Add up the conversions your ad platforms report and the total will exceed the orders in your finance system. Often by a lot. Every client we have worked with has this problem, most have quietly stopped mentioning it, and a few have built a monthly ritual of arguing about whose number is right.
Nobody is lying. The numbers disagree because they are answers to different questions, and because a meaningful slice of the data never arrives at all.
Every platform grades its own homework
Each ad platform counts a conversion it believes it caused, under rules it sets itself. Those rules overlap.
A customer clicks a paid search ad on Monday, sees a retargeting ad on Tuesday, clicks an email link on Thursday, and buys. Paid search claims it. The social platform claims it on a view-through basis — no click required, just an impression inside its attribution window. Your email tool claims it, because it saw the last click. Three platforms, three conversions, one order.
Two settings drive most of the overlap:
- Lookback windows. A 30-day click window and a 1-day view window are different products. Platforms ship different defaults, and almost nobody changes them.
- View-through attribution. Counting an impression as a cause is a defensible choice for brand campaigns and an indefensible one for calculating return on ad spend. It is frequently on by default.
Before touching anything technical, write down each platform's window and whether view-through is enabled. That table alone usually explains half the discrepancy, and it takes an afternoon.
A slice of the data never arrives
The rest is loss, and it is structural rather than a bug you can fix.
Browser restrictions. Safari's Intelligent Tracking Prevention caps the lifetime of cookies set by JavaScript — days, not years. Firefox blocks known trackers outright. A returning customer on Safari can look like a brand-new visitor on their second session, which quietly reassigns credit to whatever channel brought them back.
Ad and script blockers. A double-digit percentage of technical audiences block analytics entirely. Those people still buy; they just never appear in your funnel reports.
Consent. Where consent is required, visitors who decline cannot be tracked with cookies at all. Depending on your banner and your market, that is a substantial share of traffic — and your consent banner's design has more influence on your reported conversion rate than most of your campaign settings.
Cross-device. Someone researches on a phone and buys on a laptop. Without a login tying the two together, that is two unrelated people as far as your analytics is concerned.
What genuinely helps
Some of this is recoverable. Roughly in order of effort against payoff:
Fix the measurement plan before the tooling. Most tracking setups we audit have events firing that nobody uses and the two events that matter missing or misfiring. Decide what decisions the data must support, then instrument exactly that. This costs nothing and is skipped almost every time.
Move collection server-side. A server-side tag manager receives events from your own domain and forwards them onward. It sidesteps client-side blockers, and first-party cookies set server-side are not subject to the same lifetime caps. It is real engineering work — a container to run, a domain to configure, a deployment to maintain — and it is where marketing stops being a purely marketing problem.
Send conversions from your system of record. Enhanced and offline conversion imports let you send confirmed orders, hashed and consented, from the CRM or order system back to the ad platforms. This is the single highest-value fix for most businesses, because it replaces what the browser guessed with what actually happened, and it works for phone orders and long sales cycles that online tracking never sees.
Implement consent mode properly. Done right, declined-consent visitors still contribute modelled, aggregate signal instead of vanishing. Done wrong — and it is usually done wrong — it either breaks measurement or breaks compliance.
Where attribution runs out of road
Here is the part that gets left out of most articles on this subject: even a perfect implementation does not tell you what you actually want to know.
Attribution answers which touchpoints preceded a sale. The question you care about is what would have happened if I had not run this campaign. Those are not the same question, and no amount of tracking rigour turns one into the other. A retargeting campaign that "converts" brilliantly is often just invoicing you for people who were going to buy anyway.
The tools for the real question are experimental, not analytical:
- Geo holdouts. Turn a channel off in matched regions, leave it running elsewhere, compare. Crude, unarguable, and the most useful test most businesses never run.
- Conversion lift tests. Platform-native holdout groups. Marking their own homework again, but far better than modelled attribution.
- Budget step changes. Move spend up or down deliberately and watch total revenue rather than platform-reported conversions.
None require perfect tracking. They require the discipline to deliberately stop spending money for a few weeks, which is a harder sell internally than any technical fix on this page.
What to actually do
If you take one thing: stop trying to reconcile your platforms to each other. They will never agree, because they are not measuring the same thing.
Pick one number as the truth — usually orders in your finance or order system — and treat every platform figure as a directional signal against it. Track the ratio between platform-claimed conversions and actual orders over time. When that ratio moves, something real changed. That single metric is worth more than a quarter spent chasing a perfectly reconciled dashboard, and you can start tracking it this week.