The “Steal My Money” Lever: A Timely Tale of Data Saving the Day

Rob Collie (00:00): Hello, friends. I originally was going to talk about AI again today, but I decided instead that I want to share a timely story about the power of data that doesn't involve AI. And this story comes from the back office trenches here at P3. Now, we're not an enterprise business here at P3. So a lot of our clients, we run a pretty lean back office team and some of us end up wearing a fair number of hats.

(00:25): For me, one of those hats is being involved in our marketing efforts. Now, I did take some time off from being involved in marketing, but after a year plus of not wearing that hat and wearing others instead, I've increasingly been reintegrating into our marketing work, both in terms of our messaging, which is what we're saying, as well as our analytics, like finding out who's listening, what's getting their attention, et cetera.

(00:48): Okay, so about five weeks ago we started a new LinkedIn ad campaign. Nothing too ambitious, but definitely an upgrade to our messaging. Kind of like a return to our roots kind of thing, but in a better executed manner. And for the first week of those five weeks, we slipped up and had the please steal my money lever in the on position.

(01:08): It's like in that old animated movie, The Emperor's New Groove, where the villains keep pulling the wrong lever in their lair and it dumps them into a pit of crocodiles. And when they get back upstairs, wet and bleeding, they ask, "Why do we even have that lever?" Let's see if we can find that clip.

Audio (01:23): Why do we even have that lever?

Rob Collie (01:25): Oh yeah, that's the one. So why do ad platforms even have a please steal my money lever? Well, in my experience, it's usually not even always just one such lever. Google's Ad Console, for instance, is mostly made up of steal my money lovers. But most, if not all, ad platforms have at least one that's common to them all, and it's the show my ads on partner websites setting. Now, in theory, this is great. Why would I not want my messaging to get broader exposure?

(01:57): And they even tell me in the Ad Console UI that these partner networks are less expensive than advertising just on the main property, just on Google or just on LinkedIn or just Facebook. So more exposure cost less. What could go wrong? Well, I'll tell you what can go wrong. When you turn it on, they steal your money, like hand over fist steal your money. Because partner websites have every reason in the world to have bots clicking on your ads.

(02:29): Every time a bot clicks on an ad on say shadymcshadyclick.com, the owners of that site get paid. Now, LinkedIn gets paid too. But you see, it's not LinkedIn who's doing the cheating, it's the owners of shadymcshadyclick.com. And honestly, even the owners of that site might not know that they're paying for bots, because even they have outsourced to other firms who promise amazing boost to traffic and click rates.

(02:58): And those shady consulting firms are the ones who are hiring the bot networks. So the website owner has plausible deniability. LinkedIn has plausible deniability, and they both walk away with your money. Now, there are many, many honest websites out there who are signed up for the LinkedIn partner network. But since those sites don't generate as many clicks, guess who LinkedIn prioritizes?

(03:23): No one is ever going to investigate shady fourth tier websites like shadymcshadyclick.com. And even if they did, and even if they got caught and punished, they just fire up a new domain tomorrow like shadymcshadyclicks.com, plural, and get right back to printing that sweet, dirty money. Now, in contrast, when your ads are limited to being displayed on LinkedIn itself though, the only entity who would benefit from bots clicking on your ads then would be LinkedIn.

(03:53): And if LinkedIn ever got caught botting their own site for ad revenue, their ad business would tank. Billions of dollars would be lost overnight. And they don't get to relaunch as LinkedIns.com. The stakes are too high for LinkedIn or Google or Facebook to be botting their own properties. Now, as a side note, there is one other kind of entity who would benefit from having bots clicking on your ads on these properties, and that class of entity is your competitors.

(04:25): We've had that happen too, just thankfully not too often. Digital marketing is a lot more of a Wild West environment than you might expect. Okay, so we accidentally and unknowingly had this setting enabled on our new LinkedIn ads for the first week. And since the ads were new and since we put a lot of thought into them and believe they were likely quite a bit better than any of our other recent ads, their performance seemed somewhat believable at first.

