17
Examination of witnesses
Paul Bainsfair, Eric Salama and Keith Weed.
Q70 The Chair: Thank you very, very much for being here. I now have to read out the ‘police warning’. As you will know, this session is open to the public. A webcast of the session goes out live and is subsequently accessible through the parliamentary website. A verbatim transcript will be taken of your evidence and put on the parliamentary website. You will have the opportunity to make minor corrections, for the purpose of clarification or accuracy. I know you all well, but perhaps you would introduce yourselves for the record.
Eric Salama: I am the CEO of Kantar. We are one of the world’s largest research, insight and consulting companies. We are headquartered in the UK. We have about 30,000 people around the world. We work on a variety of issues to help our clients grow.
Paul Bainsfair: I am the director-general of the Institute of Practitioners in Advertising. That is the trade body that looks after the interests of virtually all the advertising agencies in the United Kingdom.
Keith Weed: I am an independent director and trustee of several organisations. I am here as president of the Advertising Association. Until May this year, I was the chief marketing officer of Unilever globally, which is one of the world’s largest advertisers.
Q71 Lord Harris of Haringey: You are the experts. We want to know how effective online microtargeted advertising is.
Eric Salama: The question needs to peel back a bit. What does “effective” mean? Different organisations are trying to achieve different goals. Some advertisers have a goal of driving penetration or the number of people they reach. Other organisations have a goal of improving your propensity to buy something or vote in a certain way. Other organisations have goals to do with increasing the frequency of an activity or changing your behaviour. It partly depends on what you are trying to achieve. Generally speaking, microtargeting is not as good as traditional media such as TV in driving penetration or reach. It is more expensive than TV in doing those things. It is important to understand that any microtargeting advertising depends on a number of factors, such as the quality of the creative. The quality of the creative is more important, from all the research we do, than the targeting itself.
It is not just about the media. The choice of media and the targeting accounts for around 16% of the impact of the ads, depending on the category and what you are trying to achieve. Other things, such as promotions and the quality of the creative, impact it quite a lot. Generally speaking, the best results come from combining different media: combining digital with TV, combining print with outdoor. The combinations are the most effective.
Lord Harris of Haringey: Can I clarify that? Does the analysis you have done account for the fact that, for a microtargeted social media campaign, the amount you would spend on the quality of the content is much smaller than for a cinema campaign?
Eric Salama: It looks not at the production costs, but at the costs of buying the media. Reaching 1,000 households is more expensive in digital media than on TV. It depends on what you are trying to achieve, in answering the question of whether it is effective.
Lord Harris of Haringey: What I am trying to get at is whether, as a recipient of it, I am more likely to be influenced by something I have received on my mobile phone as part of my social media, or something I am not paying much attention to, sitting in front of the television or in a cinema, catching up before the main feature.
Eric Salama: Funnily enough, people do respond when they are watching TV. It is not as poor as that. TV is more effective at driving awareness and certain brand associations. If I make the analogy — and Keith can talk about his experience at Unilever — with commercial advertising, TV helps you expand the number of people you are reaching. Online microtargeting is not as good at expanding the number of people you are reaching. It is better at reinforcing perceptions you have among the people you are reaching.
Paul Bainsfair: It is probably worth thinking about it, as we all receive these messages. If you see something on television, maybe it is not at the front of your mind, but you know it is not aimed directly at you. This is a mass broadcast. When you see a microtargeted ad, particularly one that has followed your activity — say you have been looking at a holiday in Greece and then an ad pops up saying, “We run holidays to Greece” — it is much more likely to cause what we call activation, because it is specific to the frame of mind you are in and what we know you are interested in. But the advertising itself is less likely to be entertaining or to have a long‑lasting effect on you, in terms of the brand doing the advertising. That is more likely to come from mass media — television, for example.
Lord Harris of Haringey: But your organisation has called for an interim ban on microtargeted political advertising.
Paul Bainsfair: Yes, because we feel that microtargeting for products and services is a perfectly harmless and sensible way to go about marketing. If you are into golf and you see an ad for new golf equipment, that is something you will be interested in. From research we have done, we know that people like that. When you are receiving messages of a political nature, sometimes it is not clear who is advertising to you. Sometimes you do not know whether what they are telling you is true. Sometimes you do not even know whether it is part of the manifesto or policy of the people who are advertising to you.
We think that is very dangerous. As an analogy, a hammer is very good for hammering in a nail; it is not something you would want to use as a weapon. Microtargeting for perfectly ordinary products seems to make perfect sense but, when you use it in a slightly opaque and dodgy manner, it is not a good thing.
