Why Cost Caps Broke In This Ad Account (With Screenshare Of Real Meta Ads Dashboard) transcript

All right, so I had a question from somebody about uh an overspend that happened and they said their cost caps were set, everything was going great, they were really happy to be running with manual bids, everything was working awesome. But then there's this problem, which is that uh it suddenly spiked in spend really, really badly. So, I said, "Okay, I will I Let me try to analyze what happened to see if I can best understand what's going on in your ad account and see if we can figure out was this normal or did something break or like should you stop trusting cost caps or what, okay?" So, um or bid caps in this case.

So, here So, here's the uh here's the setup for the ad account, okay? And um what you'll see is that uh I looked around and you know, let's say June 1 through 14, this is the way I would normally look at this, just kind of just poke around the data and say, "Okay, you've got 15 grand in spend here, a pretty solid setup. I would probably personally combine up these bid cap and retargeting uh as as I think you're probably separating out campaigns in a way you don't really need to, just keep everything in one campaign, don't worry about retargeting ad sets.

Uh it's questionable whether they even really work anymore because uh the way that, you know, people like cookie tracking is or, you know, tracking in general is a lot harder, etc. So, um So, anyway, but that that's a small deal. In this case, uh while I would combine them, it's not a big deal, it's not really relevant to this. So, you can see though, June 1 to June 14, you're at you know, 15 grand, 16 grand total, something like that.

You're right around um let's call it uh 1 grand per day, okay? Just a little more than that. And so, everything is going fine, about a thousand a day, we'll get to June 14th and this is the way I would always analyze this, you know, about 1,300 spent right here, 1,350, something like that. And then you get to the Saturday, June 15th, and bam, 4,000 spent and your ROAS goes way, way down. Um I've got it separated out here, here's the reporting window of 0.7, which is 70C, 1D, uh 1-day view as well, 7-day click, 1-day view combined.

I always have a 28-day click column in here as well, just to kind of see how this went. But you could see clearly a massive spike in spend and a massive drop in performance. So, how would I analyze this? And by the way, I also look, you just go to the next day and you can you can see kind of what's going on here. It's uh uh reduction in spend, performance gets a little maybe a little better, but still really bad. But then by the 17th, uh things have come back around, performance gets way better now, spend goes way back down, and and more within the normal range, especially for a Monday after the weekend.

And in fact, if you go from there to here, the rest of the week, everything looks looks really good again. This account's back to back to working. So, what happened on that weekend is the question. Okay, that's that's what I would answer. So, um the first thing I'm going to do going to isolate those 4 days so we can see sort of the before and after of that time. Um and I'm going to jump into this campaign that seems to be the big problem.

One of the um I'll actually also just filter out I always sort by amount spent, and I'm going to filter out um all of the stuff that isn't spending much cuz I don't really care about it so much. So, we're going to go filter by selection. Um and when you do this, this starts to get a little interesting. So, over this course uh over this time, one of the first things I'm going to do is I'm going to look at the um bid history in the the campaign to see if I can figure out what's going on at the ad that level, and you can see um Dan here uh has um has changed the bids a bunch of different ways.

So, um he started on June 15th, uh went from 150 to 140, um and, you know, had dropped some bids down, and then kind of uh ticked back up by June 17th. This thing had had ticked back off in spend. So, it went down and then back up, and you can see obviously during the big overspend moments especially, he he pushed those down, and then and then, you know, felt comfortable pushing them back up uh a little bit later. So, um so, that is is how this worked.

And then um and and then um from there, what we'll do is we'll go look and see at some of the other metrics cuz the big question that uh presents itself to me is like, is it is something broken here? Or did Meta actually do what it was supposed to do? And maybe what I'm always looking for usually here is people sometimes think something went really really wrong, but maybe it's just small sample size noise. Like, you had an increase in spend, and you actually should leave it and let it spend because um because if you would just let it spend some more, you know, get out of it would it would actually work out.

It's like you flipped heads five times in a row and you're like, wait a minute, is something wrong with this coin? Well, no. I mean, most cases you just keep flipping the coin and it's fine. It's just a weird thing where happens low probability chance, but it happens that heads flips you know, four or five times in a row or whatever and you should keep flipping if you want a 50/50 outcome and every time you should bet 50% heads, 50% tails, right?

