DECIDE · STEP 4 OF 4

The 12 vendor questions

1. 01Which of your priors are informative, and what informed them?

   GOOD:Named sources: category studies, our own experiments, physical plausibility bounds; plus prior-versus-posterior plots on request.

   FLAG:"Our priors are proprietary" ends the methodology portion of the call.

2. 02Show me prior versus posterior for my three biggest channels.

   GOOD:Posteriors that visibly moved and tightened; the vendor explains what data drove the movement.

3. 03What is your refresh cadence, and what changes between refreshes?

   GOOD:A stated cadence (weekly to monthly), plus a change-log discipline: what moved, why, and alerts when estimates lurch.

4. 04How do you validate out of sample?

   GOOD:Holdout periods by default, accuracy reported in business units (revenue error, not just MAPE), and honesty about when accuracy degrades.

5. 05How do estimates behave across refreshes?

   GOOD:Stability is tracked as a metric. Channel ROIs should drift, not lurch, absent real market change.

6. 06How do experiments enter the model?

   GOOD:A concrete calibration mechanism: lift tests as priors or constraints, with worked examples. This is standard in modern frameworks; managed vendors should match it.

   FLAG:"Our model is accurate enough not to need experiments" is a red flag with a confidence interval of zero.

7. 07How do you handle channels that move together (collinearity)?

   GOOD:Frank acknowledgment: merged reporting where separation is impossible, plus advice to desynchronize spend.

8. 08What does my team have to do, monthly?

   GOOD:A specific data contract: feeds, event logging, promo calendars, and the hours it costs you.

9. 09What granularity do you actually stand behind?

   GOOD:Channel-level with confidence; campaign-level labeled as modeled allocation, not measurement, if offered at all.

10. 10Who owns the data and the model outputs at offboarding?

    GOOD:You own your data and all outputs; export formats named in the contract, not the sales call.

11. 11What does pricing scale on, itemized?

    GOOD:A formula: channels, markets, refresh cadence, experiments; each priced separately so you can trim scope.

12. 12Which brands like mine have left, and why?

    GOOD:A real answer. Every vendor has churn; one that admits its failure profile is telling you the truth about fit.

Questions 1, 2, and 6 assume the vocabulary of priors and incrementality; ten minutes in the glossary beforehand makes you the best-prepared buyer that AE meets this quarter, a bar set lower than it should be. Pricing strategy for the same call lives in step 3.