Keith Weed: Targeting has been around in advertising for ever. Using the analogy of the holiday, you would be happy to hear about a promotion on a holiday to Greece if you were thinking about going to Greece. Similarly, if you have just bought a car and someone advertises for car insurance, that is a good thing. On the other side, if you are 90 years old and you receive ads for baby milk or nappies, that is a waste of money for the advertiser and annoying for you.
All microtargeting does is to take that on a step. It is also called interest-based targeting, which explains it better. The word “micro” has a technical aspect, but think of it as interest and behaviour-based. All it means is that, if I know you are a vegan and you enjoy recipes, cooking, et cetera, as an advertiser, I would not want to send recipe ideas to you with meat-based products, because it would not be very useful for you or for me.
People are learning about this interest-based approach to advertising. To give some data, the European Interactive Digital Advertising Alliance did a big piece of research in March 2019, across five countries, the UK included, and 5,000 people. It concluded that people understood this area of personalisation much better than we may collectively think they do. It sort of makes sense. We see it in all sorts of things. With Spotify, you get offered music you might like or your friends like. Amazon gives you ideas of things you might want to purchase. This is increasingly part of everyday digital life. The Interactive Advertising Bureau in the UK, in a piece of research in 2018, found that 69 per cent of people were willing to share data in return for free services and content.
The real point, which was made at the beginning by Eric, is that targeting is a good thing for both people, if done well. You do not want to see irrelevant stuff; you want to see helpful stuff. The advertiser of products or services wants the same. The quality of the ad is the main thing that drives its effectiveness. I am sure it annoys all of us if we receive a bad‑quality ad. All targeting does is to better deliver the ad to the right person. If that is done well, it is win‑win; if it is done badly, it annoys the member of the public and is a very poor financial decision for the advertiser.
Q72 Lord Scriven: This is a growing area. In the evidence that was given, you say that, while it might not be in interest‑based or micromarketing, UK digital advertising spend is greater than Germany, France and Italy combined, so it is a big area. If you are advising clients or the industry, where is the biggest return on investment, particularly for interest‑based marketing, and where is there not really a return on investment, assuming that you have good-quality content for both this and offline?
Paul Bainsfair: We at the IPA have something like 45 years of evidence on how advertising works. It has been accumulated over time and we have a rich databank. We have had academics piling through it, to work out the answer to that very question. All the normal caveats apply. Rather like Eric was saying earlier, it does matter what sort of advertising is going on and what sort of product or service is being advertised. Generally speaking, we have come to the view, from all the data we have looked at, that a split of 60 per cent on brand-building, mass-market advertising and 40 per cent on microtargeted or activation-based advertising gives you the perfect mix for the best return on investment. That is generalising but, looking at all the cases across time, we would say you should be spending 40 per cent on microtargeting or, to give it another term, sales-activation advertising, and 60 per cent on brand building.
Lord Scriven: Clearly, that 40 per cent is not mass market. That is not a return on investment. What are the drivers of me going online for this type of interest‑based or microadvertising? Is it behavioural? Is it emotional? What is it that works to get me to buy or do what you want, other than the content?
Eric Salama: I am just looking at some data. We test more advertising around the world than any other organisation and we look at the impact on the awareness of the brand and in terms of associations. Does it change your view about the organisation that is advertising? Does it lead to an action? Does your propensity to behave differently, to buy differently, change in some way?
As I said earlier, if you look at pure effectiveness and return on investment, it is most effective where you combine media. It is not a single medium. Then I can look at the data we have and say, “TV is more effective at driving awareness and saliency than any other medium, generally speaking, in terms of the return”. I can look across radio, outdoor, online video, online display, newspapers and magazines, Facebook and cinema, and the numbers will vary in terms of what you are trying to achieve. It is almost impossible to answer your question with an overall statement. It will vary according to the goals of the organisation, and it will vary across those media.
Keith Weed: It is a question that advertisers ask all the time, because you want to make sure you invest your money well. It is a well-founded question. This notion of whether digital is better or worse than offline went backward and forward quite a lot among advertisers. Eric’s point is well made. The industry has what it calls “multi-screen planning”. You see it yourself. When you are at home watching television, how often do you see people in the room also on mobile phones? One of the things that people tweet and comment about most is what is on television, including ads. The ability to optimise and plan across multiple screens is something that the industry strives for. It is not either/or. For the particular task you are doing, which is the best? To answer your question, digital gives you more insight into an individual’s interests and behaviour than mass media, where someone is sitting watching the television.
Lord Scriven: Why do we in the UK spend more combined than Germany, France and Italy? What is it about the UK?
Keith Weed: We are ahead. I did a global role until recently. The most developed digital markets in the world are the US, the UK and China. You can argue about who comes first, but these markets are very well developed. If you look at e‑commerce, similarly, the UK is very well developed. All you are seeing here is a development curve, and you will see other markets developing as well.
