Supplementary written evidence submitted by
Dr Emma L Briant, Senior Lecturer at University of Essex
The evidence below draws together additional interview material from my academic interviews with the Leave.EU campaign and further develops the key questions explored in my earlier evidence to this Inquiry. I conducted interviews with key executives from Cambridge Analytica (CA) and Leave.EU as part of my upcoming academic research on migration, political campaigns and media narratives (See my previous co-authored book Bad News for Refugees). Below I examine further the evidence on Leave.EU, Eldon Insurance and US Health Insurance Industry.
Evidence on Leave.EU, Eldon Insurance and US Health Insurance Industry
Findings of my interviews confirm that work was performed by CA for Leave.EU, and that they designed the strategy that was deployed by Leave.EU. While my interviews were inconsistent on how far CA was involved in Leave.EU, they illuminate key questions raised by ongoing investigations of both. Here I consider:
1. CA's level of involvement in the Leave.EU campaign, whether and how CA methods were used
2. What data was used in the Leave.EU campaign and how was it deployed
3. What was the relationship between Leave.EU campaign and Eldon Insurance and how does this help us understand current involvement in the insurance industry in UK and US?
Cambridge Analytica’s level of Involvement in Britain’s EU Referendum:
Several investigations have been exploring whether services were provided by Cambridge Analytica and not declared in the Leave.EU campaign – this has been denied by both parties.
The UK’s Electoral Commission are fining Leave.EU for not declaring services received by Gerry Gunster. I interviewed Andy Wigmore Communications Director for Leave.EU in October 2017, and he shared, 'they [Cambridge Analytica] didn’t give us a little box of toys and say, there you are, have a go. They just said look, if-- you gotta prepare for this because if we come in, this is what we need and what we want -- we want to do it' (Interview: Wigmore/Briant, 4th October 2017). From Brittany Kaiser’s testimony and the fact an invoice was issued for working on the UKIP component of the work in Cambridge Analytica’s Leave.EU proposal – which Arron Banks then gave UKIP money for (see the Guardian’s reporting), it does seem like Cambridge Analytica at least thought they had provided a service for which they had been hired.
In my early conversations with Brittany Kaiser it was clear that she did not know that Andy Wigmore had continued to deploy the strategy Cambridge Analytica designed for them when they lost the designation. It became apparent when Kaiser first showed me documents that she subsequently submitted to the British Parliamentary Committee on Digital, Culture, Media and Sport Inquiry into Fake News, that the methods and strategy Andy Wigmore described to me in interview had remarkable similarity to the plan she designed for them (this she provided in her evidence to the British Inquiry and I have attached it here for easy reference). Indeed the Leave.EU team told me they copied it. I interviewed Gerry Gunster, an American strategist on CA’s involvement in Leave.EU and he stated as follows: 'Cambridge Analytica, although they were involved early on, they sort of gave a bit of a backbone on how to do behavioural targeting and micro-targeting…’ and with psychographic targeting, ‘they provided some backbone for how to do it and then a lot of it was just kind of handed over to the campaign staff' (Interview: Gunster/Briant, 4th October 2017). Leave.EU Communication Director Andy Wigmore’s statements regarding CA have been inconsistent, but referring to the Cambridge Analytica method he also told me, ‘probably yes - it probably was useful because we copied it.' (Interview: Wigmore/Briant, 4th October 2017).
Deployment of Cambridge Analytica’s strategy by Eldon Insurance Staff
Cambridge Analytica had produced this dual strategy for using UKIP and insurance data to produce benefits for the insurance company. Wigmore told me Leave.EU had Eldon Insurance employees deploy the plan copied from Cambridge Analytica, with four actuaries, two marketers and a graphics team running the campaign all out of the same address (Interview: Wigmore/Briant, 4th October 2017). Wigmore also explains how this was done:
‘“So, some of the things they [Cambridge Analytica] did tell us, which were-- which were-- we did copy. And no question about that, is about, you know, these small clusters, this you need to find out in the - where these people are and what matters to them. And what we were able to deduce from that, and remember, um, ah, and as an insurance company you have actuaries that work for you. Actuaries are brilliant, they’re mathematicians. So if you give them a problem and you say right we want to look, here’s, here’s some stuff. What do you think of the probabilities. They will-- came up with the probabilities of the areas that were most concerned about the EU and we got that from our own actuaries. We had - we have four actuaries which we said right, tell us what this looks like from our data and they’re the ones that pinpointed the twelve areas in the United Kingdom that we needed to send Nigel Farage to.’ (Interview: Wigmore/Briant, 4th October 2017).
