Sometimes weird stuff happens.
One of the things Meta advertisers need to understand is that most of what happens with their ads is explainable. When you understand how things work, you see the ups and downs of performance as logical and not the work of some rogue algorithm.
Of course, sometimes weird stuff happens, too, and you have to account for it. An example of that is the focus of this post.
“Weird stuff” in this context includes the things that are either difficult to explain or completely oppose how things are expected to work. Performance shifted, but not due to reasonable explanations like creative fatigue, an increase in competition, seasonal behavior, website issues, event delays, or even randomness.
This weird stuff is reflected in data, and not on gut. And it’s rarely found by looking at the surface-level results. You need to dig deeper.
Let me share a story…
A Stable Campaign No More
I was working with a professional service company recently that enjoyed incredible stability and predictability. The goal for one campaign was leads with a budget of more than $20k per month.
You could set a clock to the results. In fact, the cost per lead was only getting better from month to month. Cost per lead hovered around $60 last summer but has slowly come down since. During the months of January and February, cost per lead had settled at a campaign best of about $50.
And then, on precisely March 1st, something happened.
Cost per lead for the first five days of March was $114, with four of the five days over $100. A bad day here or there can happen and is nothing to be all that concerned about. But this was a trend. And for context, there wasn’t a single day of January or February where cost per lead was over $100.
There had been no recent changes made to the campaign, ad set, or ads. But something had caused a drop in performance.
This had become a problem.
Considering Logical Explanations
Whenever you see a drop in performance, the initial focus should be on finding a logical explanation based on how things work. This will lead to a logical solution. You shouldn’t immediately jump to weird stuff since that will be the exception, rather than the rule.
CPM was up to $57 in those initial days of March, which was about 42% higher than normal. While CPM can fluctuate, it otherwise hadn’t been higher than $43, which was in December.
But that increased cost alone wouldn’t explain a more than doubling of cost per lead. The conversion rate (percentage of leads over landing page view) had also more than cut in half, from about 7% to just over 3%.
So the ads, which had been consistently and predictably effective, suddenly weren’t working. One of the ads Meta leaned into most had over 6,000 comments on it. I checked those comments to make sure that there was nothing there that would kill conversion rate.
And that’s when I noticed something. Comments stopped. They normally came in at a dozen or so per day, but now there were no new comments at all in March.
Why?
The Weird Stuff
The ad itself shouldn’t suddenly stop performing unless something changed. Conversion rate should remain stable, but there are reasons competition costs would increase. The ad wasn’t converting and it wasn’t even getting engagement anymore.
And this change wasn’t gradual or the result of some trend. It went from peak performance to falling off a cliff overnight.
This suggested a shift with the ads themselves. First, I performed a breakdown by platform to get a sense of where the ad was being shown in January and February to set a baseline.
Prior to the drop, Facebook made up nearly 92% of the ad spend, followed by Instagram at under 5%.
The ad set was using Advantage+ Placements, so all placements were eligible. But the older demographics on Facebook were the primary audience, so this split made sense.
This shifted completely in March. Instagram now made up 86% of the spend, followed by Facebook at 8%.
That’s not a minor variance in distribution that can be attributed to randomness. That’s a complete structural shift. And that structural shift will also mean reaching a different demographic.
In January and February, all four of the most-used placements occurred on Facebook, with 64% going to the in-app Facebook Feed. That’s where those 6,000+ comments were happening.
But during the first five days of March, both of the top two placements were on Instagram. The top Facebook-related placement was Marketplace, and that didn’t happen until later on the 5th (more on that in a moment).
The weirdest part? The in-app Facebook Feed stopped getting any impressions at all. It went from getting 64% of the ad spend to zero.
That’s undeniably weird.
So the problem wasn’t just that distribution shifted from Facebook to Instagram generally. It was that the one placement that Meta leaned into the most when things were working well was no longer used at all.
The First Attempted Fix
The hope at this point is to shift delivery back to Facebook and away from Instagram. While the old approach to fixing this might be to remove Instagram entirely, my preference was to use value rules. This way, we could adjust Meta’s bids to hopefully shift how ads were delivered.
So during late afternoon of the 5th, a value rule was applied to lower the bid on Instagram placements by 50%.
This would hopefully shift budget back to Facebook without removing Instagram entirely.
Now, data at this point is messy because changes made were happening in the middle of days. But while the value rule “worked” by shifting ad spend back to Facebook, it had unintended consequences.
Instead of shifting ad spend back to the in-app Facebook Feed, Meta sent all of the impressions to Facebook Marketplace. Meta spent a total of $404 on in-app Facebook Marketplace in January and February combined. But in under a 24-hour period, Meta had already spent more than half of that ($222).
And budget still wasn’t getting spent on the in-app Facebook Feed.
The Weird Solution
At this point, something felt structurally wrong. Meta refused to give impressions to a placement that had driven the vast majority of conversions for months. A value rule didn’t fix that, and I had doubts that removing placements would either.
On one hand, it would be easy to panic. But knowing that the performance drop wasn’t due to something explainable was strangely reassuring.
The question would be how to fix something so randomly broken. Because this felt like a bug instead of something explainable, there were a couple of weird options that are intended for such weird situations.
These “solutions” shouldn’t be solutions at all. They don’t make sense. They aren’t based on anything logical. There’s no reason they should work. But when nothing else does and it seems like something is buggy, they’re worth trying.
It seemed as though either the ad set or primary ads were corrupted. Our initial thought was to create a new ad set and start over. But first, we’d try something far more basic.
We duplicated the ad that Meta had previously given so many impressions. But instead of using the existing post, some minor changes were made to it. And then it was published.
When I say that the impact was immediate, I’m not exaggerating. The ad was quickly approved, and the initial wave of impressions made it obvious: Impressions were now flowing back to the in-app Facebook Feed. Marketplace was no longer prioritized.
After a couple of days of this, the value rule was removed. And here were the results from March 9th through the 13th:
- 87% spent on Facebook, 10% on Instagram
- 66% spent on in-app Facebook Feed
- $46 cost per lead
The biggest relief may be that cost per lead not only returned to normal, but has actually been lower than January and February. And that’s even though the original ad with 6,000 comments is no longer getting shown, even though it hasn’t been turned off.
Approach Weird Solutions as Exceptions
I was hesitant to share this story because I know what tends to happen. Anyone who hasn’t seen good results recently will assume they are part of this glitch. “It happened to me, too!”
All indications are that this was an isolated situation. We haven’t seen this across accounts.
That doesn’t mean that other advertisers haven’t seen unexpected shifts in performance. But always start under the assumption that there’s a logical explanation. Resist labeling it “weird stuff” until you’ve not only exhausted those explanations, but you’ve pinpointed what makes it especially weird.
In this case, we had months of data that highlighted expected performance and distribution. That shifted completely on the 1st of March, which impacted results. And despite the use of value rules, nothing could force Meta to show ads on what were previously the primary placements.
That was weird. It felt like a bug. And because of that, a weird solution was attempted as a last resort. Luckily, it worked.
Your Turn
Have you run into similar weird problems that required weird solutions?
Let me know in the comments below!
The post When Weird Stuff Happens to Meta Ads appeared first on Jon Loomer Digital.
