Methodology · the open part

The method.
Not a black box.

If we told you this was a black box, you should walk away. Measurement that is closed-source is just marketing. Here is how we actually do it — and what has been independently audited.

DOC: 42pp PDF
CODE: Open-core MMM
AUDITORS: Big-4 · annually
ADVISORS: 2 econometricians, 1 methodologist

01 · The method

The receipts behind the receipts.

If you cannot describe the method to an outside statistician in under 5 minutes, it is not a method — it is a sales slide. Here is ours.

01
Representative panelsN=1,200+ per market, sourced through ESOMAR-accredited providers. Age × gender × region weighted to census.
02
Matched exposed/controlPanels matched on 14 pre-declared covariates. Residual imbalance reported in every study.
03
Pre-registered hypothesisOutcome metric, cells, MDE, power, holdout — all locked in a time-stamped document before launch.
04
Difference-in-differencesDay-zero baseline subtracted from day-N read. SEs clustered. Multiple-comparison correction on secondary metrics.
05
Adstock + saturation modelledNot retrofitted after the fact. Structural form declared upfront; parameters estimated from the data.
06
Feeds the mix modelEvery lift study calibrates the Bayesian MMM. Prior becomes posterior. Posterior becomes the new prior.

02 · Audit trail

Big-4 audited. Third-party checked.

Our methodology has been reviewed by a Big-4 firm twice — annually since 2024. The audit report covers panel sourcing, weighting, statistical method, and operational controls. Available under NDA.

2024 AUDIT

First-time pass

Methodology reviewed. Three medium findings, all closed within 30 days. Zero material findings.

2025 AUDIT

Clean opinion

Zero material findings. Two minor process improvements adopted.

ISO 27001

Full certification

Information security management system. Data handling, access controls, incident response — all audited.

GDPR

Article 28 DPA

Standard DPA on file. You are a controller, we are a processor. Panel data never cross-border without SCCs.

03 · The open parts

We publish the things you can argue with.

DOCS
Methodology PDF · 42 pagesDownloadable. Every equation. Every covariate. Every panel-weighting decision. Redlined on each release.
CODE
Stat model is open-coreOur Bayesian MMM implementation is on GitHub. Proprietary parts: creator index, workflow. Measurement: open.
REVIEW
Academic advisorsTwo senior econometricians on retainer. One market-research methodologist. They veto changes we would rather not explain.

The method.Not a black box.