Search
SUBSCRIBE
WEEKLY STRATEGY
PODCAST
PROFIT AUDIT
THE SOFTWARE
calendar-dots
LET’S TALK
Apr 30, 2026
Your Ad Platforms Aren’t Broken, They’re Learning From the Wrong Signals.
Your Ad Platforms Aren’t Broken, They’re Learning From the Wrong Signals.
00:00
06:42
Transcript
0:00
A question I get fairly often from operators running decent-sized paid accounts is some version of, "Why does Meta keep finding us the wrong customers?"
0:09
The honest answer, in most cases, is that the platform is doing exactly what it was told to do. The problem is not the algorithm. It's the signal you've been feeding it.
0:19
[upbeat music] Hi, I'm Alex Orley, and this is Hard Margins, your weekly e-commerce brief, brought to you by RetentionX.
0:38
RetentionX is the only integrated growth platform built for Shopify businesses. Think one clean source of truth for your data and the tools to act on it, no data team required.
0:48
This week, we're talking about conversion API signal quality. I know, technical.
0:54
Specifically, the difference between using your conversion API to recover lost tracking data and using it to actually define what the platform should be learning from.
1:05
Those are two different things, and most teams are only doing the first one. Don't worry. I'm going to unpack this in a way that makes it super easy to follow, regardless of whether you spend your time in ad accounts.
1:18
When most brands connect their store to Meta, they watch purchase events start flowing and consider the signal problem solved. What they've actually done is handed the platform an undifferentiated feed of purchase data.
1:30
Think good orders and bad ones, high-margin customers and low-margin customers, buyers who come back and buyers who didn't or won't, with nothing in the signal to indicate that any of it should be weighted differently.
1:45
The conversion API was designed to address signal loss, and for most accounts, it does that reasonably well. The ROAS stabilizes, purchase volume looks acceptable, and your account feels under control.
1:58
But signal stability is not the same as signal quality. If you're not using your conversion API to differentiate between the orders you want more of and the ones you don't, the underlying problem stays intact.
2:11
The platform continues optimizing towards the full distribution of your customer file, including the cohorts that will prove destructive to your margin over time.
2:21
Most teams are running a single purchase conversion event and asking it to do two distinct things simultaneously without recognizing that those things have different requirements. The first job is optimization.
2:34
This is the event the algorithm uses to decide where to surface your ad and how much to bid. It needs to be stable.
2:41
If you change it, the account essentially restarts its learning process from scratch, which is why account restructuring is usually more disruptive than it appears.
2:50
The second job is evaluation, the signal that tells you whether the customers coming through are actually worth acquiring. That's not a question of volume. It's a question of whether the right people are converting.
3:02
A single purchase event used for both gives you an account that scales efficiently towards whatever it has been trained on, which, without any filter, is the full distribution of your orders, as we've already said, including the ones you wouldn't choose to repeat.
3:17
So here's what to do instead. The approach is sequential, and the sequencing matters. Keep your existing purchase event as the optimization signal.
3:26
It carries delivery history and accumulated learnings that took real time and spend to build. The account doesn't need a different optimization event. It needs a second one running alongside the first.
3:39
So next, build a filtered conversion API event for measurement purposes only. Inside RetentionX, you can create a second event scoped to the purchase quality you want more of.
3:51
New customers only, above a defined contribution margin threshold, excluding heavily discounted orders. At this stage, it's for measurement only, not optimization.
4:01
Once both events are running, use the comparison to diagnose cohort quality by campaign. Which campaigns produce high purchase volume but very few filtered event conversions?
4:13
Which produce a disproportionate share of filtered event conversions regardless of ROAS?
4:19
The gap between those two is where the account's actual problems are visible, and it's usually clarifying in ways that standard campaign reporting isn't.
4:28
From there, reallocate spend toward the campaigns where cohort quality is the strongest without touching the account architecture or the optimization event.
4:36
This is a budget decision, not a structural one, and it carries much less risk. Only after the filtered event has sufficient volume and stability should you evaluate whether it should become your optimization signal.
4:49
Measure first, optimize second. Reversing that order is how accounts get destabilized for months while chasing a cleaner signal that wasn't ready to govern bidding.
4:59
When you send a second filtered event to Meta, the breakdown surfaces directly in your campaign reporting. Qualified purchase volume alongside standard purchase volume at the campaign and ad set level.
5:11
That comparison is where the analysis becomes genuinely actionable. Which campaigns are winning on volume, which are winning on customer quality, and where those two things diverge.
5:22
Most teams treat the conversion API as infrastructure, a mechanism for recovering signal that was lost when browser tracking degraded. That framing is way too narrow. The more consequential use is definitional.
5:34
It's how you tell the platform what you actually want it to learn from when you separate the event you optimize on from the event you evaluate against. The account becomes easier to read.
5:45
Campaigns that look strong on ROAS but weak on customer quality become visible. Campaigns producing durable customers, regardless of cost, get the attention they warrant.
5:55
That distinction is where most paid accounts have the most room to improve. The platforms aren't broken.
6:01
They're learning from whatever you send them, and most brands are sending an undifferentiated signal that reflects the full range of their customer economics, including the parts they'd prefer to stop repeating.
6:12
Getting that right isn't an account restructure. It's a measurement decision, and the sequence I walked through today is how you do it without disrupting what's already working.
6:20
[upbeat music] Thanks so much for listening. I'm Alex Orley. This has been Hard Margins, brought to you by RetentionX. I will see you next week.
6:34
[upbeat music]
Hard Margins
Listen on
Apple Podcasts
Apple Podcasts
Spotify
Spotify
YouTube
YouTube
Recent episodes
How PlantPaper Built One of the Best Retention Curves in DTC
May 13, 2026
Hard Margins #15 How to Build Decision Frames
May 6, 2026
You're Not Managing Returns, You're Managing an Average.
Apr 30, 2026
Deep Dive: Your Catalog is Your Retention Strategy
Apr 30, 2026
AOV is Not a Checkout Metric, It's a Quality Filter
Apr 30, 2026
You're Not Retargeting Buyers. You're Retargeting Traffic.
Apr 29, 2026
Your Entry Product Sets Your Ceiling
Apr 29, 2026
Revenue is an Opinion. Contribution Margin is a Fact.
Apr 29, 2026
The Power of Behavioral Sequencing
Apr 29, 2026
Profit Based LTV
Apr 29, 2026
