What this checkpoint covers
The second module of the intermediate marketing analytics level: the decision-first framework in the guide to choosing an attribution approach, the comparison of mix modelling and multi-touch attribution and the inputs, transformations and outputs in the marketing mix modelling basics guide. It follows the measurement approach exercise.
Task
- Answer every question in a sentence or two, with your reason, before reading the explanations.
- Mark any answer where you named the right method for the wrong reason; those count as misses.
- For each miss, reread the linked section and rewrite your answer.
Questions
- A retailer spends heavily on radio and posters. Its AttributionAssigning conversion credit to touchpoints. reports give both almost no credit. The finance director proposes cutting them. What do you say?
- A team wants a mix model, but spend on every channel has been almost flat for two years. What will the model be able to tell them about those channels?
- Search and social budgets have always been raised and lowered together. The first mix model gives search a large contribution and social a small one. How much weight do you put on that split?
- A modeller includes branded search volume as a control. Television's estimated contribution comes out close to nothing. What is the likely problem?
- A channel shows a strong average return over the model's history but a weak marginal return at today's spend. Which number should guide next quarter's budget, and why?
- The mix model's optimiser recommends tripling spend on one channel. What do you do with that recommendation?
- The paid social manager wants to use the mix model to decide which creatives to pause this week. Is that a good use of it?
- Attribution says a remarketing campaign is the best performer; the mix model says paid social overall is near saturation. How do you settle it?
Answers explained
- Absence of credit is not absence of effect. Radio and posters produce no clicks, so path-based attribution cannot see them. Their value is a channel-level question for a mix model, ideally checked by a regional test, as the comparison of the two methods sets out.
- Very little. A model learns from variation; a flat series gives it nothing to relate to sales, and the result reflects the modeller's assumptions. Planned bursts and pauses are the cheapest fix, as the mix modelling basics guide explains under data requirements.
- Little. Channels that move in lockstep cannot be separated, so the split rests on assumptions rather than evidence. Present it with wide intervals, and break the lockstep with a test before acting on it.
- A control caused by the media. Television drives branded search, so treating brand search as a control hands it television's effect. Model it as an outcome instead.
- The marginal return. It is the slope of the response curve at current spend, so it says what the next unit of budget would earn. The average describes the past and can hide saturation. Plan on marginal, report on average.
- Treat it as a direction to test. The optimiser extrapolates beyond spend levels the history contains, where the curve is least certain. Move part of the way and run an IncrementalityConversions caused by marketing. test on the extra spend, then recalibrate.
- No. Mix modelling (MMMStatistical model of channel impact.) works at channel level, over weeks, refreshed periodically; it cannot see creatives. Creative choices belong to the platform's attribution and in-platform tests.
- With an experiment. Remarketing often reaches people already on their way to buy, so attribution credits it heavily. A holdout that withholds remarketing from a random group measures what it adds, and the result can calibrate the model, as the guide to choosing an attribution approach describes in its staged roadmap.
Where to go next
If you missed the attribution questions, reread the decision table in the comparison of mix modelling and attribution. If you missed the model questions, work through the inputs, adstock and saturation sections of the mix modelling basics guide. Then start the third module with the spend reconciliation exercise.
