make the link to the national insurance contribution record. HMRC does not have the records. Do you want to say a little bit about what we are now going to be doing in terms of the process, Richard?

Richard Hawthorn: Sure. It is exactly that: child benefit did not require the national insurance number up until the year 2000, when it became mandatory, which is why we believe the gap here is between ’78 and 2000, and stops at that point. The 2000 to 2010—

Q51            Chair: But even a child born in ’99 is now well beyond child benefit, and the records would be destroyed. So, just to be clear, this is a big problem.

Richard Hawthorn: It is a big problem, and quite a challenging one because of the records that we do not have any longer. We are doing various scans on our IT systems to identify the very likely cohort of parents who could have been eligible for HRP. We think, on the basis of early scans, that that could be several hundred thousand people, although, as the DWP accounts have said, we think the actual number of people affected will be smaller. We will write to all those people progressively. We expect to start doing that this week, coincidentally, or if not, then next week—certainly by the end of September. To begin with, it will be a small volume, so that we can start to understand the number of people who will come to us.

Q52            Chair: Just to be clear, what will you ask people? Did they have a child and were they not working?

Richard Hawthorn: We are going to say, “We believe you may have been eligible for HRP during this period.” We will set up the basis for that, and invite them to make a claim, which we will then assess.

Q53            Chair: Will you have a control, effectively, through your tax records about that individual, so if someone had a taxable income that suggested that they did not have the home care responsibility, then they would not qualify? I have a child in that cohort, and I could say that I qualified. As it happens, it is pretty obvious from my public life that I was working, but how would you know? Surely you would have to match it against records.

Richard Hawthorn: Absolutely. It is not the case that people will come in and apply, and we will just give it to them all. We will apply a level of due diligence, and will aim to make sure that people who are claiming are eligible. We have an online checker that customers can already use. It has been available since August.

Q54            Chair: We have looked at online checkers and discussed them before, and for pensions models, it was going to be too complicated to have them. Are you sure that this one will catch most people?

Richard Hawthorn: The online checker will allow customers to understand whether or not their pension credit is full. Where it is already full, HRP will not make any difference. Where it is not, it invites them to think about whether they are one of the customers who had a child, claimed child benefit and did not get the benefit of HRP. We will be able to identify them. The plan is to start small, and learn as we get feedback from customers, understand the behaviours and how they are responding to it. We plan to roll that out to cover all those several hundred thousand—whatever it is—within the next 18 months.

Chair: Eighteen months is quite fast.

Richard Hawthorn: We recognise that these are people already in the state pension—

Q55            Chair: But that is for getting letters out to several hundred thousand. How much time have you estimated or allowed to deal with each case and claim?

Richard Hawthorn: We have a general post turnaround time. That is, we aim to deal with our priority post—80% of it—within 15 working days, and we expect to be able to do that for these customers as well.

Q56            Chair: Then once you have got the information, you pass it over to DWP?

Richard Hawthorn: We will assess it, and where we think there is a legitimate HRP claim, we will notify the customer that that is what we have done. We will correct the national insurance records, and at the same time send a notification to colleagues in DWP, who will do the necessary work to assess the impact on the state pension.

Q57            Chair: That means lump sum back-payments of pension.

Peter Schofield: Yes. We have a “business as usual” process for people who, for whatever reason, identify a gap in their national insurance record and have that corrected by HMRC. We can deal with that, set it right going forward, and deal with any underpayment.

Q58            Chair: Do you have an estimation of the amount of money—or a range—that claimants might get?

Peter Schofield: We have made a provision in the annual overall accounts, and we have set out in detail how that has been arrived at. Catherine can probably better take us through that.

Chair: Ms Vaughan, you are the money woman.

Catherine Vaughan: Yes. There is obviously a significant amount of uncertainty about the provision, and we will get more certainty as we start seeing live cases and live information. We have set out a range in the annual report and accounts. Our central assumption is that the average arrears for the people above state pension age is £5,000, and £3,000 for those who are deceased. We have put in assumptions about take-up, delivery and completion, and how much those change the overall value of the provision, but it is very early days.

Q59            Chair: So it is quite a life-changing sum, potentially, for people. With regard to those who are deceased—we have obviously touched on this before, with the pensions issue—what efforts will you make to pursue next of kin, Ms Vaughan? And Mr Hawthorn, it is hard enough to prove all this if you are alive, but what if you are the child of a deceased pensioner who should have had this money? What are you doing on that? Ms Vaughan first.

Peter Schofield: Shall I speak on the next-of-kin thing? We have made a provision—I think we are assuming 75% of cases—

Q60            Chair: I was interested in that figure; it is quite high.

Peter Schofield: It is based on what we are seeing in terms of the Category BL recoveries. We have set up a next-of-kin portal. Basically, if you think that you have a relative who might have been entitled to an underpayment payout, you can register with us on the portal; that is a way of helping us to identify next of kin where there is an underpayment to be paid out. There are other things we are doing to try to address that. We have talked about that in this hearing before. We are basically learning the lessons from previous experience.

