Programmatic Attribution

Uniquely hard. Post-cookie reality, models, view-through, clean rooms, MMM, incrementality.

Why this is uniquely hard

Programmatic spans many surfaces (display, video, audio, CTV), many devices (mobile, desktop, TV), and operates within shifting identity rules. Single attribution methodology doesn't suffice; triangulation is mandatory.

Post-cookie reality

• Third-party cookies gone or degraded.
• Identity stitching via Universal IDs (UID 2.0, ID5, RampID).
• First-party data activation grew.
• CTV measurement requires new methodology.
• Walled garden inventory black-box.

Attribution models

• Last click. Severely underweights display/CTV.
• Multi-touch. Possible where identity supports; degraded.
• View-through. Standard for display/CTV.
• Brand lift. For awareness-focused.
• MMM. Aggregate; works without identity.
• Incrementality. Causal inference.

View-through

• Conversion within window after view.
• Window typically 1–30 days depending on category.
• Doesn't prove causation; suggests influence.
• Discount factor often applied to view-through credit.
• Verification partners measure viewability before credit.

Clean room measurement

• Amazon Marketing Cloud (AMC).
• Google Ads Data Hub.
• Snowflake Data Clean Rooms.
• Habu (acquired by LiveRamp).
• Cross-platform measurement without raw user-data exchange.
• Mature for AMC; emerging for others.

MMM for programmatic

• Aggregate weekly impressions and spend by channel.
• Model contribution to outcomes.
• Calibrate with incrementality.
• Bayesian frameworks (Robyn, LightweightMMM).
• Vendor solutions (Recast, Haus, Northbeam).

Incrementality

• Geo holdouts for programmatic.
• User-level holdouts where DSPs support.
• Brand lift studies.
• Conversion lift studies.
• Quarterly cadence ideal.

Advanced playbook

• Triangulation: platform-reported + GA4 + MMM + incrementality.
• View-through with discount factor.
• Clean room measurement where supported.
• Bayesian MMM annually with calibration.
• Incrementality quarterly on top channels.
• Identity strategy (Universal IDs).
• First-party data activation in DSPs.
• Cross-screen measurement methodology.
• Stakeholder communication of measurement limits.
• Annual measurement methodology review.

Common mistakes

• Last-click attribution applied to display/CTV.
• View-through credited without discount factor.
• Single methodology trusted as truth.
• Clean rooms ignored.
• MMM not built.
• Incrementality not tested.
• Cross-screen attribution naive.
• Identity strategy missing.
• Stakeholders not educated on limits.
• Annual methodology review skipped.

Operating checklist

• Triangulation across methodologies
• View-through discount factor
• Clean room measurement where supported
• MMM annual rebuild
• Incrementality testing quarterly
• Identity strategy aligned
• First-party data activation
• Cross-screen measurement
• Stakeholder education on limits
• Annual methodology review

Part of the Programmatic / CTV series.