Eric Salama: Keith mentioned Twitter. I do not know how many of you watch “Strictly Come Dancing” on a Saturday night; it looks pretty much like a “Strictly Come Dancing” audience. Tweet activity and TV are totally intertwined there. If you want to understand the effectiveness of Twitter and TV on a Saturday evening, or a Sunday evening when the results show is on, the audiences are totally intertwined. It is very hard to look at Twitter in the context of what is happening with “Strictly Come Dancing” and TV, and separate them out, because they are symbiotic. They feed off each other.
Q73 Baroness Kidron: I am interested in whether you would make a distinction between contextual advertising online and targeted advertising online. I understand that, while targeting uses and collects a great deal of data, contextual advertising online is similarly effective.
Paul Bainsfair: Could you elaborate on what you mean by contextual advertising?
Baroness Kidron: It is not directed at an individual person, but just in the environment they happen to be in.
Paul Bainsfair: I am not sure we have any data one way or the other on that. Common sense suggests that it would be effective. Whether it is more effective or less, I could not say.
Baroness Kidron: I would be happy to provide some.
Keith Weed: You were talking earlier about brand building and brand awareness being 60 per cent of investment, and sales activation being 40 per cent of investment. Those two things can be done offline and online. You can do brand awareness campaigns online, not necessarily targeting individuals.
Baroness Kidron: This may become more relevant when we get on to political advertising.
Q74 Lord Lucas: How do you measure the effectiveness of political advertising, particularly of the directed kind? How good are those measures, really?
Eric Salama: On the commercial side, we measure the level of activity of political advertising in a market such as the US. I can provide the Committee with a paper on political advertising trends in the States, where we have done the analysis of the last few campaigns, not just federal but state and local. There is data on how that has changed. We are measuring the amount of advertising and the spend, and how they have changed.
Lord Lucas: Yes, but how effective is it?
Eric Salama: We do not measure the effectiveness of political campaigns. We measure the effectiveness of commercial campaigns, by looking at people who have been exposed to advertising and relating that to changes in their perception of the brand and their motivation to purchase or behave differently. We look at direct sales impact. We will match people’s exposure to a campaign with whether they actually bought something. It is both an attitudinal measure and a sales measure, but we do not do that in the political sphere, so I cannot tell you how effective political campaigns are.
Keith Weed: What is an effective ad? One side is about creating a favourable perception, and that might be awareness or some attributes and associations. The other is about a desired action taken. You went straight to the sales, which is the ultimate one for a commercial proposition, but a desired action could be a test drive or signing into a website. Looking at a combination of those two, I would argue that commercial and political advertising are quite similar in how you might want to think about them.
From experience, as Eric says, the best way to test an ad is A/B testing. You have a sample here to whom you show an ad, and a sample here, exactly mirrored to your original sample, to whom you do not show an ad. Which is better or worse? That was quite difficult in previous days, because you would have to run a whole TV campaign, which was very expensive, and you would always find the results afterwards. In the digital world, you can do A/B testing quite easily with smaller groups. In fact, they now start talking about A/Z testing: you can put out many, many ads and test them. Those approaches are open to commercial and political testing. The proxies we use are reach, changing of attributes, click-throughs, et cetera, but the best thing to do is a test, side by side.
Paul Bainsfair: How we can monitor the effectiveness of political campaigns is difficult to say, because they are not really in the public domain. We measure effectiveness for conventional packaged goods and other products and services. Interestingly, we also measure campaigns that the Government run for stopping people smoking in October or stopping people drinking and driving. In the public opinion area, we have measured the effectiveness of such advertising over time. Using all the techniques we have heard about, we can show that advertising is very effective in shifting public opinion. It is not much of a jump from there to political advertising, but that is the only area we have a lot of data on.
Eric Salama: It is harder to measure. When you are measuring the effectiveness of advertising, let us say, for Unilever’s products, it is a continuous stream of activity. You have lots of things happening, lots of intermediate measures and final measures, and lots of individual sales. In a political campaign, ultimately, there is just one measure, which is a vote. That happens infrequently, so there are not as many things. You could look at intermediate measures, in terms of willingness to behave in a certain way, to take part in civic activity or other things. But, ultimately, I think you are asking about impact on voting activity. That happens very infrequently and no one really knows how anyone has voted.
Q75 Baroness Kidron: Can you tell us something about the data used to target audiences for commercial advertising and where it comes from? Do you know or imagine that there are different data sources used for political advertising?
Keith Weed: There are basically three types of data available for anyone, commercial or political, to pull from. They are called first-party, second-party and third-party data.