Q61            Chair: Sadly, you are becoming an expert in this. Mr Hawthorn, someone has died, and a family member thinks they may not have claimed full home responsibilities provision. How do they go about finding that out, and can you do that from a third party and match records?

Richard Hawthorn: You are right: it is incredibly difficult to find the right next of kin, given the age of some of these cases, so that will be a limiting factor for us.

Q62            Chair: It is from ’78 to 2000, so that would be anyone with a child of 25 to 45.

Richard Hawthorn: Exactly that, so it is going back some way. That is why, alongside the targeted intervention I have described, we are building a communication package, working with stakeholders and others, to say, “Actually, there’s a range of customers who could be involved. Please help us identify them and get the message out, so that where there is a next-of-kin situation, those customers can come in and talk to us.” The next of kin can use the checker, and can use the same form in the same process. We have already seen one or two of those come through.

Q63            Chair: What if you have several children, and they all try to be the next of kin? Let’s say someone is already deceased. How do you qualify someone as next of kin in those situations?

Richard Hawthorn: We will make a judgment as those come through, so that we make sure that we don’t double-pay.

Q64            Chair: Well, not double-pay. My mother has a lot of children. If one gets in first, there are an awful lot of disappointed siblings behind in the queue. Is it just first come, first served?

Richard Hawthorn: We will need to work that one through; we have not got into the detail of that yet.

Chair: The person has died and their estate has been settled.

Richard Hawthorn: It is a challenging area; we accept that.

Q65            Ben Lake: I do not envy you your task at all, Mr Hawthorn. You have just set out the process by which you will try to identify the number of people affected. Forgive me: you may have mentioned when you hope to have a firmer idea of the number of people affected by this.

Richard Hawthorn: Well, we are constantly reviewing the scans. I am due another scan this week. That is what will trigger the first of those. As we get that feedback, we will be able to reassess, working with colleagues from DWP. It is worth saying that we have set up a joint committee, which I chair jointly with the director of DWP, to oversee the recovery of this. We will look at that and refine the scans, and hopefully that will give us greater clarity as we go forward, but it is a very iterative process.

Q66            Ben Lake: I am sorry to ask again, and I appreciate how complicated this process is, but what sort of timescale do you hope we are looking at for ascertaining the number of people affected?

Richard Hawthorn: We hope to have written to them all in the next 18 months. We will iterate, and learn as we go about whether that is realistic or not.

Q67            Ben Lake: Mr Schofield, a similar question, really: if everything goes to plan, what sort of timescale are we looking at for any repayments being completed?

Peter Schofield: As I say, the happy fact or the silver lining to the cloud of the state pension underpayment exercise is that we have a group of qualified, experienced colleagues who are ready to go with this, virtually as soon as Richard’s team have started their process, because the category BL folk are finishing their exercise at the end of this calendar year; they are ready to go.

We do not have a clear, formal delivery plan. We need to work that through. That is one of the things that we are working on. We have based the provision on different assumptions, which are all set out, of how long that would take. Obviously, Richard is thinking that they can get through in 18 months. We would want to try to be as close to that as we can, but the provision will be based on completing in four years. That is because we have to make provision on a conservative basis, but maybe we can do it quicker than that. We will report regularly on this, and I am sure the Committee will be asking questions.

Q68            Chair: Do you know what percentage of those involved will be women? I guess that in both cases—the previous pension issue that we looked at and this one—will disproportionately affect women. What percentage of overlap is there, where one interacts with the other?

Peter Schofield: I do not think we know that. We think it is about 210,000—

Q69            Chair: But let us take people who have already been dealt with, through your remarkable achievement of getting through cases of underpayment of pension. If it is then discovered that they had home responsibilities protection rights, what happens? Would they be entitled to more, or have they been given the uplift?

Peter Schofield: That is a complicated one, because if they benefited from their husband, as it would be in those cases, it might well be that they then made up—

Neil Couling: They had the full contribution.

Peter Schofield: They got the full contribution after all anyway, so this may not make a difference. That is why this is an incredibly complicated exercise; we have to build it on to what people have already done. The other complication I should mention is that people might be receiving pension credit, and that would then be offset. We have to take that into account the other way. It is a complicated process, but we have experts—

Q70            Chair: Does pension credit open up other benefits? Is it a gateway benefit for certain things? I think it might be.

Neil Couling: Yes, certain cost of living payments.

Q71            Chair: You have that as well. People might not have qualified for pension credit. Will you try to recoup any of those cost of living payments, or once they are out of the door, is that it?

Peter Schofield: I do not think we have any plans to do that.

Q72            Chair: That will be helpful for people to know. How will we make sure that this does not happen again, Mr Hawthorn and Mr Schofield?

