When attribution does not reconcile, the tool is rarely the problem

"Our attribution numbers do not reconcile" is one of the most common complaints growth teams have. Add up the conversions each channel reports and you land at one and a half, sometimes two times what your own system counts. The first instinct is usually to change tools. After the migration, the gap is generally still there.

Because the problem is rarely the tool. It is that nobody ever defined the basis.

The gap comes from four places

Inconsistent attribution windows

Channel A uses a 7-day click window, channel B uses 30 days, your own system uses 24 hours. The same user gets counted once in each. Of course the total overshoots.

Standardizing the window is step one. There is no universally correct length — it depends on your product's decision cycle. Impulse-purchase commerce and enterprise software cannot share a number.

Inconsistent attribution models

Channel dashboards default to last-click, and specifically to last click as that channel can see it. A user who sees A's ad, then clicks B's ad and converts, gets claimed by both.

Resolving that requires a single arbiter — your own tracking. Channel-reported data is useful for reconciliation. It is not a basis for allocating budget.

Organic contamination

Users arriving through brand search or organic app-store ranking, if counted to paid channels during a campaign, make paid look extraordinary.

The test is simple: pause for a week and read the baseline. If new users only drop 20% while you are dark, a large share of what you were calling paid was organic all along.

Genuine tracking gaps

Cross-device journeys, web-to-native handoffs, reinstalls, and platform privacy restrictions all break the chain. That loss cannot be fully eliminated, but it can be quantified — and knowing you have a 15% gap is far better than assuming the data is complete.

Define the basis, then pick the tool

When we build attribution, the first step is never tool selection. It is writing down, with the client:

• At what moment a "new user" is counted, and on which identifier they are deduplicated
• How long the window runs, and how click and view attribution rank against each other
• How organic is separated out, and how the baseline is measured
• Which source governs when channel and internal data disagree, and what variance is acceptable

By the time that document exists, half the problem is usually visible — before anyone has touched a tool.

A practical caveat

Attribution is never 100% accurate, and chasing precision is a waste of time. The point of an attribution system is not to trace every unit of currency. It is to rank channels correctly against each other — so you know where to add budget and what to cut.

As long as the basis is consistent, the ranking holds even when the absolute numbers drift. That is enough to decide on.