First-party data is data directly owned and understood by a body. That might be people on websites or you might have a direct-to-consumer shop as a brand or service. You know then quite a lot about that person. It is a one-to-one relationship and they might come back to buy more of that product, et cetera. Second-party data is data you get, say, from a retailer, Sainsbury’s or whatever, if you were a branded goods business. Third‑party data is data from Nielsen, Kantar or whatever, which you buy into.
This is available to all. The big differentiator is expertise to understand and manage it. It is Eric’s day job to do that. In that way, insight wins. All that brand advertisers are trying to do is serve their consumers better. Just like a vote for politicians, in a buying decision, they either buy your product or someone else’s product. Understanding consumers better in order to serve them better is how businesses either grow or become irrelevant. Companies, therefore, have lots of market researchers and marketers analysing this. They employ experts such as Eric and his company to get better insight.
One should be focusing less on the availability of the data, which is available to everyone, and more on the acquisition and leverage of the data, which is much harder than it might appear, because of the vast amounts of it. There has always been data there. There is now even more data and we have machine learning to help us analyse the data, but we are not yet where we would want to be.
Paul Bainsfair: This might be behind your question, but we are all leaving behind evidence of our internet browsing data. These so-called cookies are there on our computers. The real‑time or programmatic buying that allows relevant ads to appear in a fraction of a second is based on that information. That is done by computers analysing data in, as I say, a fraction of a second. That is a new thing for us, compared to where we were 10 or 20 years ago, in terms of targeting.
Eric Salama: I will answer the second part of your question first. The data available to political institutions and parties is no different from that available to commercial organisations. That data is available to everyone. It is really a matter of what people care about and what they want to use in order to achieve their objectives.
Political parties in the US put much more store on voter records, which are available there. In over half the states in the US, voter registration is based on affiliation as a Republican, a Democrat or an independent, so you already have something about the person’s political affiliation. There are voter records showing past voting behaviour from people. The political parties in the US rely more on that.
Now, different political parties will rely more or less. To go back to 2016, the Clinton campaign spent six per cent of its media budget on digital. The Trump campaign spent 40 per cent of its media budget on digital. The same data is available to everyone, but people make different choices about what to use, coming back to Keith’s point about where they feel they are going to get the most insight.
A lot of column inches and TV programmes were made about Cambridge Analytica. The segmentation approach that Cambridge Analytica took, called Ocean, is publicly available. It was not invented by Cambridge Analytica, but Cambridge Analytica used it. I went online before coming here today, and you can print off the questionnaire and the attributes online. Cambridge Analytica used that to decide whether to target certain types of ads at certain people, and to relate it back to people.
Coming back to Keith’s point, all this information is available more or less to everyone, if you really care to dig for it. Some of it requires you to survey people and actively ask them for their opinions about things. People will fuse data. They will fuse people’s IDs, demographics, voting behaviour, purchase behaviour, whether they are employed, what they have bought, what their attitudes are. Organisations can do that if they have the resources and the will. It then depends on what use they put it to and what aspect of that data they care about most.
Baroness Kidron: We have already heard that you think political ads should not be targeted at individuals. If it is all available to everyone, there is the question of who has the money to apply the expertise, but where is the potential democratic deficit in this system? Why are we all sitting here worrying about the use of data, from your perspective?
Eric Salama: I am not answering as the CEO of Kantar. I suspect it is partly because, if you think about the way that advertising has grown up, in this country or globally, it has been regulated. If you want to put an advert on TV, it needs to be pre-vetted. If you want to put an ad in the newspaper, it needs to be pre-vetted. What is probably causing concern is the amount of misinformation, fake news and advertising that is not regulated. You are not quite sure who it is going to, how it is getting to those people and whether it is true. It is to do with a world that we are used to, and then a world where things can be at an extreme scale that we have never been used to, in an unregulated way.
Baroness Kidron: At what point in the value chain would you regulate? I assume it is the content, but would you then regulate the delivery mechanisms?
Eric Salama: You say “regulate the delivery mechanisms”. What do you mean?
Baroness Kidron: I mean the targeting and the other factors. It is not simply the content. When you say “regulate”, do you mean the content itself, the fact-checking and who paid for it? Would you also make an intervention in how it gets to the consumer or citizen?
Eric Salama: I am going to see if anyone else wants to answer.
Paul Bainsfair: I will say something, to give you a chance to think. This whole issue of regulation is a difficult one. You had the ASA in earlier, and you know that the content of political ads is not regulated. Speaking for the IPA, we believe there should be a register. It should be transparent and probably machine-readable, because there are millions of ads of a political nature, as we know, so that it is possible to find out who is saying what. That is probably as far as it is feasible to go, because it is not practical to regulate and pre-vet millions and millions of ads, as we do with conventional ads.
