Attribution Reconciliation

If you add up the conversions every ad platform claims it produced, the total almost always exceeds the conversions that actually happened. Meta says it drove the sale; Google says it drove the same sale; the affiliate network claims it too. Each platform, optimizing to its own numbers, counts generously and within its own walled view. Attribution reconciliation is the unglamorous but essential work of resolving those overlapping, inflated, and sometimes contradictory claims into one coherent account of what really drove results. It compares the self-reported figures from each platform against an independent source — the company's own analytics, an order database, or a measurement partner — and works out how to credit conversions without double-counting them.

The discrepancies have real causes, not just error. Platforms use different attribution windows, different rules for what counts as a conversion, and different definitions of a view-through versus a click. One counts a sale if its ad was seen within a week; another claims the same sale on a click the day before. Privacy changes, modeled conversions, and cross-device behavior widen the gaps further. Reconciliation does not assume any single platform is lying; it accepts that each measures within its own boundary and asks how those partial views fit together against the ground truth of actual orders. The output is a reconciled view a marketer can budget from, instead of a pile of conflicting dashboards that cannot all be right.

Attribution reconciliation is often confused with choosing an attribution model, but they solve different problems. An attribution model — first-touch, last-touch, multi-touch — is a rule for how to split credit among the touchpoints on one known customer journey. Reconciliation comes first and sits above that: it deals with the fact that different systems report different totals for the same period, so the journeys themselves do not agree across sources. You can apply a beautiful multi-touch model inside one platform and still be wrong at the company level, because that platform never saw the touchpoints that happened elsewhere and counted some it should not have. The model organizes credit; reconciliation makes sure there is a real total to organize.

In practice the two work together. Reconciliation establishes a trustworthy total — how many conversions truly occurred, from an independent source of record — and then an attribution model or an incrementality test distributes credit for that total across channels. Skip reconciliation and you apply your model to inflated, overlapping inputs, producing tidy percentages built on double-counted sales. This is closely tied to install attribution on mobile, where an independent measurement partner reconciles the competing claims of many ad networks against the app's own event data. Reconciliation is the check that keeps attribution honest; the model is how you allocate once the totals are trustworthy. Do them in that order, or the elegant model just launders bad numbers.

Doing reconciliation well starts with a single source of truth for the outcome that matters — usually the order or revenue system, the one place a sale is recorded once. Everything else is compared against it. Then you align definitions and windows across platforms so you are comparing like with like, quantify the overlap and double-counting, and decide deliberate rules for crediting shared conversions rather than letting each platform keep its full claim. Cadence matters: reconcile every reporting period, not once a year, because windows, privacy rules, and platform behavior keep shifting. Where the stakes are high, pair reconciliation with incrementality testing, which answers what the platforms cannot — how many of those conversions would have happened anyway.

The traps are taking any platform's self-reported number at face value, summing them, and acting surprised the total is impossible; and mistaking a fancy attribution model for reconciliation. Another is reconciling once and assuming the gaps stay fixed, when measurement rules change constantly. The subtlest trap is treating reconciled attribution as proof of causation — even a perfectly reconciled count still credits correlation unless incrementality testing shows the ads actually caused the conversions. There is also the quiet failure of reconciling once in a spreadsheet no one revisits, so the rules drift out of date while the dashboards keep disagreeing and someone keeps trusting them. Done well, attribution reconciliation gives an organization one set of numbers it can budget from; done badly, it lets whichever platform counts most aggressively quietly set the budget it never earned.

Reconciliation borrows from accounting, where it means matching two records to resolve differences; attribution reconciliation applies that discipline to conflicting marketing conversion data.

Etymology: source.

Search interest for this term over the last five years:

View interest-over-time on Google Trends →

Areas of marketing where attribution reconciliation is a core concern: