The Causality Gap in MarketingAppendixRE-FG3 1.0
7 questions to ask before you believe a marketing result
One page for the board · from "The Causality Gap in Marketing" · Seifelden Alhajj · Revenue Engineering · Riyadh
Attribution tells you what happened around a marketing action. Causal analysis asks what happened because of it.
| # | The question | Why ask it | An acceptable answer |
|---|---|---|---|
| 1 | Compared with what? | Without a comparison there is no effect, only a number | Holdout branches, a matching Hijri period, or an earlier experiment |
| 2 | Who chose where it ran: a coin or a person? | People pick the branches or customers most likely to succeed | A coin flip within similar pairs, written down before launch |
| 3 | What was the calendar doing? | Ramadan moves about 11 days earlier each year; paydays and holidays move sales | Comparison on the Hijri calendar, not "March vs March" |
| 4 | Did we start from a dip? | A branch picked because it fell will rise on its own | A long pre-period and a control group picked the same way |
| 5 | Who marked the exam? | A platform dashboard measures the success of what it sells you, and each platform claims the same order | A number from your own point of sale, not from the seller alone |
| 6 | Where did the sales come from? | They may have moved from a neighbouring branch, another channel, or next week | Control branches with non-overlapping delivery radii, plus a post-period for pulled-forward demand |
| 7 | Was the decision rule written before the result, and could the test see the effect at all? | A result must not rewrite the question; a weak test says "no effect" even when the offer worked | A signed pre-registration card, and a minimum detectable effect smaller than the expected lift |
Board language: every marketing number carries one word
- Attributed: what a platform dashboard or attribution model says.
- Incremental: a measured difference against a control group or experiment.
- Modelled: an MMM output, stating whether it was calibrated by an experiment.
The rule: no campaign budget is doubled on an attributed number alone.
Sources and tools, including "Design your branch test", are in the full paper. seiflden.com