So, so random little things happen in small variations and that I think is a really common problem when people are buying on manual bids. They see some like quote unquote overspend, but but actually it's just small sample sizes and and the the reality is that weird stuff happens there. So, is that what's happening here? Well, I don't think so and I'll tell you why. So, on Friday everything is working great. $1,000 in spend on these ads, 2.2k 28 DC revenue.

So, if we're at 2.1, we'll just stick with 28 DC reporting to to just make it easy. And what I This is one of the only times where I really will look very much at these these sort of lower funnel metrics or upper funnel metrics, excuse me, to to try to isolate what's going on. You're paying $2.35 for a click at a 1.5% click through rate, okay? And your CPM's $34 in this case. So, then you go to Saturday and what you'll see is the spend goes way up, the ROAS goes way down.

The sample size is small here, by the way, so you know, I I did wonder about that. But what you'll see is the CPC goes all the way up to $4.10. That's way up and the CTR gets cut in half, CPM actually goes down. That looks to me like something went haywire with the delivery of the ads, maybe on a different format to a different age or gender or something like that and and that's probably what I would go look at next.

Here, let's just we'll just do it really fast. Let's do this in real time and see like what about age and gender? Is there anything going on here? So, you've got let's see, the top spenders are 25 to 45 male. What about on Friday? Any difference there? Oh, heavily female. That's interesting. So, for some reason, a bunch more delivery happened to that male audience um before. Now, I don't know why that would be, but maybe that's part of it is it suddenly some some signal came into practice where um or it came to be where where uh for some reason the machine learning thought, "Okay, this is actually um going to be delivered needs to be delivered to a different audience." And so, maybe it did that.

Let's also check placement. That's another really common uh thing that you could see happen here. So, um June 14th, so this is the day before the spike in spend. You're heavily Facebook mobile, Instagram mobile, Facebook desktop. Really especially um Facebook mobile, Instagram mobile um for for at least this campaign. Let's see, the Dan Explainer one is the actual one that went crazy. So, this one, Facebook mobile, Instagram mobile, both feeds um feed ads, a little bit of stories, a little bit of little bit of reels as well.

So, let's keep our eye on that. When we go to top spenders here, um it's Facebook mobile feed, Instagram mobile feed, reels. Yeah, it's it heavily mobile feed. So, no issues with placements that I can see. Everything looks pretty normal for placements. Let's Let's check Let's just make sure we're going to actually hone in on this one um Dan Backyard Explainer ad set again. So, let's actually even Yeah, dial in a little bit more closely here.

We'll filter just to this one so we can make sure we don't get anything confused. We've got 25 to 54-year-old males um is what the audience is on June 15th. It does not perform well in that audience. Let's see, that this same audience Oh, okay. So, in this case, it's the same audience actually. So, I was looking at a different ad ad set. Uh maybe a little It changed a little. It does 25-year-old males suddenly jumped up a whole bunch.

What about here now? Same ad set. Yeah. Yeah, maybe a little in there. I'm not sure. I'm not sure that's the culprit either, actually. That's a little bit weird. For whatever reason, hmm I'm not sure what's going on there. Um but for whatever reason, something seems to have changed in the delivery over that time period. Let's look one more time. 35 to 54 male one more time here. 25 to 34, 35 to 44 male. Yeah, and then that this audience actually performed, you know, a little poor for one day, but that not so much.

Okay, so I don't know. I don't know if there's a delivery issue or what on that ad and on this um campaign in this ad set in particular. Um but it looks to me like something did happen with the delivery in a way that is weird. Something these ads were delivered to meaningfully different people because they're behaving really differently with the ads, right? Again, the thing you can look at here is the click-through rate goes way down.

CPMs within the range of normal uh relative to what it was before. CPC goes way up. Again, we'll just look at this again and CPC, you know, 235 versus a 1.46. And again, we'll we'll we'll kind of dial this in a little bit more against all these ads. Now we're back to normal here basically. Really similar CPCs, all that, and everything's working great. Um so what happened? Here's my guess. Uh I think you got a weird thing where the machine learning sort of for some reason uh uh screwed up a little bit.

