Attribution Window

Almost nobody buys in the moment they see an ad. The normal shape is returning three days later to sign up and paying ten days after that, and somebody has to decide whether that payment belongs to the ad.

Leave it undecided and the same month produces two numbers. That happened at Pagedeck. The platform report showed 480 conversions and Ravi's own count showed 260. Two weeks went into finding the cause, and neither dataset was wrong: the platform was counting 30 days after a click, and the internal query was counting seven.

Set one deadline for clicks and a separate one for impressions, and attach only the conversions that land inside them.

1. Keep the click window and the view window separate. Give someone who clicked 7 to 30 days; give someone who merely scrolled past about one. Stretch the view side and people who never registered the ad start arriving as results, which is what makes a view-through conversion untrustworthy.
2. A longer window makes performance look better. Moving from seven days to 30 reorders the channel ranking on identical data. Without touching the model, last month's top channel can land third.
3. Match it to how long the decision takes. When the trial is 14 days and the click window is seven, everyone who paid at the end of a trial drops out. Products sold to companies sometimes use 60 days, because security review and sign-off alone take weeks.
4. Use the same value across every channel. Platform defaults differ, so pulling each report as-is and lining them up isn't a comparison.
5. Write the value down, and overlap when changing it. A new deadline can't be read against the old numbers, so count the changeover month both ways.
6. There's a ceiling on lengthening it. At 90 days, people who arrived for reasons unrelated to the ad come in as results. Extend only as far as the influence plausibly lasts.

"The platform was counting 30 days and we were counting seven."

The explanation for two different conversion counts in one month.

"A 30-day window makes this channel first. Is there a reason to extend it?"

Blocking the habit of choosing the deadline that flatters the result.

"The trial runs 14 days, so a seven-day click window won't do."

The deadline is shorter than the decision, so whole conversions vanish.

Ravi fixed the standard at 30 days for clicks and one day for views, and wrote that line at the bottom of the reporting template. Once the same sentence appeared on every channel report, arguments about whose number was right turned into discussions about which standard applied. In the month the standard changed, both the seven-day and the 30-day figures went in so the gap was on record.