(04:50): A week later though and we're looking at the numbers and going, okay, wait, this is too good to be true. And that's when we found the steal my money switch was in the on position. And of course, we immediately turned it off. Okay, so that's the power of data, right? The ability to say, wait, these numbers are too good to be true based on past experience. That's particularly important when you want to believe them. And we had every reason to want to believe these numbers.

(05:17): But without that grounding baseline data of historical performance, we'd have likely fallen prey to our own hubris. Man, we killed it with those new ads. We'd have been too exhausted from all the high-fives we were giving each other to think critically. And if that's where the story ended, fine, it would be a good story and an excuse for me to warn you about the steal my money switch, which is one of my favorite stories to tell. But that is not the end of the story.

(05:43): Because about a week ago, we launched our retargeting campaign. Now, a retargeting campaign is one where you deliberately show ads to people who have already engaged with your previous ads. You don't show them the same exact ad again, but you show them something that kind of continues the conversation that they first engaged with. Now, you see this all the time in your own travels on the internet, things follow you around.

(06:07): Sometimes it's annoying, sometimes it's spooky. But when it's done right, it actually feels kind of good to have the conversation continuing itself without you having to do anything. And that's the spirit we're aiming for with these retargeting ads. So when we fired up the retargeting campaign, you better believe we did not turn on the steal my money switch. We said, nope. Just display this retargeted messaging on LinkedIn itself. Leave the bots out of it. Thank you.

(06:34): Now, remember, retargeting is kind of supposed to perform well. Yeah, more often than not, people still don't click them because they already checked us out and decided something even like, well, not now, but maybe later. And two weeks later is not later enough. But in general, people who clicked on your ads before are still more likely than the general public to click again. So guess what?

(06:56): The early returns from retargeting looked great, even with the steal my money switch set to off. So we had no real reason to distrust it. We were believing it again, for a couple of days anyway. But then I looked at the overall click volume from the retargeting campaign, and then I compared that with a number of legitimate clicks that we'd gotten from the original ads.

(07:18): Now, what's the legitimate click? It's the clicks that came in after we turned off the steal my money switch. And guess what? In a few short days, the retargeting campaign already had more than double the total clicks as we had legitimate clicks on the original campaign. When you think about that for one or two more seconds, you realize, oh wait, this is totally bogus.

(07:41): If the entire retargeting audience is say 1,000 people, 1,000 legitimate people, who'd clicked on your ad during the legitimate window of a few weeks, and you suddenly get 2,000 retargeting clicks in a short period of time, you know something is up. Really no matter how long the retargeting campaign runs, you'd never even reach or get close to 1,000 retargeting clicks because that would be 100% re-engagement, which is just insane.

(08:10): Even if you got to 3%, you would be absolutely killing it. So 30 would be the grand total lifetime on a retargeting campaign that you would expect if you had 1,000 original clicks. 2,000 out of 1,000, that means that everyone not only clicked, but clicked twice. So this sounds obvious perhaps when you say it that way, no one would miss this, right? Except it's all too plausible that it would be missed for a little while.

(08:36): The reason it jumped off the page at me was because I'd baked a page-wide date filter into our dashboard so that for the main campaign, I was only looking at results that came in after we turned off the steal my money switch. If you go back to the LinkedIn Console on the LinkedIn website, the totals there are going to reflect all of the bots. But my dashboard allowed me to very clearly see the total number of legitimate clicks without having to navigate all kinds of crazy LinkedIn website filters.

(09:04): And then with a cloned dashboard page with a different campaign filter, I was looking at the retargeting results. Now, imagine if I hadn't had those things set up. Imagine if I'd been looking at the entire performance to date of the original ad campaign, including both LinkedIn clicks and those dirty, dirty partner website clicks. The total number of clicks then would be just orders of magnitude higher, completely polluted by all of the bot clicks.