Peter Schofield: At the heart of this are various things that we have talked about already, but there are also other things that have come out in the NAO Report on the accounts. For example, there is the issue of having assurance about the national insurance record, and seeing what the internal audit work comes up with. We will know more about that towards the end of this calendar year. That goes to our audit and risk committee, and the National Audit Office is part of that. That is about full disclosure. I am sure we will be able to say more about that in next year’s annual report and accounts.

There is also the wider work that we are doing that the NAO recommended on how we pull together early intelligence of underpayments. On historical underpayments, sampling and the work that we have done has been quite helpful in the past, but there is more that we can do. Our quality teams are working on how we take into account underpayments that have been identified, complaints that are coming up in the system, or how we improve the tier-2 quality—the sampling that we do in-year, not for the statistics, but to make sure that we are getting things right as we go along. There is quite a lot of intelligence to pull together. Elizabeth Fairburn, who came with us to last year’s hearing, leads on that work. I hope there will be more to say on that.

Neil Couling: You put it rather well last year, Chair, when we raised the fact that the new state pension is a lot less complicated than the kind of work that we are trying to step through. At Q101, you said: “There’s a lesson for all future Ministers.”

Q73            Chair: Indeed. Ministers, I am sure, ought to be tuning into us all the time. Do not worry, we flag things to them.

Mr Hawthorn, there is a potential tax charge here. If you get a lump sum of £5,000, that could tip you over, especially with the state pension increases, which are tipping people anyway. Is HMRC going to be sympathetic to people receiving a lump sum that they should have had over a long period of time?

Richard Hawthorn: You are right, in that the tax applies at the point when the charge is due, but we are able to apply a level of discretion, and there are discussions going on with the HMRC commissioners to understand the level of discretion we are prepared to apply. Generally, we will be as generous as we can be, recognising the situation—

Q74            Chair: “As we can be”. Are there limitations in law about how much?

Richard Hawthorn: There are some limitations, but we will aim to do what we have done with the LEAP exercise, which has created the same issues. That also creates a tax charge. We will try to apply consistency.

Q75            Chair: That could tip people over the capital amount, if they are claiming some other benefits, potentially—very few, I suppose, because they are pensioners by then.

Richard Hawthorn: Yes, less so on that. It is the tax charge that will be the issue.

Chair: I hope that there is some contrition across Whitehall on this. It is pretty devastating for people. Well, it is good news if they get the money, if they are alive to receive it.

Q76            Mr Djanogly: I would like to look at PIP underpayments, which have risen sharply from 3.8%, or £570 million, in 2021-22 to 5.1%, or £900 million, in 2022-23. It seems that most of the rise was due to recipients of PIP failing to let the DWP know that their medical needs had increased. Mr Schofield, can you comment on why medical needs have increased so dramatically in that period? Why would people be unlikely to update?

Peter Schofield: That is a good challenge. We touched on a bit of this at the hearing we had back in July on the health transformation programme, and I will come on to some of the links to that in a minute. We have seen a significant increase in demand for PIP over the last year. There has been quite a dramatic increase in claims. The year before, there were 715,000 claims, and that rose to 851,000 claims last year.

The other factor to be aware of with these statistics is that the last time we measured underpayment in PIP was in 2019-20, before the pandemic. We used to do a sample based on face-to-face interaction with claimants, which we were obviously unable to do throughout the pandemic. We wanted to work out if we could do it through telephony, and that took us a while. It has been three years since we last measured this.

Q77            Mr Djanogly: Sorry, is that separate from the automatic health reassessments that have been frozen, or is that the same point?

Peter Schofield: It is separate from that; it is the way that we sample to do the statistics. My point is basically that we are seeing two effects. One is that there has been a big jump from last year to this year, but we had not measured it last year: we based the assumption on a percentage that we rolled across from 2019-20. Maybe the increase in underpayments has been a more gradual thing that has happened over the three years since we last measured it.

The other thing is that we are seeing this growth in demand for PIP across the whole of DWP. That brings in your question about why that might be happening. We think that it may well be a range of factors, but it could well be increasing ill health across the wider economy and wider society playing into more people making claims. You then have another point: “Okay, if there is a tendency for more ill health throughout society, why are people not coming forward and letting us know?” That plays into the conversation we had back in July, because I fear that one of the things we are trying to do more of, and one of the reasons why we are taking forward the health transformation programme, is improving trust in the way that we operate. If customers are on the standard level of PIP, but their condition has worsened in a way that would probably entitle them to an enhanced level of PIP, we want them to feel confident about coming forward promptly to let us know, and not to feel that there is a risk because they do not trust or understand the system. They do not come forward because they worry that they might lose what they have.

