Before you cut a channel, four things have to be true.

1. 01

   The revenue figures agree, or the gap is quantified

   Platform-reported revenue and booked revenue must be on the same basis before they are compared. If they are not, the channel that looks worst may simply be the one whose customers take longest to pay.

2. 02

   The target was derived, not inherited

   A channel is only underperforming against a break-even calculated from your own contribution rate. Judged against a target somebody typed in last year, a profitable channel can look like a failure.

3. 03

   The period is comparable

   In the UAE, Ramadan, the summer exodus, the shopping festivals and the school calendar move demand more than most channel changes do. Comparing July with April is not a test of the channel.

4. 04

   The channel has been tested, not only measured

   Reported performance tells you what a platform recorded. A holdout or a staged reduction tells you what happens to the business without the spend. Only the second answers the question a cut is asking.

• Reduce rather than stop. A 30% reduction held for a full purchase cycle produces evidence. A full stop produces a confound you cannot undo for a quarter.
• Stage it. Cut in one emirate, one branch, or one product line, and keep the rest as the control.
• Fix the measurement window before the cut, and set it to at least one full purchase cycle plus the payment lag.
• Write down in advance what result would reverse the decision. A cut with no reversal condition is not a test.
• Name the owner and the review date at the same time as the cut, because an unreviewed cut becomes permanent by default.

This is the part usually skipped, and it is the reason the same argument returns every quarter. Six months after a cut, with none of it recorded, nobody can say whether revenue fell because of the cut, because of the season, or because of something nobody was watching.

• The baseline the decision was judged against, and the date it was taken.
• The evidence used, including what it did not cover.
• The expected effect, in a figure, with the reasoning.
• The person who approved it and the limits they approved.
• The result that would reverse it, and the date it will be reviewed.
• The measured outcome against that baseline once the window closes.

• Condition one: reconciling platform revenue with the accounts
• Condition two: deriving a break-even ROAS from contribution
• How PAS95 keeps that record as part of the decision