Keith Weed: I totally understand the concern about political advertising and that people want to work out how to make the system better. I would argue that political parties have some responsibility to take this forward, along with the Electoral Commission. People often turn to the digital platforms and ask what they are doing about it. That always perplexed me a little. Why should they be responsible for campaigners’ behaviour? In the same way that advertisers pay for the ASA and then stand up to the idea of being legal, decent, honest and truthful, a similar approach to political advertising would be a great reassurance to the public.
Q76 The Chair: This interests me a lot. Baroness Morris stood for election in her constituency and at the time there was a very precise limit to how much she could spend, as a constituency candidate, on her campaign. Now, regrettably, we have accidentally stumbled into a form of ‘Citizens United’ situation, whereby through micro-targeting individual voters you can avoid those campaign cost restrictions. In fact, we have run a coach and horses through what was the accepted means by which we ran our elections. It is lamentable. It is doubly lamentable that the Electoral Commission made recommendations to prevent abuse 16 or 17 months ago and nothing has happened. What could this Committee be recommending, or pushing for, to at least to move us back to the type of level playing field we thought we had?
Eric Salama: I cannot speak for everyone who works at companies such as Google, Facebook, Twitter and Snapchat, but everyone I have met in those companies is decent and has a sense of morals. They are in no way irresponsible people who are not trying to do the best thing. We are dealing with issues of incredible scale. They have spent a lot of money to deal with brand safety issues and to make sure that ads do not appear next to certain types of content. Advertisers have whitelists and blacklists: “We can only have our ads appear here; we will not have our ads appear there”. Huge progress has been made by both advertisers and platforms to improve the brand safety situation. It is virtually impossible to make sure that something does not appear somewhere 100 per cent of the time, but things can be done to make what we have at the moment better.
The Chair: I think you are saying, and I would agree, that the situation that has been created is very difficult for the digital companies, because of the lack of regulation. As Facebook, I would be saying, “For God’s sake, Government, work out some sort of co‑regulatory process with us which we can sit behind, and that the citizen can have confidence in”. I hope we will begin to move towards that.
Q77 Lord Scriven: Paul, you raise this in your written submissions. The basis of your submission is that democracy is about competing ideas, people being exposed to them and being able to make a rational choice about what is best for them, their community and their country. You say in your written evidence that one issue we might have to look at is this: “We have not set a fixed threshold under which targeting would be defined as microtargeting. We volunteer to help create a mechanism for calculating thresholds”, which is welcome. Is there anything else that needs to be thought of, so that this concept of democracy, of people being exposed to competing ideas and making a choice about what is best for them, can be preserved? That is what we need to ensure. It is not just that we have people of good intent, but there are unintended consequences of this technology, in how it disrupts democracy, and we need to address them. One idea here is that microtargeting might need a fixed floor.
Paul Bainsfair: Behind our comment is this fear we have. You have heard the term “echo chamber”. It is this aggravated political polarisation that goes on when you target groups of people who have certain attitudes. We have seen people pushed to the extremes of what starts as a fairly right-of-centre or left-of-centre point of view. This seems to have been enabled by microtargeting. As for whether one goes so far as to completely ban microtargeting by political parties, we would say it should be considered.
Lord Scriven: Is there anything other than microtargeting? Based on your collective expertise, if we are going to have this concept that democracy is about people being exposed to different ideas, is there anything else we need to change or think of, to help us get back to a level playing field, so that that is the basis of our democracy?
Keith Weed: To directly answer your question so there is clarity, you were talking about one aspect. An ad has two things: the content and the delivery mechanism. It is fair to say that the public would assume that commercial and political advertising were held to similar standards. As I said before, commercial advertising is held in regulation to being legal, decent, honest and truthful. There needs to be consideration of what political advertising would be held to, for the confidence of that very debate.
Eric Salama: To give you a couple of examples, at the moment in the UK you can go on a website and see who has paid for certain ads politically. You cannot see which audience segments they were trying to target. I see no reason why the regulation should not say you need to publish the audience segment you are targeting. If there was nothing to worry about in the audience segment you are targeting, you would not have anything to worry about. From a transparency point of view, you could make clearer the way in which microtargeting is happening, so it is published: “It was this group. It ran this ad for this amount of money on this platform. The programmatic buy, the audience it was targeting, was this”. You could see whether it excluded anyone or targeted certain people. You would hope that the transparency led to an improvement.
There are other things you can do. Some are to do with the amount of time, let us say, Google has to check whether there is misinformation, discrimination or anything else going on in an ad. The argument is that, due to their volume, we cannot pre-vet them all, which is probably true. But you could delay things to give the human and machine checkers enough time to check that before the ads are aired. It happens at the moment. If you want to advertise on Facebook in a commercial setting, it takes time for Facebook to approve the ad to go on there. There is no reason why you could not say, “Until this political ad reaches a certain number of viewers, or has been viewed a certain number of times, we will not allow it to go on”. That would give the machine and human checkers time to check things.
Q78 Baroness McGregor-Smith: Reflecting on your comment that all these executives at Google and Facebook are really great and decent people, they have spent year after year arguing that they are only platforms and are not responsible for the data that sits on them. Yet they are all involved in microtargeting individuals with advertising in both the commercial sense and the political sense. If they are so great, why have they done nothing?
This is the thing I do not understand about this debate, if they are so moral, so decent and so good, they will know the impact it is having on citizens in countries around the world. They will know what it is doing to elections. They will know about the lying we have seen in political campaigns. Why are they not doing anything about it? Although we say they have done this and that, they have not. Today, as a citizen, I cannot go on to any of their platforms and find out what they know about me or who they have told anything about me, if I have ever been on them. If you are on social media and someone knows something about you, they have not held their hands up and even said they are responsible for that. I am just interested that you think they are all so great. They might be nice people, but they are not. That is because they think they are running a tech platform.
Eric Salama: I do not think I used the word “great”.
Baroness McGregor-Smith: You said “decent” and “good”.
Eric Salama: No, I said as individuals they were decent people with a sense of morals. I happen to believe that they are publishers, not just platforms. I do not think those two things are in contradiction with each other. As to why they have not done anything, you need to ask them as opposed to me.
Paul Bainsfair: We all know that Twitter has done something about it. It happens to be very much smaller, but we support its point of view in saying, “We will not take any of these political ads”. As Eric said, it is impossible to answer for the other giants.
Baroness McGregor-Smith: On the basis that the regulation is not there, and it will take years for regulators to catch up and Governments to work out what to do, who should take the lead? Should it be Governments? Should it be us or them trying to make sure this is dealt with properly and people are not exploited, which ultimately they have been?
Paul Bainsfair: Speaking personally, I think the Government need to step in and tell them to do it.
Keith Weed: First, they have done a huge amount. I have been working with them closely over many years. I have been amazed by the scale of the challenge and how they have engaged with it. I totally understand the frustration. Everyone would prefer for us to be in a different place than we are right now. That is part of what happens when you have rapid evolution. Before, when new industries evolved, it was in humankind years. This is rapidly evolving, with enhanced technology. Every minute, 500 hours of user-generated content is uploaded; that is 82 years of video a day, or 2.5 billion daily contributions across Facebook’s different platforms. The scale is massive and that is part of the reason why one might feel frustrated. Why have they not got their hands around this? Are they not big companies? It is a very big area.
AI, artificial intelligence, I think will ultimately help. Right now, it is very good on text and static pictures. It is not so good on videos. It is very good on spam and fake accounts. It is less good on the nuances of bullying and misinformation. That will improve. What you need to add is a lot more humans. Facebook now has 14,000 people working on this, and likewise YouTube has bolstered this area a great deal. To the point on political advertising, as I said, I do not think it is their responsibility to hold political parties to account. The political parties and the Electoral Commission should jointly create a framework to hold themselves to account.
Q79 Lord Black of Brentwood: Before asking a supplementary, I want to draw attention to the fact that I am a director of the Advertising Standards Board of Finance and sit on the council of the AA.
To pick up on Eric’s very important point about publishers versus platforms, through the Online Harms White Paper, the Government are going to almost contorted lengths to deal with these issues. Some of the issues this Committee is grappling with are highly complicated. Would it not make life a lot easier if the Government simply legislated to say, “These are publishers, not platforms”, and made them subject to the laws that all other publishers are subject to, including electoral law?
Eric Salama: I do not know whether it would be simpler. Algorithms are not totally neutral. They drive certain content to certain places in certain ways. The companies we have been talking about have taken steps to make sure that those algorithms work in as good a way as possible. I am not apologising for them or saying they are great. They have made huge efforts in those areas, but algorithms are not neutral.
Lord Black of Brentwood: They are also not transparent.
Eric Salama: They are not transparent. Personally speaking, I think they are publishers. They are not just platforms on which stuff appears. I do not know what the solution is, whether it is simpler to do what you have suggested or whether there are other means to do it.
The Chair: Would you agree, Paul?
Paul Bainsfair: Yes, I would. They provide a fantastic service. I am not sure what the unintended consequences of taking that step might be, in limiting the way in which they provide many people with a fantastic service, as I said. We all just tick the box to say “I agree” when we go on to a site, because we want to use the product. Whether that would no longer be the case, if we were to take that step, I do not know. But, in principle, I agree with you.