Together with the health transformation programme, as we said in July, we are working on trying to improve trust. We are trying out using case management to help us to get alongside the customer, and help them to understand the decision that we are making, so that the whole PIP process does not feel like such a mystery, as it does for many people at the moment. It is really about making it easier for people to report a change of circumstance, and helping them to trust the system, but being aware of the fact that rising levels of ill health across the economy may be driving more demand, and aware of the worsening of conditions among people who are PIP customers already.

Q78            Mr Djanogly: Thank you. I can see that it is a complicated scenario, but at the same time as that is going on, there has been a significant backlog of health assessments building up during the pandemic. How is that interacting?

Peter Schofield: There was a big increase, but as the NAO says in the Report, that has fallen away. We are back down to the new claims, which have typically been at about 13 weeks. The other aspect of that is trying to ensure that we prioritise cases where someone phones up with a change of circumstance. That is what we are talking about. We take that through the same overall journey, so we are trying to ensure that that is also done quickly.

Q79            Mr Djanogly: But you have paused automatic health reassessments.

Peter Schofield: There is a wider issue there around health assessments where someone has come up for a reassessment. The best way to deal with the underpayments that we have described is for the claimant to let us know promptly when there has been a change of condition. Also, many PIP awards are for a finite period, and then they come for reassessment. What we have tended to do when there has been huge demand for our service is prioritise new claims; there has been an increase in the wait time for reassessments because we prioritise new claims. Again, as we increase our capacity and make greater progress on this, we are able to address both.

Q80            Mr Djanogly: Because you have cleared the backlog, you can now look at increasing the number of reassessments.

Peter Schofield: Yes, we have prioritised the new claims journey, but as we reduce the amount of time back to standard levels, that frees us up to go into the reassessments that are due. I always say this to customers, and I will use this opportunity to say it as well: if you are a PIP customer and you feel that your condition has worsened, please let us know because you may be entitled to an enhanced level.

Q81            Mr Djanogly: But you are not talking about keeping the number of reassessments low because of sensitivity issues—that is not the reason for it.

Peter Schofield: No.

Q82            Mr Djanogly: On targeted case reviews, paragraph 22 of the AG’s Report says that the “most significant component of DWP’s counter-fraud plan by value is a £443 million project to review millions of universal credit cases to root out incorrect payments. DWP expects this project, which it calls Targeted Case Reviews to generate some £6.4 billion of savings by 2027-28.” That project is going to cost about £443 million. The AG also speaks about whether you have enough people to do this. I think we are talking about 8 million reviews. Do you have enough people?

Peter Schofield: Yes, we do. I might bring in Neil in a second, but I should be clear: the £443 million is for the spending review period, so that takes us up to March 2025. There is more detail on page 41 of the NAO Report, but you will see that the whole programme lasts through to 2027-28, and that gets you to the £6.4 billion that we are talking about. We got additional funding for additional people, and we currently have about 1,700 folk working on the targeted case review. There is an opportunity to meet some of our colleagues doing that work in Basildon. Neil, I think you were in Tŷ Taf near Cardiff seeing teams.

Q83            Mr Djanogly: So you are talking about doing it within existing resources? You don’t need to train up more people to do it?

Peter Schofield: Yes, we do; there is a massive recruitment process. Neil, do you want to say a bit about the process?

Neil Couling: In the space of a little under 20 months, we have gone from zero people doing this to 1,764 at the end of August—close to that. We are trying to get to 2,000 by the end of this month, and then we will keep growing ultimately to about 5,930, from the top of my head—that sort of order of magnitude. We are not going to use all civil service resources to do this. We announced this recently. We are just talking to the market about augmenting probably about 3,600 or 3,700 civil servants with some private sector resources.

It is a big operation to get going. We have grown it in universal credit style, so we have grown it from small acorns, learning as we go along, and developing the products, functionality within the system and everything we need. But we are on track, and, as the National Audit Office Report says, the results from it are encouraging. We now have to increase productivity, as the NAO says. We have to develop the learning from this as well. One thing that it does is give us intelligence on what’s going on and feeds it into our continuous improvement programme within universal credit. It is a big task and a big project—I am SRO for it—but we are on track and I am encouraged about how it is going. It is certainly finding the fraud and error in the system. It is also finding some underpayments.

Q84            Mr Djanogly: You may have answered my second question there, which is that you have this huge project going on, you are building up staff and you—DWP—have found that around 30% of claims reviewed so far by targeted case reviews are incorrect, which is similar to the level of incorrectness found by random sampling exercises, so obviously my question is: why the project? Why do you need to spend all this money if the result is the same as if you kept it random?

Neil Couling: Well, if we kept it random there is a chance that, on those sorts of volumes, we miss fraud. We may be able to get better as this gets on, so we may be able to get a higher hit rate. But right now I am kind of resisting the urges of lots of people around me to go faster and snatch at success here, because what I do not want it to do is to turn into a process. I want intelligent inquiry, and when I was in Tŷ Taf and you, Peter, were in Basildon, this is what we were saying to the staff there: “I want you to get the cases right. It’s no use to me that you’ve cleared”—

Q85            Chair: This is all laid out in figure 11, isn’t it?

