Two numbers claim to be your cost per lead. The one in the ad platform and the one you get from counting enquiries yourself. They almost never match, and the distance between attribution and reported CPL is where a great many bad decisions get made.
The gap is normal. A gap that suddenly changes is the thing worth investigating, and almost nobody is watching for it.
This is for you if your platform reporting and your inbox tell different stories.
Why attribution and reported CPL drift apart
Each system can only count what it can see, and they see different things.
An ad platform counts conversions it can tie back to a click it served, within its own attribution window, using whatever tracking survived the journey. Your CRM counts people who contacted you, regardless of where they came from.
Four forces pull those apart. Blocked or degraded browser tracking. Consent declined, so the conversion is modelled rather than observed. Cross-device journeys, where somebody clicks on a phone and enquires on a laptop. And attribution windows, where a conversion is credited to the day of the click rather than the day it happened.
None of that is anyone behaving badly. It is the limit of what each system can observe.
Which way attribution and reported CPL disagree
It runs both ways at once, which is what makes it confusing.
Platforms under-report where tracking was blocked. They over-report where view-through conversions and modelled estimates are included in the same total as observed ones.
Those two errors do not cancel out into a reliable middle. Each figure carries an unknown amount of both, and you cannot tell which dominates without checking against your own records.
The other reliable oddity: add up conversions reported by every platform and the total often exceeds the number of enquiries you actually received. Each one claims what it can see, and they overlap.
What the tracking changes cost your reported CPL
This is measurable rather than anecdotal, which is rare in this area.
Research by Aridor, Che, Hollenbeck, Kaiser and McCarthy, published in April 2025, measured a 36.6% relative reduction in click-through rates on conversion-optimised Meta campaigns after Apple’s App Tracking Transparency, with a 95% confidence interval of 18.2% to 54.5%, across 1,221 firms. Meta’s share of online advertising spending declined by 4.4% over the period, with most of that shifting to Google.
The point for your reporting is that campaigns relying on off-platform data were measurably degraded, and a share of what your dashboard shows is now estimated rather than observed.
That is also why running the Pixel and the Conversions API together matters. A server-side signal survives things a browser-side one does not.
Working with attribution and reported CPL honestly
Pick one source of truth for business decisions. It should be your own records, because they count people who paid you.
Use platform figures for steering, because they update fast and they know which ad was clicked, which your CRM does not.
Then measure the gap between them, monthly, as a ratio. If your platform reports 100 conversions and you counted 80 enquiries, that ratio is 1.25. Write it down.
A stable ratio is entirely workable. You can plan around a known bias. What matters is when it moves, because a gap that jumps from 1.25 to 1.9 in a month means something broke, and you will spot that long before anyone notices the underlying problem another way.
Which attribution model to use for reported CPL
Less important than people think, provided you are consistent.
Last click gives all credit to the final interaction, which undervalues everything that created the demand in the first place. Data-driven models spread credit more sensibly and are harder to explain to anybody.
The genuine mistake is switching models partway through a comparison, then concluding a channel improved or collapsed. Nothing about the channel moved. The accounting did.
Pick one, know which way it is biased, and leave it alone for at least a quarter.
The one attribution and reported CPL check worth building
Most measurement work is about being right. This one is about noticing when something changed.
Build a single line: platform-reported conversions divided by enquiries you actually received, each month. It takes minutes once your enquiry count exists.
That ratio will not be 1.0 and it does not need to be. What it needs to be is stable. A steady 1.3 for six months tells you your platforms are consistently generous by about a third, and you can plan around that.
When it jumps, something specific happened. A tag broke, a consent banner changed, a form was rebuilt, or a campaign started counting a different event. Every one of those is cheap to fix in week one and expensive to discover in month four.
Nobody builds this report because it is nobody’s job. It is the highest-value line in a monthly review and it fits in one cell.
What this changes about budget decisions
Three practical consequences.
Do not compare a platform-reported figure from this quarter against a CRM figure from last quarter. That comparison is meaningless and it is made constantly.
Do not move budget between channels based on platform-reported cost per lead alone, since each platform is generous about its own contribution. Comparing channels properly needs your own qualification data.
And do not treat modelled conversions as though they were counted ones. They are reasonable estimates. Reporting them without saying so makes your numbers look more precise than they are, and precision you have not earned is how people become confident about the wrong thing.
What to change this week
Four steps.
Count your genuine enquiries for one month by hand and compare against what each platform reported. Write down the ratio.
Check whether your reporting distinguishes observed conversions from modelled ones. If it does not, find out the split.
Pick your source of truth for business decisions and say out loud which one it is, so nobody quietly uses the other in a meeting.
Then put the ratio in your monthly report as a line of its own, so a sudden change is visible rather than discovered months later.
The wider piece on judging lead cost covers the figure this is all trying to measure, how the calculation is built covers which inputs are definitions rather than facts, sending closed-won data back is the fix that puts real signal back into the platform, and why Meta and Google moved in opposite directions matters when you compare their reported numbers. To have your measurement checked properly, get in touch.