No single measurement method survives signal loss, so the ones that work now run as a system: a model for contribution, experiments for validation, granular signals for tuning. The scorecard is changing with the methods. Pick the decision in front of you and this page shows the method built for it, the method that checks it, and the number that reports it.
One question sits under everything here: what did marketing actually cause, and where do buyers now decide? Start with the decision in front of you, or read straight through. The methods come first. The scorecard follows, with the traditional KPIs and the new ones on one structure.
Six decisions run most measurement conversations. Pick yours and the page highlights the method built for it, the method that checks it, and where the result lands on the scorecard.
Media mix modeling is built for this. Check its answer with incrementality tests. Report contribution and marginal ROAS.
The marginal read from your media mix model decides it. Validate the biggest moves with a test. Report marginal ROAS.
Run an incrementality test. Use the model for context. Report causal lift and incremental ROAS.
Tactical attribution and platform signals carry this. Spot-check them with incrementality tests. Watch the channel tier.
Clean-room analysis is the only surface with the data. Design lift tests inside it. Report partner lift and incremental ROAS.
Track Share of Model and AI Inclusion Rate. There are no sessions to count, so visibility is the metric.
If the decision is which of these systems to build first, that is a different kind of call. The note at the end of this page is for you.
Four methods, each broken alone, each covering another's blind spot. Media mix modeling for contribution, incrementality testing (designed experiments against a holdout) for validation, multi-touch attribution and data clean rooms for in-flight tuning. The map plots them by cadence and granularity, and the lines between them show which methods check and complete each other. Click a line or a method to see what each pairing does.
The map does not fit a phone, so here is what it says: how the four methods check and complete each other.
Experiments validate the model's contribution estimates and keep it honest.
The model allocates across quarters; attribution tunes inside the quarter. Each checks the other's channel readings.
Clean rooms give the model the partner-level detail it reads too coarsely on its own.
Tests confirm whether attributed conversions were actually incremental.
Clean rooms restore the partner and walled-garden visibility attribution lost.
Clean rooms host the designed lift tests you cannot run on partner data any other way.
Contribution from the model, validation from the experiment, tuning from the signals. The pairings make the system. Positions on the map are directional, not measured.
Every measurement organization runs some version of one cascade: business goals at the top, channel telemetry at the bottom, latency rising as you climb. The new metrics extend that structure rather than replacing it. This section shows both on one scorecard: what you already track at each tier, and what is arriving next to it.
This scorecard combines established measurement practice with emerging metrics for AI-mediated discovery and agent operations. Each new metric carries a maturity label. Adapt the emerging ones to your own data and decision context.
Established: standard practice in measurement teams. Emerging: named in industry research, adoption still low. Proposed: a framework measure; adapt before adopting.
Few, episodic, slow to move. Did marketing do what the business needed, and at acceptable cost? Effectiveness and efficiency live here.
These two are how the top of the cascade stops running on correlation.
Always on, moderate latency. Are the programs working, from acquire through onboard, retain, and winback?
mROAS runs allocation here; ARC splits conversion reporting so agent-initiated transactions read separately from human ones.
Low latency, sensitive, high volume. What is happening right now, by channel and by system?
Visibility metrics for a surface that sends no referral traffic.
The tier the cascade never had, because the workforce was never software. It measures the agents that now run parts of the execution.
Nothing sits here yet. That is the point.
Built with room for what joins it next.
A working measurement system contains an allocation model, an experimental validation program, tactical signals for tuning, and, where buyers discover you through assistants, visibility tracking. The scorecard separates causal outcomes, marginal allocation signals, visibility, and agent operations, and it keeps the traditional tiers underneath. The common mistake is asking one method to do every job on this page. Three questions for your measurement team: Do we have one method for allocation, one program for validation, and one set of tactical signals? Can we explain where they disagree? Does the scorecard distinguish correlation, incrementality, and marginal return?
If the decision in front of you is which of these systems to build first, that is an operating-model conversation, and one I have often. Get in touch.
Measurement is one of eight capability areas in the Marketing Maturity Framework, and the methods here map to its Analytics & Insights climb from dashboards to agentic analytics.
Open the framework →