Neil Couling: Yes. I am trying to hold people’s nerve across Government pressuring me to say, “Go faster, Neil. This really seems to work. If you could just go faster, you could crack fraud quicker”—if I could, but it depends on me keeping the quality there, and we are still learning. We are still developing the service, and I will not have finished that until we are ready to go to market, and can put this out there for—

Q86            Mr Djanogly: I assume what you are saying there is that you learn more from the targeted, rather than just looking at data coming in?

Neil Couling: Yes. The truth is that we will target as things happen, so we will spot something maybe in some cases. I will keep some random cases going on anyway.

Mr Djanogly: You will? Okay.

Neil Couling: Partly that will allow me to do a double-check on the measurement we do for fraud and error, because it is the same approach as our fraud and error measurement, but it will also, by its randomness, allow me to find new things that in the targeting we are not finding. It will always be iterative. We will target—the clue is in the name—but we will also do enough random so that we are picking up new frauds or error, and so forth.

Catherine Vaughan: I was just going to add, hopefully for clarification, that in the fraud and error statistics we are taking a sample to understand what is going on in the wider population from a fraud perspective and then acting on those cases. What Neil is doing here is applying that logic at a much bigger scale to then intervene on more cases, rather than just from a sampling and measurement exercise. So it is coming up with similar answers from a percentage of fraud and error in the system, but this intervention allows us to tackle a much greater volume of actual cases of fraud.

Q87            Chair: Ms Hillyer, this work is going on with someone coming to you because it has been picked up, but how does the loop go round so that you say, “Yes, these are ones we need to target more?” and how do you then incorporate that? Is this still part of the learning?

Bozena Hillyer: It is very much part of the learning. This project is really important to us—tactically because it is about finding the fraud and error in the stock, and that is one of the things we have to do, but strategically because we need to understand how it got in there, so that we can design it out and therefore reduce the flow. So it is really important to us.

We work together very closely on it. As Neil said, he is the SRO and I actually chair the project board for targeted case review, so it is a very collaborative, cross-Department, multidisciplinary effort to make sure that we take all of the learning, because as you point out, streams of work will come out for my counter-fraud officers, and eventually it will end up in debt and, indeed, possibly in other parts of the Department as well. We are building it slowly to make sure that we understand the best way of making that work, and the actual work itself comes through the integrated risk and intelligence service, which is part of CFCD—counter-fraud, compliance and debt—so it is a whole-Department effort. We have to do all of these things. Not one of the things will get us where we want to be; it is doing all of it together that is going to get us there.

Peter Schofield: It is also worth saying that we identify underpayments as well. Going back to what we were talking about in terms of early warning on underpayments, there is the opportunity to learn that way as well.

Q88            Chair: And to iron those out. Mr Couling, you touched on the issue of everyone wanting you to do it faster to get more money. Any politician or, indeed, permanent secretary would be keen to see that happen. When the deal was done with the Treasury on the money coming in—the £900 million—to get some money back, of which some is being spent on this, was that funding agreed on the basis of a target of what you were going to save? How much do you have to prove to get the next bit of money?

Peter Schofield: If you turn to page 45, figure 12, you will see some of the metrics that were taken into account.

Q89            Chair: And you agreed that with the Treasury.

Peter Schofield: Yes, so the £6.4 billion overall assumes hit rate only marginally increasing, but productivity per agent increasing quite significantly. But we have time to get there. I am probably in the category of people saying to Neil, “Please get on with it, please can you go a bit faster?”—I am coming clean on that.

Q90            Chair: I am just looking at the footnotes here on productivity, which say, “New reviewing officers are assumed to be underproductive whilst in training then to clear 0.5 cases per day”. It takes two days to do a case—am I reading that right?

Peter Schofield: Yes, but ultimately it should build up—because it is detective work and it is complex. Any one caseworker may have a number of cases under way at the same time, so they have asked questions on one and are waiting to hear back on another.

Neil Couling: We spoke in June about learning from major projects, and your report came out the other day—it is a very good report. In setting the agreement with the Treasury, in the assumptions we made on productivity I have given ourselves time to learn and do this properly. The fact that I knew I had to recruit nearly 6,000 people across a period of time meant that that allowed me to put some phasing in here.

Q91            Chair: And you know how long it takes to recruit those people.

Neil Couling: I know how long it takes to recruit those people. Often, there’s a slip twixt cup and lip there. If the Committee wants full disclosure, I am about two weeks behind on recruitment at the moment, but that is not material in a five-year project. It is something we watch very carefully through the project board, don’t we? In this plan working with the Treasury, the key things are: where is the hit rate, what is happening to productivity, is it improving? Everything I know after 37 years of leading people in this organisation is that as people become more experienced, they become more productive. We have seen that through the LEAP exercises as well.

