Data - A grown-up view of attribution post-cookie - Ask a consultant

Insight · Data

Attribution and measurement in 2026.

Cookies are fragmented, platforms self-report aggressively, and AI features blur the funnel. Here is the measurement stack Npexpos now recommends - platform-honest, holdout-tested, and CFO-readable - without pretending any one view is the truth.

12 min readNpexpos consulting team

TL;DR

  • No single method tells the full story. Triangulate: platform, MMM-lite, incrementality.
  • Platform-reported ROAS is a planning input, not a source of truth.
  • Lightweight marketing mix modelling is within reach of most mid-market brands now.
  • Invest in geo-tests and holdouts - they are the only honest arbiter when platforms disagree.

The honest starting point

There is no one measurement method that captures modern marketing. Last-click is wrong, data-driven attribution is a black box, platforms report what flatters them, and view-through remains an article of faith in most orgs. The only grown-up position is triangulation.

The three-legged stack

A defensible measurement stack has three legs:

  • Platform data: directional, good for campaign-level optimisation, never taken at face value for channel-level truth.
  • Marketing mix modelling (lite or full): gives a top-down view of channel contribution. Weekly or monthly, never daily.
  • Incrementality testing: geo-splits, ghost ads, holdouts - the honest arbiter when platform and MMM disagree.

Platform-reported ROAS is a plan, not a truth

Use platform ROAS for planning and optimisation inside a channel. Do not use it for cross-channel budget reallocation. The cross-channel decision belongs to MMM plus incrementality.

MMM-lite is now realistic for mid-market

You do not need ten years of weekly data. Modern open-source and commercial MMM-lite tools work with 18-24 months and give usable channel contribution estimates. Use them as planning aids, not as dashboards.

"All models are wrong. Some are useful. The useful ones are the ones that change decisions."

First-party data underneath it all

Attribution quality is bounded by data quality. Server-side tagging, enhanced conversions, consent-mode compliance and clean identity resolution are the infrastructure that lets the three legs stand up. Without them, every method gets noisier.

Reporting upwards

The CFO does not want three attribution views. They want one planning number with a confidence interval. Build a single "blended contribution" report, show the sources, and be explicit about what changes your confidence in it.

Need this built?

Npexpos's analytics and attribution engagement rebuilds tracking, measurement and reporting in 8 phases - from raw events to CFO-ready dashboards.

See analytics & attribution →

Related reading

Rebuild your measurement stack

Share your current analytics setup and the cross-channel decision you cannot confidently defend - we will come back with a shape for the rework.

Request a scoping call