blog: Don Marti


Nudgestock 2018 transcript

03 July 2018

(This is a cleaned-up and lightly edited version of my talk from Nudgestock 2018.)

First I have to give everybody a disclaimer. This is 100% off message. I work for Mozilla. I am NOT speaking for Mozilla here.

If you follow Rory, you have probably heard a lot about signaling in advertising, so I'm going to go over this material pretty quickly. Why does Homo economicus read magazine advertising but hangs up on cold calls? To put it another way why is every car commercial the same? You could shoot the "car driving down the windy road" commercial with any car. All that the car commercial tells you is: if it was a waste of your time to test drive our car then it would have been a waste of our money to make this little movie about it.

There's a whole literature of economics and math about signaling involving deceptive senders and honest senders. With this paper, Gardete and Bart show that when the sender wants to really get a message across, counter-intuitively the best thing for the sender to do is deprive themselves of some information about the receiver. If you're in the audience and you know what the sender knows about you, then you can't tell are they honestly expressing their intentions in the market, or are they just telling you what you want to hear? Anyone who used to read Computer Shopper magazine for the ads didn't read for specific information about all the parts that you might put into your computer. You read it to find out which manufacturers are adopting which standards so you don't buy a motherboard that won't support the video card that you might want to upgrade to next year.

There are three sets of papers in the signaling literature. There are papers that have pure math where you devise kind of a mathematical game of buyers and sellers and see how that game works out. And there are papers where you take users in an experimental setting. Ambler and Hollier took 540 users, showed them different versions of expensive looking and cheap looking advertising that conveys the same information. Finally you've got the kind of research that looks at spending across different product categories, and in this study they found that types of product that have different advertising to sales ratios really depends on how much extra user experience it takes to evaluate that product.

The feedback loop here is that when brands have signaling power, then that means market power for the publishers that carry their advertising, which means advertising rates tend to go up, which means the publishers can afford to make obviously expensive content. And when you attach advertising to obviously expensive content, that means more signaling power. It's kind of a loop that builds more and more value for the advertiser.

Some people compare this to the signaling that a bank does when they build this monstrous stone building to keep your money. Really, the stuff that a bank does, having a stone building doesn't do any more for keeping money in it than having a metal building or a concrete building, but it just shows that they've got this big stone building with their name on it so if they turned out to be deceptive it would be more costly for them to do it. That's the pure signaling model. But the other area that we can see when we compare this kind of classic signal-carrying advertising to online advertising, the kind of ads that are targeted to you based on who you are, is what's up with the norms enforcers?

Rory has his blue checkmark on Twitter which means he doesn't see Twitter ads. I'm less Internet Famous, so I still get the advertising on Twitter. A lot of the ads that I get are deceptive issue ads. This is one. A company that's getting sued for lead paint related issues is trying to convince residents of California that government inspectors are coming to their houses to declare them a nuisance. This is bogus and it's the kind of thing that if it appeared in the newspaper that everyone got to see then journalists and public interest lawyers, and everyone else who enforces the norms on how we communicate, would call it out. But in a targeted ad medium this kind of deceptive advertising can target me directly.

So let me show a little simulation here. What we're looking at is deceptive sellers making a sale. When a deceptive seller makes a sale that's a red line. When an honest seller makes a sale, that's a green line. The little blue squares are norms enforcers, and the only thing that makes a norms enforcer different in this game from a regular customer is when a deceptive seller contacts a norms enforcer the deceptive seller pays a higher price than they would have made in profit from a sale. So with honest sellers and deceptive sellers evolving and competing in this primordial soup of customers, what ends up happening to the deceptive sellers that try to do a broad reach and hit a bunch of different customers is, well you saw them, they hit the norms enforcers, the blue squares lit up. Advertisers who are deceptive and try to reach a bunch of different people end up getting squeezed out in this version of the game. An honest advertiser like this little square down here can reach over the whole board because they don't pay the penalty for reaching the norms enforcer.

So what does this really mean for the real web? On the World Wide Web, have we inadvertently built a game that gives an unfair advantage to deceptive sellers? If somebody can take advantage of all the the user profiling information that's available out there, and say, "oh I believe that these people are rural, low-income, unlikely to be finance journalists, therefore I'm going to hit them with the predatory finance ads," does that cause users to pay less attention to the medium?

Online advertising effectiveness has declined since the launch of the first banner advertisement in 1994. That's certainly not news. This is a slide that appeared in Mary Meeker's famous Internet Trends presentation, and as you can see blue is percentage of ad spending, grey is percentage of people's time. So TV is 36% of the time 36% of the money. Desktop web 18%, 20%, about right.

What's going on with print? Print is 9% of the money for 4% of the time. Now you might say this is just inertia, that that this year people are finally just cutting back on spending money in print because of people spending less time on print and it'll eventually catch up. But I went back and plotted the same slide from the same presentation going back to 2011, and I've got time plotted across the bottom, money plotted on the y axis, and what do we see about print? Print is on a whole different trend line. Print is on a trend line of much more value to the advertiser per unit of time spent than these other ad medium. My hypothesis is that targeting breaks signaling and this means an opportunity.

Targeting means that when you see an ad coming in targeted to you it's more like a cold call. It doesn't carry credible information about the seller's intention in the market.