Q92            Chair: This is interesting. You can call us cynics on this Committee, but we look at a number of projects where a number, like in a quadratic equation, moves somewhere—up usually—and optimistically in order to justify the end result. Has there been any pressure on you? You must have been grilled by Treasury officials to push harder. You have been around a long time, Mr Couling, so you are quite robust to push back on that. Was the push there, though?

Neil Couling: The push was there, yes; you imagine correctly, Chair. But my ally in this was the Office for Budget Responsibility, which has to make decisions around central forecasts and the like. In terms of what the original estimates were for this activity and what appears in the forecast for the Government, there was action there to scale back any optimism bias. They are forecasts. In ’27-28, I might not be down to the last penny and shilling and everything here, but I think the profile is an achievable profile.

Q93            Chair: You are also on a bit of a hamster wheel, because you have new fraud and error coming in and you are trying to clear out the old fraud and error. How fast can that hamster wheel go? When do you get to the point where you are tackling it faster than the new fraud and error is coming in?

Neil Couling: This is Bozena’s point. This is one of the things we are doing. It will have a big impact. The high numbers on fraud and error will not start to come down in any marked way until we hit big volume here, but the bath is filling up as I throw out water.

Q94            Chair: Big volume is £6.4 billion in ’27-28.

Neil Couling: Yes, and we have a series of other responses that are about stopping the fraud coming in, such as the disrupt activity that Bozena leads and the prevent activity that we will take. We are developing work, which is also in the NAO’s Report and our accounts, that looks at how you spot activity before it happens or just as it happens, rather than picking it up, which is what targeted case review does.

Q95            Chair: What is interesting about this, as you say, is that it is forecast that there are moveable elements to the calculation. There is always some fraud that could be perpetrated from outside, but even Ms Hillyer might not have noticed that that was coming, so there will be challenges there. We recently did a report on fraud across Government, and HMRC, DWP and Defence came out with relatively glowing colours from the PAC on actually having a fraud function. Other bits of Government do not. When you look at this, are you ensuring that the learning goes in through the Cabinet Office so that other Departments know what is working and what is not so that we do not see over-optimism?

Peter Schofield: We do. We have regular conversations with the permanent secretary for the Cabinet Office. He wants to know what we are doing and understand how we are learning from our experience and sharing that. The Public Sector Fraud Authority is a really good way of doing that; we talked a little bit earlier about the operations there. It is clear that we face huge challenges in DWP. It is nice of you to say what you said, but actually it feels like—

Q96            Chair: The accounts were still qualified.

Peter Schofield: The accounts were qualified. The propensity for fraud and error is—

Neil Couling: The headwind is very strong.

Peter Schofield: The headwinds are strong. I do not think that there is more that we could do, to be honest, on putting resource into this. The targeted case review is a huge, huge intervention.

Q97            Chair: My point is that it is an expensive intervention, so you have to have something to prove for it, or credibility and—

Neil Couling: It is an expensive intervention, but there is payback on it, as the National Audit Office points out. To be fair to the National Audit Office, it has been saying this to us for a number of years. It has been challenging us on this. I sat down faced with the figures that we had two years ago and said, “What can I do about this?” The lightbulb went on in my head that we could do this measurement. We did 3,700 cases or similar in the sample. I said to the team, “What if we did, I don’t know, a million?” They went, “Do you think that is possible?” and I said, “We would need funding”. That was the basis for putting this together: the National Audit Office’s challenge of us when faced with this. This is an intervention that is working and will work; we just need a bit of time to get it up to the volumes that we want.

Chair: We want to share your optimism. In 2027-28 not all of us will be around this table, but our future Committee will be challenging you on this.

Q98            Ben Lake: Mr Schofield, you just referred to one of the factors that you face—Mr Couling referred to it as the headwind—which is the underlying propensity for fraud. I understand that one of the assumptions made as part of your forecast is that there will be an annual increase of some 5% in that underlying propensity. Can you elaborate a little on the reasons why you made that assumption and on what basis you made it?

Peter Schofield: It is a judgment that I had to make. There is not a number out there that you just pick on. We looked at a number of different elements that we could learn from. To give you a couple of examples, we work closely with CIFAS, which is the fraud sharing database. It found that over the last year there has been an 11% increase in fraud against organisations.

We have talked a lot about our work with the Public Sector Fraud Authority. It does a report each year on the cross-Government fraud landscape. It found an increase of 7% in fraud outside of tax and welfare. You have seen a number of areas elsewhere where there is this relentless rise. More broadly, if you look at the ONS crime survey, 41% of all crimes are related to fraud, so we are seeing this type of crime going up and being an increasing element.

