Deduping Conversions When Every Platform Claims the Same Order
Add up platform dashboards and revenue exceeds the store's actual revenue. Where double counting comes from, order ID discipline, and the one source of truth rule.
Do the party trick in any marketing meeting. Add the conversions claimed by the ad platforms, the email platform, and the affiliate network for last month. Compare the sum with orders the store actually took. The claims will exceed the reality, sometimes by a third or more, and every platform's number is internally honest. They are all counting the same orders, each with a rule that favours itself.
Deduplication is unglamorous and it is the foundation everything else stands on. Incrementality tests, channel budgets, and freelancer bonuses all inherit whatever double counting lives in the base reports, and no model sophistication downstream fixes a numerator that counts orders twice.
One order, one ID, one owner
The working rule is boring: your store's order system is the source of truth, every platform event carries the order ID, and any order can have exactly one credited channel in the deduped view, assigned by a rule you wrote down. Browser tags, server events, and platform imports all firing the same purchase is the common mechanical cause of inflation inside a single platform, before cross platform claiming even starts. Audit the firing first: one purchase, one event per platform, checked in the tag debugger against real orders, including a refund and a partial refund to see what those do.
The assignment rule matters less than its consistency. Last click, first click, or a split for defined cases all work as bookkeeping, provided the rule is stable across months so trends mean something. What breaks teams is the rule quietly changing, a new server side integration here, a changed window there, until a performance jump turns out to be an accounting change wearing a growth costume.
Windows are where the bodies are buried
Every platform defaults to the longest attribution window it can defend, because longer windows claim more orders. A thirty day click window and a one day window produce different truths about the same campaign, and a view through claim on a first order and a repeat order in the same report multiplies the overlap. Line the windows up across platforms to one agreed standard for the deduped view, keep each platform's native view for in-platform optimisation only, and never let the two be quoted in the same meeting without labels.
Refunds belong in the same discipline. Platforms rarely subtract them on the same schedule the store does, so claimed revenue drifts further from reality at exactly the moments, big promotional pushes, high return categories, when decisions are being made from the numbers.
What deduped reporting is actually for
Not for declaring a winner. The deduped view exists so the totals tie to money, so a month can be compared with the month before, and so the arguments happen about incrementality, the genuinely hard question, on top of arithmetic everyone trusts. Teams that skip it end up litigating whose dashboard is lying, which is a question with no productive answer, because none of them are. They are answering different questions and calling all of them conversions.
Related reading
If this topic is useful, these earlier guides go deeper on connected questions: Reading a Lift Test When the Conversion Counts Are Small, Coupon Extensions and Last-Click Credit: When the Final Click Did Nothing, and View Through Conversions: A Sceptical Read Before You Pay for Them.
Frequently asked questions
Why do platform conversion claims add up to more than actual orders?
Each platform counts orders under its own window and rule, so the same order is claimed by several. Tag duplication, browser plus server events for one purchase, inflates counts further.
What should be the source of truth for conversions?
The store's own order system, with every platform event carrying the order ID. The deduped view assigns each order to exactly one credited channel under a written, stable rule.
How much do attribution windows matter?
Enormously. Longer windows claim more orders, and mismatched windows across platforms guarantee double counting. Agree one standard for reporting and keep native platform views for optimisation only.
Does deduping decide which channel works best?
No. It makes the totals tie to real revenue so incrementality testing, the real effectiveness question, can happen on arithmetic everyone trusts.