Keith Weed: This is a hugely complex issue. Looking at the community guidelines that most of these platforms have, I think we would all sign up to them. The challenge is how they enforce stronger compliance with their own community guidelines, which would deliver some of what you are looking at.
Q80 Lord Holmes of Richmond: How effective is platforms’ algorithmic detection of advertiser-friendly content? Are you concerned by the suggestions that YouTube’s algorithmic detection of advertiser-friendly content discriminates against LGBT and other content?
Keith Weed: What I said before is relevant to this answer, about the amount of content and the current effectiveness of artificial intelligence, which is getting literally better by the day. Given that it is an evolving area, my answer to your question today and in three months’ time would probably be very different. It is literally happening in real time.
There are various ways that platforms currently help in this area of advertising-friendly content. The first is to have algorithms with exceptions: “We do not want to be next door to a disaster”, for example, or nudity. The advertiser can help define that. The second is on premium content, where there has been greater scrutiny of the content to try to make it more advertising-friendly. The third is monetisation. As Eric was saying earlier, it is making monetarisation more of a privilege than a right, so things must be viewed and checked accordingly.
Those are the ways that advertising-friendly content currently works. Advertisers want to be diverse in their approach and engage with the population at large. The last thing they would want is for diversity to be stifled. Is there the potential for this to happen? We are in an evolving situation. I have heard comments that would support your question and we need to work this through in real time. Some of the artificial intelligence uses words that come out of the very communities you are referring to. That is how that happens.
Paul Bainsfair: There were lots of issues with brand safety and ads appearing next to pretty dangerous content. The platforms have done a very good job in trying to solve that problem, but they might have been overzealous in applying some of these algorithms, which might have led to the problems you are talking about.
Lord Holmes of Richmond: On that, neither of you thinks it is deliberate.
Paul Bainsfair: It is absolutely not.
Keith Weed: It is definitely not. It is an unintended consequence of taking action to address an issue that was very real.
Q81 Lord Lipsey: How often do publishers or platforms, whichever you want to call them, turn ads down?
Keith Weed: To my knowledge, it is not very often. If an ad was illegal or indecent, that could be a reason. Indeed, if an ad had viruses, worms, Trojan horses, et cetera, they would not want to bring those on to the platform. In the normal placement of an ad, there is the ability to scrutinise, but I think we are talking about programmatic advertising here. If you would not mind me explaining what happens with a programmatic ad, you arrive on a platform and you have an interest in cooking or vegan ice cream—I do not know—and, from the moment you are on that site, that information is then going out: “Who would like to buy an ad for this person?” There is bidding between computers, and then that ad is served up.
When you are on a site, you might think, “This site is choosing to give me the ad”. No, the reason why you are seeing that ad is because of you, not the site. That particular site has not chosen the ad; your interests and behaviour has brought that ad to you through a programmatic buy. There is not a screen in that sense.
Lord Lipsey: I do not find that exactly reassuring.
Keith Weed: You did not ask for reassurance; you asked me to answer your question.
Paul Bainsfair: The very first thing that Keith said is the case: very few ads get turned down straight off the bat. They do take them down. If there are complaints and the ASA is alerted, it will contact the platforms and they will cease to run them, but it is not pre-vetted, as we were saying earlier. A lot of ads get up there that perhaps should not.
Lord Lipsey: If they are political ads, which is after all the focus of this Committee, the ASA cannot do anything. It is nothing to do with them. That makes it even more worrying that these ads can go out without being pre-vetted, apart from possibly by an AI computer.
Keith Weed: A commercial advertiser would not want the censure of the ASA saying it had done something that was not right.
Lord Lipsey: Political parties do not suffer that hazard. Therefore, they can lie as much as they want.
Q82 The Chair: Can I ask a question, one word that has not cropped up very much is “reputation”? Keith, your career has been spent promoting and protecting the reputation of your brands. The big issue around here, this morning and yesterday, was the non-publication of the Grieve report on Russian interference. How troubled are you that you could quite accidentally find your brands being promoted adjacent to some clear disinformation that has been microtargeted at the same audience you are seeking? All of a sudden, you are faced with the possibility of reputational damage. You are the paymasters to this entire industry. Yet you are effectively funding the process that could ultimately do extraordinary damage to individual brands or reputations. First, do you have any solution to that? How do you feel about that risk?
Keith Weed: The reputational risk of being placed against inappropriate content is a genuine concern. Any brand advertiser would be concerned about where that ad appears. It could be an airline not wanting to advertise in the middle of an airline disaster film, or a life-giving vitamin ad not being on an outdoor poster next to a funeral parlour. This notion of adjacency is real.
