No single measurement method survives signal loss, so the ones that work now run together. The scorecard is evolving along with the methods.
What did marketing actually cause, and where do buying decisions actually happen now? The page covers the methods first, then the scorecard, with the traditional KPIs and the new ones side by side.
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.
Marketing mix modeling is built for this. Check its answer with incrementality tests. Report contribution and marginal ROAS.
Your marketing mix model's estimate of what the next dollar returns in each channel 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 user-level partner data. Design lift tests inside it. Report partner lift and incremental ROAS.
Track Share of Model and AI Inclusion Rate. Referral counts understate what assistants actually influence, because buyers often make their decision before they click anything, which makes visibility the place to start.
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.
None of these four methods works on its own. Each one covers a blind spot in another. Marketing mix modeling estimates contribution, incrementality testing (designed experiments against a holdout) validates it, and multi-touch attribution and data clean rooms handle the 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.
Each line on the map is one method covering another's blind spot. Select one to read what the pairing does.
The model tells you what each channel contributed to the business, the experiments tell you whether it's right, and the signals are for deciding what to change next. What makes it a system is the pairings. Positions on the map are directional, not measured. These four methods all sit in one layer of a much bigger picture. The Martech and Adtech Reference Architecture maps the systems underneath them.
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.
Every measurement organization runs some version of the same cascade. Business goals sit at the top, channel telemetry at the bottom, and the higher you go the slower the feedback. 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, and the Proposed label is my call. I think a Proposed metric belongs on the scorecard before the industry has fully adopted it, and it's credited to its source where it appears. Adapt the emerging and proposed 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 put experimental evidence behind numbers that have mostly run on correlation until now.
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 the assistants, whose influence on buying barely shows up in referral numbers.
A tier the cascade never needed before, because the workforce was never software. It measures the agents that now run parts of the execution.
Nothing sits here yet.
The tier has room for what comes 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. Four 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? When a read arrives, does it change what happens next quarter?
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 →