Analytics & measurement agencies, reviewed and ranked

Marketing mix modelling, attribution and incrementality. Scored on whether the method is published, whether the model is validated, and whether the firm grades media it also buys.

17 evaluated.

You are buying a number you will then spend against

Every other category on this site sells work you can look at. This one sells a claim about causality, and you will reallocate budget on the strength of it. That makes the usual questions secondary and one question primary: how do they know the model is right?

Ask how the model was validated, not how it was built

Any firm will describe its method. Far fewer will tell you how they checked it: a holdout period, a backtest, or a real experiment the model was reconciled against. ‘Proprietary algorithm’ is not a method — it is the least checkable claim a measurement vendor can make, and it should lower your confidence rather than raise it.

Correlation sold as causation is the category's characteristic error

Mix models and attribution describe what correlated with what. Establishing that spend CAUSED the outcome needs an experiment: a geo holdout, a matched-market test, a switchback. Firms that run those, or reconcile their models against them, are selling something different from firms that do not — and the price rarely reflects the difference.

Ask whether they also buy the media they are grading

A measurement firm inside a media agency is marking its own homework. That is not automatically disqualifying, and some do it honestly, but the conflict should be disclosed by them rather than discovered by you. Where it is not disclosed, our records name it.

Find out what happens when the engagement ends

Whether the model is refreshed or frozen, whether you keep the code and the data pipeline, and what a refresh costs. A model delivered once and never updated is a photograph of a market that has since moved.

Evaluated