Conversion completeness

It shows up as a number that will not stay still. A report is sent on the third of the month, someone opens the same view a fortnight later, and the figures have all moved up. Nothing was changed and nobody made a mistake. The month was simply still arriving.

The version that causes arguments is the month-on-month comparison made too early. The current month is compared against a fully settled prior month, comes up short, and gets read as a decline. Every cost figure derived from that comparison is overstated at the same time, because the spend is complete and the conversions are not.

A conversion is recorded against the date of the click, not the date it happened. So an outcome that lands three weeks after the click is added back into a month that already looked finished. The longer the lookback window and the longer the real sales cycle, the more of any recent month is still in transit.

That is why the same reading date means different things on different accounts. An account with a short window and same-session purchases is close to complete within days. An account with a ninety-day window and a considered purchase can still be missing a third of a month a week after it closed. The difference is not measurement quality, it is the shape of the business.

Offline imports add their own delay on top, because they run on a schedule and backfill in batches. An account can look flat for days and then jump when a single import runs.

The direction of the error is always the same, which is what makes it dangerous. A recent window understates conversions and therefore overstates every cost derived from them: cost per acquisition looks worse, return on ad spend looks worse, and a campaign in its first weeks looks like it is failing at exactly the moment somebody decides whether to keep it.

Acting on it is expensive twice over. Budgets get cut on a number that would have corrected itself, and the cut then removes the very spend whose delayed conversions were about to arrive, which makes the following month genuinely worse and appears to confirm the original reading.

Measure completeness per account rather than applying a blanket waiting period. Take several closed months, record what each looked like at three, seven and fourteen days after close against its final total, and you have that account's own maturity curve. It takes an hour and it is reusable for as long as the sales cycle holds.

Then use it. State the completeness alongside any recent figure you report, exclude in-progress periods from volume comparisons, and hold any judgement about a recent window until the curve says the number has settled.

Rate metrics and volume metrics deserve different treatment here. Volume in a partial period always reads as a fake drop and should be excluded. A rate is still real information about the traffic that has been counted, so include it, marked as partial.

It is not conversion lag in the actuarial sense, although they share a cause. Lag describes the distribution of delay between click and outcome. Completeness is what that distribution has done to a specific period as at a specific reading date.

It is not a tracking fault. An account that keeps adding conversions to a closed month is working correctly. The fault would be reading it as if it were finished.

It is not fixed by shortening the attribution window. That makes the month settle faster and makes it settle lower, which trades one distortion for a different one.