Over-delivered to the wrong audience that for whatever reason its prediction engine like {quote} unquote believed was working. Um here's what I'll say about that. Um it's I I'm very surprised. I very rarely see this kind of thing. I think that's probably what happened and I think it's really notable here that it self-corrected the way that it did. Um I think that is not too surprising to me that it self-corrected the way that it did once it got better signal back that actually it's delivering to the wrong ad.

You could try to go chase down a refund. I doubt you'll get it cuz it doesn't look like you got like a massive CPM spike. I don't think there's a wide outage here. I think in this case the machine learning just got the wrong signal and and delivered to the wrong person and certainly that can happen. Um so does that mean the whole bid cap strategy is wrong? No, I'll tell you two reasons why. Number one, because um this is incredibly rare.

It's incredibly rare. And for me, I am willing to pay the cost of something like this this happening so rarely versus the cost of trying to manage this all myself over time where um I just know that I'm going to make mistakes like this incessantly. Like I'm just going to do all kinds of things that are going to deliver the wrong ads to the wrong people the more that I get involved. So, the machine learning in the aggregate going to outperform me and and so that's point number one that this is this is really really rare.

Point number two I would make here is that uh is that you I think of probably I think of manual bidding similar to the way good poker players think. And that is you want to you want to make the right bet the most total times. Machine learning is not perfect and I don't think it's perfect and anybody who tells you otherwise is wrong. And that's because machine learning is fundamentally trying to predict the future and it is impossible to perfectly predict the future.

You cannot do it. It can't be done. And therefore, what you want to do instead is try to think in terms of probabilistic forecasting. And and so there's a lot of ways to think about this, but basically what I would say is um or the basic idea here is if you're a poker player you think like even if you get beat because a player you're playing against gets an ace on the river, right? The last card flips an ace and it was a very low probability chance that was going to happen.

If you put all your chips in or the right like that you make the right bet quote unquote that the bet that works out the most times, okay? But you get beat. A good poker player makes that bet again every single time even if it stings to get beat that one time because what they know is that they're not actually trying to they're never going to win every hand. It's not going to go perfectly every time. There's no way to predict the future like that for it to be perfect every time.

What you're trying to do is predict the future the most possible times probabilistically and win in the aggregate. And there's a very similar thing going on here where in Meta except that you don't have anybody you're playing against. Meta's machine learning is going to predict the future better than you most times. So, you may get a weird outlier scenario where an ad that could have spent gets suppressed, an ad that doesn't spend spends when it shouldn't or rather an ad that's in this case like you know some ads spend when they shouldn't.

Whatever. And again, this actually may legitimately be like some kind of a weird bug that happened that day. You may reach out to to Meta and see if that could happen. But that's good. So, one last thing, the way I mitigate against this uh um kind of thing happening is that I make a point of um keeping my budget above my campaign budgets above my actual spend, but not too high above it. And that will end up solving this.

So, um I used to keep budgets much higher, but then at the beginning of this year there was some uh it's 2024 at the beginning of this year there was some uh outages that affected a lot of accounts and I had my budgets way too high. Um and so now what I'll do, like in this case, right? If you um let's say you uh yesterday spent 760 on this ad account. You got $2,500 budget. That's about right probably. Like I wouldn't push it much above 2,000, 2,500 for the day because now if you get hit by some weird thing like that, it won't hit you as hard, especially if it's actually legitimately an outage.

Um and and you know if you leave it 10 grand or whatever, it's not the case. Now, if you keep it there and then you have ads start spending and they start working out really well, you may lose a little bit of upside day up, but of course if you spend good money at your budget and it works and um and you're hitting a high ROAS, then just go raise your budget the next day. So, you lose a little bit of upside in case an ad takes off and goes crazy, but not not too much.

So, that's what I think is happening here. A weird situation, very very weird. It's it's honestly not what I expected when I when I said I would look at this because it happens uh in my experience like I in fact I really can't think of a time when I've had this experience exactly happen to me outside of a scenario where I get extremely high engagement ads where the ad gets a ton of engagement, Meta predicts a high conversion off the ad gets a ton of so much engagement, but then the the people don't convert normally cuz the ad was sort of click-baity.

That's the only other scenario where I see um this kind of thing happen, but uh yeah, that's what I think is going on here. Uh thanks Dan for sharing and being willing to share this with other people. And uh best of luck to you going forward. I think your account build looks really good. Now go build some more creative and some landers and all the things you got to do and generate some more money.