I looked particularly at those 11% and 7% numbers. I felt that we needed to take account of the fact that, more broadly, fraud is being better detected than before. That might give you a case for going down a little bit from that, so 5% seemed like an appropriate number to base our forecast on. We then worked with the OBR, because this is taken into account in its twice-yearly forecast, and that was a figure that it felt comfortable with. The numbers that you see in that chart in the NAO Report are based on the 5%. I hope that gives you a bit of a sense of how we got to that number.

Q99            [1]Ben Lake: Mr Hawthorn, are you making a similar assumption in HMRC?

Richard Hawthorn: As Peter said, we do not forecast a propensity to that. Perhaps the equivalent figure is for our tax gap, and that is holding. In fact, if anything, it has come down very slightly this year with the number of interventions we are making; at the moment the tax gap estimate for ’21-’22 is around 4.8%—slightly down on the previous year.

Q100       Ben Lake: Finally, Mr Couling, given all the headwinds and problems that have been encountered, when do you expect or hope that fraud and error in universal credit will fall to the 6.5% outlined in the business case?

Neil Couling: I think I explained this last year. The 6.5% figure in the business case is implicit rather than a target. What it relates to is the difference in the counterfactual between an assessment of fraud in the legacy system and the assessment of fraud in UC. There is no reason to think that that delta has changed, so we could still achieve the savings. Fraud might well have moved up in the legacy system and then the reduction from UC would still be a reduction; you will all see the headline figures for fraud and say that that cannot be true, but it is. That is just how the business case works. So it is not a target. Peter and I have said before that it is a thing we would like to work to because it is sort of understood out there.

So it still is possible to achieve my responsibilities as SRO and deliver the business case and have higher levels of fraud than 6.5%. The key thing to do here is to understand just how low you can get it. The point that Peter made earlier was that nobody has said to me or Peter that there is something we are missing, something we are not doing in this fight against fraud. If we were missing something, we would want to know and we would take a look at it. It will be what it will be. There are, as I said, particularly at the moment, strong headwinds here affecting all organisations trying to tackle some of this.

Peter Schofield: To build on this in terms of what we are trying to do, we start by being very clear about measuring what fraud and error is out there and being better at doing that. As the NAO say, we are world leading or up there in terms of the transparency of our data. This year we put more data than ever before in the public domain, including new experimental work on valuing the prevention activity that we do. The first thing is to be able to measure it; the second element is to be able to identify the causes of fraud and error. You will see in the annual report and accounts, particularly on universal credit, that we go through the main causes of error to give us a sense of where it is coming from and where the challenges are. Then we put into place the strategies to address that—that is, a fighting fraud and error plan.

We talked a lot about the targeted case review. That tries to identify the fraud and error that is already in the system, but we can learn from that to make the barrier into the system ever stronger, particularly picking up on the changes of circumstances problem. As we do that, we evaluate the effectiveness of our strategy. That is in the conversation we had with the National Audit Office about whether we can demonstrate that we have a cost-effective control mechanism for managing fraud and error in the system. It might be that we are doing everything that you could possibly do and fraud and error is still not down to the level that we would want it to be, but for me, as accounting officer, the key thing is to be able to be very clear that we have a control regime that is cost-effective and I do all I reasonably can to keep fraud and error low.

Q101       Peter Grant: Mr Couling, you have told us previously that you are beginning to look at the possibility of using artificial intelligence machine learning so that it flags up potentially suspicious patterns in claims. Do you appreciate the concerns of some of the people who have submitted evidence to us that, first of all, the machine might learn to do it wrong and carry on doing it wrong, but more importantly, the machine might learn to do things in a way that builds in unintentional bias. Do you understand why people are raising those concerns?

Neil Couling: Yes, and I share their concerns. I do not want to preside over a system that does that. The Department is doing two very important things to try to mitigate that risk. The first is that we always put a human being at the point of decision making about a claim, and if we are using any of this information, we put to the decision makers false positives, as well as cases that we think are flagged for attention. The second thing we are doing—are in the process of doing—is that we are taking this carefully, again, in the universal credit way, with small numbers at the start so that you can observe what is going on. You can then run analyses against this to check for bias.

The systems are biased; they are biased to finding fraudsters. For example, when we looked at UC advances, we were finding a bias against older people in there. You are paid less under age 25 and then you are paid more. People were presenting for advances as if they were older than 25 to try to get more money, so the system was throwing up more older cases; there was a bias against age, but of course there was, because that was the very fraud we were trying to identify. So we check on this.

The Committee said to us last year that, in terms of gaining public confidence, we need to set out our plans for how we do this. We will do that; we have said we will come back in November to set those plans out. We will make public our findings from our analyses around bias, to try to maintain public confidence, because public confidence is vital in this. These tools and techniques could really unlock this problem—as I was saying before, stop the new frauds coming in—so it is vital that we take the public with us on this, as well as all the politicians and Committees like this one, but we also need some space to explore and learn a bit. We are not at the point where we can share all that with you now.