Phase I of Cambridge Analytica’s plan seems to have been carried out with Cambridge Analytica – the plan submitted to the UK Fake News Inquiry by Former Cambridge Analytica Business Development Director Brittany Kaiser states under Phase I: ‘Significant work has already been completed, and this should take no longer than ten days including sign-off from Leave.EU and its partners on the content of the political questionnaire’. The following also sounds like Phase I of the Cambridge Analytica plan: ‘We started by doing the biggest poll we possibly could. We polled at least 10,000 people and found out what the main concerns were. So the main one that stood out like an absolute sore thumb was immigration and what, you know, what came up in immigration was housing, schools, overcrowding, never racism… and people didn’t say they didn’t say they didn’t want immigration, they just didn’t like immigration. Because it was changing their lifestyles.’ (Interview: Wigmore/Briant, 4th October 2017).
My interview with Andy Wigmore seems to indicate that Leave.EU also deployed parts of ‘Phase II’ of Cambridge Analytica’s plan – Phase II involved ‘online surveys’. He states: ‘from a standing start we sent out on facebook ‘if you are interested in leaving the EU… or are you against the EU? Support us…' so people ‘liked’ which is nothing… can’t do anything with that… OR you can join… and you pay money to join, so we ended up with close on 200,000 members. And a lot of, almost a third of our money - which is a lot - we spent almost £8 Million - was got from small donations. Anything from a pound to a thousand pounds in some cases. If they join… we asked them two things: would you like to be polled …and’
EB: But this wasn’t on facebook though -
AW: ‘No no no, on facebook you recruited them, so if they clicked on a link, then they came through to our website and they filled out ...then you’ve got the data. And that we were using anyway but we didn’t need to give it to Cambridge Analytica or anything like that.’ (Interview: Wigmore/Briant, 4th October 2017).
Importantly, my evidence shows that Leave.EU copied and were able to deploy an effective campaign based on Cambridge Analytica’s methods following the pitch that Kaiser gave them. This raises questions of whether other entities who received a similar pitch could also have replicated the methodology - this is of particular importance in relation to Lukoil for example, a Russian state-owned company that Cambridge Analytica pitched their methods to around the time SCL were delivering training in methods to Eastern European countries to ‘counter Russian threat’.
Leave.EU, UKIP, Eldon Insurance and the US Health Insurance Industry
Andy Wigmore claimed it wasn’t possible to deploy Cambridge Analytica’s AI methodology without access to ‘electoral roll data which you can then use [...] Because Cambridge Analytica artificial intelligence requires data - if you don’t have it, you can’t do it.’ - this implied that they couldn’t do CA’s artificial intelligence because they didn’t have enough data (Interview: Wigmore/Briant, 4th October 2017). But Wigmore later told me:
AW: ‘So. Um, so, if you take… You talk about data. So you have a lot of data when you’re an insurer. And that data is, it’s, there’s layers and layers and layers. You know, you have, um, ah, lifestyle data, of course you do. You have, um, credit check data which of course you do. All that data you put that together, the way you can actually then make risk against an individual is incredibly strong. So our ratios, in other words, you always say, right, if we take on this customer what are the chances of them making a claim within a twenty-four year- uh, twelve month period, twenty-four month, is it high? Because they’re so de- so low social demographic grouping, they don't have a proper job, they have to take out a loan to pay for their car insurance, you can make a risk assessment of that. So makes it more expensive.’ (Interview: Wigmore/Briant, 4th October 2017).
Subsequent evidence indicates that the campaign may have used both Eldon Insurance’s data and UKIP data. Asked about whether they might have used a political database for commercial purposes by the Observer, Banks said ‘Why shouldn’t I? It’s my data.’ Andy Wigmore confirmed in interview with me, ‘We used our data. Gerry took our data, which we’d… and we started to look at polling.’ (Interview: Wigmore/Briant, 4th October 2017). Both of these datasets were factored in to the original Cambridge Analytica plan which Brittany Kaiser submitted as evidence which indicates how the Leave.EU ‘psychographic survey’ would be used to build ‘psychographic insurance’. While Andy Wigmore was happy to admit copying Cambridge Analytica’s methods for the Leave.EU campaign, he sought to present the Leave.EU team’s deployment of Artificial Intelligence and knowledge of how to draw on this for Eldon Insurance as an apparent happy accident resulting from their own visionary brilliance:
AW: ‘Let’s say the referendum’s just finished. What we discovered, we were actually quite bloody good at artificial intelligence. And we’ve applied what we learned in the referendum to our business model for insurance. (Interview: Wigmore/Briant, 4th October 2017).
But this indicates they deployed the complementary workstreams part of Cambridge Analytica’s dual strategy. If, as this evidence shows, Cambridge Analytica’s plan was indeed used it is hard to understand how this would not be considered a gift in kind or receipt of services, particularly as Kaiser described seeing the datasets of the political campaign and insurance company being used in parallel during her visit to their offices. My own evidence and that of Brittany Kaiser directly contradict the conclusions drawn by the Electoral Commission which declare ‘the relationship did not develop beyond initial scoping work’.