As for whether there is reputational risk in social media, which I think is the direction of your question, if the risk was strong, people would not be advertising. They are advertising, because the controls in place are doing, so far, a good job in that area. There are controls in many areas. Companies have their own controls in their own right. Media agencies have their own controls. I am going to pass over to Paul to talk about some of the industry controls as well. There are controls at all different levels.
You are right. The way we look at brand safety is twofold. The first is being adjacent to content that you do not like or on a website that promotes unsafe or divisive behaviour. The second is funding a site that you would not want to be on. There are a few industry initiatives. GARM, the Global Alliance for Responsible Media, is one. The IAB, the Interactive Advertising Bureau, gold standard is another. Managing this on a daily basis to minimise risk is the responsibility of all brand advertisers, because they would not want to have what you described.
Paul Bainsfair: I touched on this earlier. We have been involved in calling for particularly Facebook and Google to join the cross-industry brand safety initiatives that have been going on. They have done that. We are in a much better place on this topic than we were a couple of years ago. There are still problems, as we heard earlier, to do with the sheer volume of stuff being uploaded all the time, but we are pretty happy that progress is being made here. In the end, it is a conflict between trying to find the audience you are after and the adjacency of the content. Because, as we were saying earlier, these decisions are being made by computers in fractions of a second, despite all the checking, things sometimes go wrong. It is a reputational risk for the brands concerned. I think I am right in saying that some big manufacturers pulled their advertising from the platforms for a while, to indicate their displeasure at this and give them a boot up the you-know-what, to make sure things changed.
Eric Salama: As we have all said, a lot of progress is being made, and it is not just about appearing next to fake news or hateful content. People have blacklists to make sure they do not advertise on sites that are non‑GDPR compliant or webpages that have certain keywords. There is a lot more activity to make sure that you do not appear on certain sites. The platforms and the advertisers have done a huge amount on that.
Ultimately, to achieve 100 per cent safety, you would need the opposite of the blacklist, which would be to say, “We will advertise on only these sites”. Programmatic advertising, as Keith was saying, does not work that way. From a cost point of view, people are buying space on thousands of different sites, as opposed to saying, “We will advertise on only these sites”. For the industry, it would be cost prohibitive to do that. You have a conflict. If you really want to get to 100% brand safety, the cost of doing that would be huge.
The Chair: Some of the views of the Committee have become apparent this morning, including during the evidence of the ASA. I absolutely accept that 99 per cent of the people working in this area are decent and reasonably well motivated. I totally accept that. As an ex-ad man, I know the fact checking mechanisms that have been put in place over the years are pretty robust and, together with the work Keith has done, they ought to promote an environment of trust. Here is my worry: having done all that work, if I were a bad actor looking to place a very destructive message, I would seek out the most trusted place I could put it, in order for it to be effective. I am concerned that we do not have the mechanisms in place to spot and prevent that bad actor.
Q83 Baroness Morris of Yardley: If you cannot answer that one, maybe you can answer this one. It is just a round-up question at the end, really. From everything you said, if the Government could do just one thing to improve the regulation of digital technologies, what would you suggest it might be?
Paul Bainsfair: I will go first, because we have already stated our wish on this one. We should have message transparency, achieved by a platform-neutral, independent, real-time, machine-readable register of all advertising: the messages, the content, the targeting, and the spend. It would be possible to go to a place and know unequivocally who has said what to whom, and how much they have spent doing it.
Eric Salama: We have talked about transparency, as Paul has just mentioned. I assume your question is more about political activity than commercial activity.
Baroness Morris of Yardley: Yes. In the context of our report, what might be a strong recommendation for us to put forward?
Eric Salama: Personally speaking, not in a corporate capacity, I think that having a more level playing field between political and commercial advertising would be beneficial. I do not see why there is such a distinction between the standards to which commercial companies are held and those to which political organisations are held. That affects the organisations and the places where those ads are placed.
Keith Weed: We spent a lot of time, quite rightly, talking about the new world of targeting, data, et cetera. Overwhelmingly, the impact of an ad is driven by its content more than its targeting. The delta or the change is in targeting, but this is about content, as Eric was saying. The most important thing I would like to land is that, if we are going to build public trust in advertising, this notion that political advertising is held to a different standard to commercial advertising would not be well understood. If commercial advertising needs to be legal, decent, honest and truthful, as I think the public would accept, I would argue that political parties and the Electoral Commission should look to create an environment for political advertising to be held to similar standards.
The Chair: On behalf of everyone on the Committee I am very, very grateful for your evidence this morning. It is hugely helpful to us. We are beginning to pull together some of these threads, which the Government habitually refer to as the ‘’wicked issues. You have made it a lot easier to identify them. Thank you very much indeed, all of you.