From the point of view of who has an incentive to to support signal-carrying ad media instead, the people who have an interest in that signal for attention bargain in that positive feedback loop are of course the publishers, high reputation brands that want to be able to send that signal, writers, photographers, and editors, people who get paid by that publisher, and people who benefit from the positive externalities of those signal carrying ads that support news and cultural works.

So if the signaling model is such a big thing then why are there so many targeted ads still out there?


Let's have a look at, just to pick an example, the Facebook advertising policy. As you know, the Facebook advertising platform will let you micro target individuals extremely specifically. You can pick out seven people in Florida, you can pick out everyone who's looking for an apartment who doesn't have a certain ethnic affinity, that kind of thing. But the one thing you're not allowed to do with Facebook targeting is put anything in your ad that might indicate how you're targeting it. The policy says:

ads must not contain content that asserts or implies personal attributes

You can't say, I know you're male or female, I know your sexual orientation, I know what you do for a living. The ad copy has to be generic even if the targeting can be extremely specific. You can't even say other. You can't say meet other singles because that implies that the advertiser knows that the reader is single. Facebook will let you target people with depression but you can't reveal that you know that about them. Aanother good example is Target. They do targeting of individuals who they believe to be pregnant, but they'll pad out those ads for baby stuff with ads for other types of products so as not to creep everybody out.

Back to our shared interest in signal for attention bargain. Pretty much everybody has an interest in that original positive feedback loop of getting the higher reputation for brands of getting reputation driven publishers that'll build high quality content for us. Writers and photographers have an interest in getting paid, and people who are shopping for goods are the ones who want the signal the most. All that stands on the opposite side is behavioral tricks to conceal targeting. Now I'm not going to say this as a privacy issue. I know that there are privacy issues here but that is really not my department. Besides, Facebook just announced a dating site so they're going to breed privacy preferences out of their user base anyway.

Can the web as an advertising medium be redesigned to make it work better for carrying signal? We know from the existence of print that this type of signal carrying ad medium can exist. Print is an existence proof of signal carrying advertising. We also know that building that kind of an ad medium can't be that hard because print was built when people were breathing fumes from molten lead all day.

The prize for building a signal-carrying ad medium is all the cultural works that you get when somebody like Kurt Vonnegut can quit his job as manager of a car dealership and write for Collier's magazine full-time. This book is still on sale with the resulting stories. And of course local news. Democracy depends on the the vital flow of information of public interest. Some people say that the problem with news and information on the web is that it's all been made free, and if people would just subscribe we could fix the system. But honestly if if free was the problem, then Walter Cronkite would have destroyed the media business in 1962. It's a market design problem and a signaling problem, not just a problem of who has to pay for what.

And the web browsers got a bunch of things wrong in the 1990s. There are certain patterns of information flow that the browser facilitated, like third-party tracking, where browsers enable some companies to follow your activity from site to site, and data leakage. Things that that just don't work according to the way that people expect. Most people don't want their activity on one site to follow them over to another site, and the original batch of web browsers got that terribly wrong. The good news is web browsers are getting it right, and web browsers are under tremendous pressure now to do so. As a product the web browser is pretty much complete and working and generic. The whole point of a web browser is it shows web sites the same as all the other web browsers do, so there's less and less reason for a user to want to switch web browsers. But everybody who is trying to get you to install a web browser needs for there to be a reason, so the opportunity for browsers is to align with those interests of users that the browser wasn't able to pick up on previously.

At Mozilla some user researchers recently did a study on users with no ad blocker installed and users within the first few weeks of installing an ad blocker. Anybody want to guess on the increased engagement? How much more time those ad blocker users spend with that same browser than the non ad blocker users? Anybody shout out a number. All right, 28%. From the point of view of the browser those kinds of numbers, moving user engagement in a way that helps that browser meet its goals, that's something that that the browser can't ignore. So that means we're going from the old web game where everyone tries win by collecting as much data on people can without their permission to a new game in which the browser, high reputation publishers, and high reputation brands are all aligned in trying to build enough trust to work on information that users choose to share.

I know when I say information that users choose to share you're going to think about all these GDPR dialogs and I know I've seen these too, and they're just tons of companies on these. To be honest, looking at some of these company names it looks like most of them were made up by guys from Florida who communicate primarily by finger guns. Users should not have to micromanage their consent for all this data collection activity any more than email users should have to go in and read their SMTP headers to filter spam. And really if you think about what brands are, it's offloading information about a product buying decision onto the reputation coprocessor in the user's brain. It's kind of like taking a computational task and instead of running it on the CPU in your data center where you have to to pay the power and cooling bills for it, you offload it and run it on on the GPU on the client. It'll run faster, it'll run better, and the audience is maintaining that reputation state.

The future is here, it's just not very evenly distributed, as William Gibson said. This picture is the cyberpunk of the 1990s. Today all of that stuff he's carrying, his video camera, his laptop, his scanner, all that stuff's on a phone and everybody has it.

Today, the privacy sensitive users, the ones who are already working based on sharing data with permission, they're out there. But they're in niches today. If you have a relationship with those people now, then now is an opportunity to connect with them, figure out how to build that signal carrying advertising game, and and create a reputation based advertising model for the web. Thank you very much.