Q102       Peter Grant: I appreciate the point that there is always a human being involved in the decision, but one of the comments that the Comptroller and Auditor General makes in his Report is that there can be a detriment to somebody just from the fact that their claim has been picked out for further examination, because it could mean, for example, that the claim is delayed. Is that a fair comment to make, and are you taking steps to make sure that if an innocent person’s claim is picked out for closer examination, the innocent person will not lose out by their claim being either reduced or delayed as a result?

Neil Couling: I am nervous about speaking for the Comptroller and Auditor General, as he is sitting so close to me, but he is pointing out a hypothetical risk; there is no evidence of that risk manifesting now. But it clearly is a hypothetical risk and it is not something that you would want to be happening to individuals who are making legitimate claims. The kind of approach we want to take will, hopefully, sort the sheep from the goats here effectively. I am taking it very slowly because I think you have to have high levels of finding the right sorts of case before you release this into bigger numbers of cases. That is what we did with the UC advances model. We tested it, and it did not work terribly well the first time. We learnt with the machine learning, and then we tested it again; it was a bit better. We then tested it again. We did all this on small numbers of cases, until we got to a point where it was working at a very high level of efficacy, and then, at that point, we released it into wider use. I tend to follow that model.

Peter Schofield: All I would say is that you have to see this tool in the context of a wider delivery system. You have to get the accuracy of the tool to a level where you have enough colleagues who can process the alerts that come through, to avoid precisely the risk that you have just described, Mr Grant, of unnecessarily holding up someone who should be getting their money. We do not want to roll this out until we know that the machine learning has got to where the level of accuracy is such that we can then handle all the cases that come out, which we were able to do on advances fraud in universal credit.

Q103       Peter Grant: How do you teach the computer the dangers of discrimination on the basis of what on the surface does not look like a protected characteristic, but could be a proxy for one? For example, there may be particular patterns of family life that are more common in different cultures, or some sections of the population may be more likely to change jobs or change address a lot more often than others. How do you make sure that a frequent change of address within a claim, which might appear to be something suspicious, does not end up meaning that people from a particular background or with a particular set of protected characteristics are more likely to be flagged up, simply because they are the ones who have particular changes in family circumstances, such as moving around a lot?

Neil Couling: I am going to take the fifth on how we will do that, but that is what we will do. The reason why I am taking the fifth is that I do not want to expose how our models will work to the outside world, because unscrupulous people will work their way around them. But you make a very good point, Mr Grant; I absolutely agree with what you are saying.

Peter Schofield: It is also worth bearing in mind the work we are doing with the Information Commissioner. Catherine, do you want to say a bit about that?

Catherine Vaughan: It is incredibly helpful that there is independent scrutiny of our approach. We worked with the Information Commissioner’s Office last year, which did an independent investigation into the use of artificial intelligence in the welfare system. It found that there was no evidence that people in the welfare system were subject to any undue harm or financial detriment, to come back to the point raised earlier. It also found that there was sufficient meaningful intervention by humans in the process. Obviously, this is quite an early stage in our work, as Neil and Bozena said, but we expect the Information Commissioner’s interest to continue and we welcome that independent insight into our processes.

Q104       Peter Grant: I understand that you have told the National Audit Office that you intend to start reporting annually to Parliament on the results of what you described as a fairness assessment. So you will basically be reporting to Parliament for us to assess what you have done to ensure that there is no unintentional bias going on.

Peter Schofield: We are going to come back in November in response to that recommendation. Our thinking is that the right way would be to report annually in the annual report and accounts.

Q105       Peter Grant: Okay. Finally, are you able to tell us when you think you will be ready to go live on a big scale with this, or would you prefer not to tell the baddies when you are going to be looking at them?

Peter Schofield: I think you have probably answered the question.

Q106       Peter Grant: Is that a decision you can take yourself? Does it need any ministerial approval or is it just regarded as part of the delegated authority you have?

Peter Schofield: We would work closely with all the relevant authorities in terms of how we take that forward, and Ministers would be aware.

Q107       Sir Stephen Timms: I have a question for HMRC. You said that the tax gap has come down a bit in the last year, but are you seeing an inexorable increase in the propensity to commit fraud among taxpayers as a whole?

Richard Hawthorn: That is not my area of expertise, but I am sure we can come back and give you an answer. As a general answer, we are seeing changing vectors of, if you like, areas of attack. As I say, the tax gap is thereabouts as it has been.

Q108       Sir Stephen Timms: If you or your colleagues do have any information about that propensity, it would be interesting to see.

Richard Hawthorn: We will come back on that.

Chair: I think we are done. I thank you very much indeed for your time. The transcript of the session will be available on the website, uncorrected, in the next couple of days—I thank our colleagues at Hansard for that—and we will be producing our report at some point after the conference recess.