Signal Width. Any conversion tracking setup either holds this one or it does not, whoever built it.
Signal Width
You are not just flying blind. You are actively teaching the algorithm to waste your money.
The propertySignal Width is the events either side of the money event, and the right one per platform.
Why it gets worse the more you spend
The pattern is familiar enough that people describe it to me the same way. Small budgets worked. Then the account scaled, and the quality of what came back went down as the spend went up, so the obvious answer was that the audience is exhausted or the creative is tired.
Sometimes it is. Often the account is simply being taught, week after week, from a partial and slightly wrong version of what happened, and it is doing exactly what you asked with it.
Why I built this
On a call in July 2025 I was explaining why a broken setup is worse than no setup at all, and I said the thing that has stayed with me since. The conversions that never make it back to the platform do not sit in a neutral pile marked unknown. They get lumped together with the people who actually did not convert.
That is what makes this expensive. Those buyers are in the model, filed on the wrong side of it. The platform now has real customers of yours sitting in its no column, and it will go and find you more people like them, because that is what you appeared to ask for. Every week the account runs, the mistake gets reinforced with more budget.
Picking the right events has a second half, and it applies to an account where everything is working. If your sale takes six months to close, closed business cannot be the thing you optimise on, because no ad network waits that long to connect a click to an outcome. So you go looking for the earliest event that still means something: nurtured or qualified, whatever your own funnel calls the moment a lead stops being a name. Near enough to the click for the platform to use, far enough down the funnel to be worth something.
Either side of the money event is meant literally. The events before it are what let a platform learn early enough to act on what it learned. The events after it are how it finds out whether the money stayed: a refunded sale or a subscription cancelled in week one counts as a win forever if nothing goes back to correct it, so the platform keeps hunting for more buyers exactly like that one.
Each platform draws that line in a different place, because each one has its own attribution window. The right event on one platform can be the wrong event on the next, so we pick them per platform, deliberately. Signal that is perfect and arrives too late counts as zero.
What it looks like when it is on
The guarantee is that the ad platforms are learning your business from events that mean something to your business, on both sides of the money event, chosen per platform so that each one gets the version it can still act on.
There is one measured example I can point you at, and it belongs to the whole rebuild rather than to this property: the account in the Ceiling Break case study. Leads grew from about 490 a month to about 4,100 a month in the four months following the tracking rebuild. Results are from one client’s own ad-platform data and are not a guarantee of future performance; individual results vary.
This property’s contribution to that was the choice of what those platforms were allowed to learn from. A share of the result, and only ever a share.
The mechanism underneath it is the part I am confident about, because I have watched it happen on accounts I run. Send the real qualified events back and the traffic changes character within weeks. The obvious junk stops being rewarded, and cost per acquisition follows it down. The sales team tends to say so before the dashboard does. I stop short of saying it happens every time on every account, because I would need evidence across clients to say that, and I do not have it yet.
Can the algorithm work it out on its own?
Usually put to me as“Just trust the algorithm and let it run.”
An ad platform's optimisation is a learning system, and it learns from one thing only, which is the events you send it. It works out nothing on its own about the buyers you failed to tell it about. One event is one lesson. A thousand form fills teaches it to go and find people who fill in forms, which is a different population from the people who buy. Feed it the ones your sales team qualified and it goes looking for those instead. A stretch of silence, because something upstream broke, reads to it as a stretch where nothing worked, and it rebuilds its idea of your customer around that. Trusting the algorithm is reasonable. Trusting it to overcome the material it was trained on is not the same thing.
Two things are worth checking on your own account this week: whether the event you optimise on lands early enough for the platform to still act on it, and whether you are sending the same one to every platform when their windows are nothing like each other. Send us your busiest landing page on each traffic source you run and we will tell you what your platforms cannot see of it, in writing, for nothing.
Find what your platforms can't see