(09:30): And if I'd been looking at retargeting and at the original campaign squished together into an overall single performance dashboard, it still would've eventually dawned on us, but only after we'd burned untold amounts of time and money. So to catch this quickly isn't trivial. Okay, so what was going on behind the scenes here? Well, how could there still be bots? Those retargeting clicks, clearly bogus in number. They still were coming from LinkedIn itself.

(09:55): Those clicks were happening on LinkedIn, not on some partner website. So here's what I think was happening. In order to serve an ad to a partner website, the user who came to that website, even if it's a bot, the user who came to shadymcshadyclick.com that user, whether they're real or not, must have a LinkedIn account. Because in LinkedIn, when you target your ads, you can target specific roles and specific industries.

(10:22): And so the ads won't be served on shadymcshadyclick.com to a user unless that user has a LinkedIn account which matches those filters. So even if it's a bot, it has to be tied to a LinkedIn account. A bogus and fake LinkedIn account? Yep, absolutely. Not a real person, but it's still a valid LinkedIn account as far as LinkedIn is concerned.

(10:43): And since our retargeting campaign tells LinkedIn, please show this ad to people who engaged with our messaging before, well, all of those bot accounts that clicked on our ads in the partner websites before are still in fact on LinkedIn. LinkedIn has those "people" registered as real people on LinkedIn, even though they're not. And those bot accounts on LinkedIn are going to see are retargeting ads because again, they're in the audience definition.

(11:13): So we've pulled up the drawbridge by saying, don't display on partner websites anymore. So no new bot thieves are getting in, but the bots that stole from us the first time in the first week are still inside the castle with us, and LinkedIn is absolutely going to let them click on our ads. Now, why would the bots click on the ads on LinkedIn at all? Why would they even bother at this point?

(11:37): Because when they click on that ad, because it's not happening on Shadymcshadyclick, Shadymcshadyclick isn't getting paid. Well, there are two reasons I can think of. One is that Shadymcshadyclick might actually still be getting some money. They might get some sort of residual payment from LinkedIn if the bot re-engages on LinkedIn proper. I don't know. I didn't even bother to look it up because I don't care. I already know these are bots.

(12:00): The numbers are just too obvious. The other possible reason is that the bots need to click on the retargeting ads on LinkedIn in order to look real. So for instance, if 10,000 fake bots clicked on ads on shadymcshadyclick.com last month, and when those 10,000 users see followup retargeting ads on LinkedIn itself, if zero of them click the ads, LinkedIn is going to know something is up with shadymcshadyclick.com.

(12:30): Plausible deniability will be lost, and Shadymcshadyclick is going to get banned from the ad network. Sure, they'll just relaunch to something new, like I said, but it's an absolute bummer. You can totally imagine. It's just a total bummer to have to interrupt the flow of stolen money even for a day or two. So it doesn't really matter which of those explanations is true, or both of them.

(12:49): It doesn't matter because I'm already positive that the bots sneaked back in for one last bite at the apple. So we're pausing retargeting. We're going to limit the time window so that only recent clickers, aka the legitimate ones, will be targeted when we relaunch. Summing up, being on top of our own data in a clear and visual low-friction fashion saved us a lot of money and polluted thinking. And I'm not sure actually that the money is the bigger deal.

(13:17): For instance, if we had believed in a durable sense over a long period of time that retargeting was generating amazing re-engagement, that doesn't just mean we continue spending the money, which we would have, but it also would've impacted our thinking and our planning in other ways. It might have been even more damaging to our operations to have been deceived about it, to have the false beliefs in our head than even the cost and wasted dollars.

(13:41): But if we've been staring at that LinkedIn ad portal or exports to Excel, or even less thoughtfully constructed dashboards than the ones we have, we would've been defenseless to both, both the loss of money and the polluted thinking. And of course, got to mention the underlying Power BI semantic model that we use behind those dashboards. It will be very useful when connected to AI, which we'll be doing shortly, of course. So there's your AI mentioned for this episode. Check. All right, until next week. In the meantime, be careful out there.