The desire to take credit for developing the expertise themselves may have been because they are now selling their expertise in the United States insurance market. Wigmore explained to me:
‘So for- for risk and, um, making sure you get the right person who’s not gonna be fraudulent, not gonna make claims they, you know, you know so much about them. So we’ve started an operation in Ole Miss University in Mississippi which is the centre for artificial intelligence in the world, who knew? […] And, and, and the guy that runs it, he’s like the most extraordinary data scientist. So when we explained all our sorta, d’you know what, you should have a chat with him. When we explained what we did- […] He thought we were completely nuts. He said, how on earth did you create? This is how we created it. He couldn’t get his head around it. But, if you took the application of what we learned in, you know, learned, self-learning, you know, with the ads that we did, what worked, what didn’t, who responded, how they responded, et cetera. Et cetera. And you apply that kind of learning to-‘ (Interview: Wigmore/Briant, 4th October 2017).
The algorithms developed through the ‘complementary workstreams’ allowed them to evaluate risk for their insurance company ‘imagine if you had the ability to say, this person is absolutely brilliant. We know as an insurer, it’s worth giving them a good price because we know they’re unlikely to make a claim, they’re solid, they’ve got a good job, um, you know, they’ve got a family life, blah, blah, blah, blah, blah. So that in artificial intelligence terms is the holy grail in insurance. So that was a by-product of what we discovered, brilliantly. And that’s all about data. That is all about data. So um, that was- that was the upshot. So we’ve set this up in Mississippi. It’s been going for nine months, we’ve been testing for twelve months now, testing all the insurance against it and it’s extraordinary.’ (Interview: Wigmore/Briant, 4th October 2017). This was lucrative, Wigmore said the disparity was ‘Massive. Massive. Our loss ratios have dropped by about 13 - 14 per cent. And in- in insurance terms that’s millions of pounds. Millions. Because what happens, for every pound of insurance you give someone, you have to put two pounds into like an escrow account to cover just in case that person ever makes a claim. It’s called solvency. Very dull, very boring. But, so if you imagine now actually I only need to put a pound in against two pounds because we’re confident that person isn’t going to make a claim…’ (Interview: Wigmore/Briant, 4th October 2017). Asked about how reliable it was Wigmore said:
AW: ‘Yes. This is the thing. So, in a twelve month period, you take out insurance and in the twelve month period you renew and you haven’t made a claim because you’re a good client and you’re likely to renew because you have that type of-, and when you shop--’
EB: So, taking that back to campaign... it was that reliable?
AW: ‘No. We only discovered that after the campaign when you analyse everything that happened. Why did people like-- so our polling was 0.01 percent out.’
EB: You couldn’t analyse it while it was going on?
AW: ‘No, we didn’t know about it. It was only when we looked back and went, crikey, we’ve got something here. Because our polling was 0.01 percent out in the twelve weeks. 0.01 percent!’ (Interview: Wigmore/Briant, 4th October 2017).
AI in the US Health Insurance in Rust Belt States
Wigmore described how he continues to go to Washington once a month as Arron Banks’ business partner and ‘We are still doing the insurance, the artificial intelligence in Mississippi has been very interesting, really interesting, because- […] what they are going to start to look at is healthcare because addressing an individual, you are so individual, it is like insurance -- how do you know what’s wrong with that person, what kind of insurance... health insurance (because that’s how America works) do you have?’ (Interview: Wigmore/Briant, 4th October 2017). Wigmore said ‘We’ve proved it in motor insurance and they are extrapolating those, these are data scientists.’ Which raises the question of whether these data scientists at Ole Miss have the British data, or whether they are reverse engineering modelling and algorithms developed using British citizens’ data (Interview: Wigmore/Briant, 4th October 2017).
Criticising Obamacare as ‘one size fits all’ and open to exploitation by insurers, he said, ‘Big insurance companies, have now gone ‘okay’, they need to come up with something different, so the two biggest in insurance they are starting to think differently, so they are the ones who are piloted these little schemes in the rust belts, where poverty is absolutely [unclear] and actually as a benchmark for trying to sort the healthcare system, which you will never do, it’s a good start, so artificial intelligence would allow those insurance companies to make their assessments quickly based on knowledge which is already there, it just needs to be gathered.’ (Interview: Wigmore/Briant, 4th October 2017). Wigmore framed this in individualistic terms of helping people to help themselves and the state making provisions for prevention in the cases of high risk individuals ‘every state has a duty of care’ to ensure they are ‘making people aware and helping them to help themselves’ (Interview: Wigmore/Briant, 4th October 2017). He said what matters is ‘this person smokes therefore he is high risk, therefore we need to treat him differently and pathway him differently etc. […] based upon a much stronger knowledge.’ (Interview: Wigmore/Briant, 4th October 2017). The AI advancements may be seen as a method to identify high risk cases for health insurance, those groups in poverty in the rust belt perhaps (where some places have been hit by an opioid epidemic). On identification of risk-associated predicted unhealthy behaviour, if they are relocating responsibility on the individual and state to reduce those risk-associated behaviours this may call into question whether or how they would be insured, which is essential for healthcare provision in the United States.