The Measurement Maturity Model: Six Levels from UTM to Incrementality
Most brands are stuck at Level 2 trying to run Level 5 programs. A diagnostic framework for understanding where your measurement infrastructure actually is — and what it takes to get to the next level.
“Most brands are stuck at Level 2 trying to run Level 5 programs. That’s the gap where measurement programs fail — not because the methodology is wrong, but because the data foundation underneath it can’t support the question being asked.”
The Six Levels
Level 1 — UTM-Based Attribution (Multi-Claim)
What it looks like: GA4 + UTM codes matched to channels. Any channel present in the conversion journey claims credit — last touch, but with multi-channel visibility. Can get granular to campaign name, ad name, or creative.
The failure: Structurally overcounts. The same conversion is claimed by every channel that had a touch. No deduplication, no path weighting.
Why it still matters: More honest than raw platform self-attribution. At least you’re working from a consistent tracking layer you control.
Who’s here: Most small and mid-market brands. Anyone who has set up GA4 properly but hasn’t invested beyond it.
Level 2 — Normalized Cross-Channel Reporting
What it looks like: One dashboard with consistent metric definitions across Google, Meta, TikTok, DV360. Deduplicated reach. One view of spend and performance across all channels.
The failure: Still platform-reported attribution underneath — the normalization fixed the display problem, not the measurement problem. You’re looking at cleaner numbers built on the same flawed inputs.
Who’s here: Most agency reporting platforms. Brands that have invested in a reporting layer but not a measurement layer.
Level 3 — First-Party Data + User Journey Stitching
What it looks like: A CDP or data warehouse where every interaction is recorded with a timestamp. Full user journeys can be constructed. CRM data joined to media exposure data. You know which customers were exposed, not just who clicked.
The failure: Signal gaps are structural — physical retail, linear TV, out-of-home, and direct mail don’t produce a touch event. Identity resolution is hard; the join key between ad platforms and CRM is often inconsistent (PII normalization, hashing differences across ingestion pipelines).
Additional nuance — Privacy-safe collaboration: Clean rooms (Snowflake, Google Ads Data Hub, LiveRamp) are how you solve the match problem when you can’t share raw data. Publisher exposure log meets advertiser CRM without either party seeing the other’s raw records. This is a capability that layers on top of Level 3 infrastructure, not a separate stage.
Who’s here: Brands that have invested in a CDP or data warehouse. Enterprise advertisers with active clean room programs.
Level 4 — (intentionally not a separate level)
Clean room collaboration and privacy-safe data sharing are treated as a nuance on Level 3, not a distinct maturity stage. The infrastructure is the same; the collaboration layer is an extension of it.
Level 5 — MMM (Marketing Mix Modeling)
What it looks like: Top-down, observational. Models the contribution of spend to outcomes at aggregate level using historical data — including channels that produce no digital touch event (TV, radio, OOH, print). Bayesian MMM with clean spend inputs and aligned offline sales signals. Answers: where should the next dollar go across the full channel mix?
The failure: MMM revival is real but fragile. Most brands rebuild on the same fragmented, inconsistently-defined spend data that made their last program unreliable. The model converges on an authoritative-looking answer built on bad inputs. Garbage in, authoritative garbage out.
Who’s here: Large brands with dedicated marketing science teams. CPGs, large retailers, brands with significant offline spend. Anyone trying to measure channels that Level 3 can’t touch.
Level 6 — Incrementality
What it looks like: Experimental and causal. Geo holdouts, synthetic control groups, A/B at scale. You know what actually caused a conversion, not just what correlated with it. Pre-planned holdout groups. Answers: did this specific campaign, channel, or creative actually work — or would those customers have converted anyway?
The failure: Requires pre-planned experimental design. You can’t run a holdout after the fact. Most teams don’t build the discipline into their planning process. Geo experiments require sufficient geo-level volume to produce statistical power.
Who’s here: Sophisticated marketing science teams. Brands willing to deliberately withhold spend from a control group. The methodological gold standard — but operationally demanding.
Diagnostic Questions
Use these to place a customer on the curve in a discovery conversation:
- “Where does your ROAS number come from today?” → Level 1 or 2
- “Do you have consistent UTM taxonomy across all your channels?” → Level 1
- “Do you have a single dashboard across all paid channels with consistent definitions?” → Level 1/2 boundary
- “Do you have a clean join key between your ad platforms and your CRM?” → Level 2/3 boundary
- “Can you construct a user journey — every touchpoint, timestamped, tied to a person?” → Level 3
- “What’s your match rate when you join media exposure to CRM records?” → Level 3
- “Have you run an MMM in the last 18 months? What did you feed it?” → Level 5
- “Have you run a geo holdout or incrementality test in the last 12 months?” → Level 5/6 boundary
The Line
“The brands that invest in measurement governance today — solving the join key, fixing the identity resolution layer, building the clean room architecture — are building the foundation for safe agentic AI tomorrow. Everyone else will be automating guesswork at machine speed.”
Supporting Frameworks
Medallion Architecture — The Real Version
Most measurement programs fail at Silver Step 1. Everyone focuses on the dashboard (Gold). Nobody invests in encoding what “impression” means consistently across their channel mix.
The standard Bronze → Silver → Gold model hides two completely different operations inside “Silver”:
Silver Step 1 — Semantic layer: Metric normalization. “Impression” = served impression, minimum 1 second in-view, for all display sources. “Conversion” = last-touch within 7-day window, per agreed attribution model. Campaign taxonomy normalized across sources. Define once, version it, apply everywhere. Failure mode: every analyst re-derives these definitions in a spreadsheet. They diverge. The institutional knowledge walks out the door with the analyst.
Silver Step 2 — Identity resolution: CRM data joined to exposure data. Email → hashed token → RampID or UID2. Match rate calculated and documented. Privacy controls applied. This layer cannot live in a SaaS aggregation tool — it requires a governed data warehouse with full audit trail and consent management. Failure mode: trying to do identity resolution in a SaaS reporting tool, or skipping it and running a clean room on unresolved identity — producing a 48% match rate and not knowing why.
The MMM inputs, clean room queries, and incrementality analysis all live in Gold+. They’re only trustworthy if Silver Step 1 and Step 2 were built correctly.
Change Readiness as Design Principle
The question to ask when reviewing any measurement architecture: what happens when Meta changes their attribution window? Because they will.
Three principles:
1. Version everything in the Silver layer. Metric definitions, normalization logic — versioned with effective dates. When Meta changes their attribution window, you can restate historical data using the old definition and forward-calculate using the new one. You don’t lose your trend line.
2. Schema evolution at Bronze, not Silver. When a new channel is added (CTV, DOOH, retail media), it adds a new Bronze schema. The Silver normalization layer absorbs it without rewriting Gold. Bronze is schema-agnostic; Silver Step 1 is a mapping layer, not hardcoded transformation logic.
3. Institutional knowledge in the platform, not the person. When the analyst leaves, the normalization rules should still run. If those things live in that analyst’s head or their personal dbt repo, you start from zero when they leave.
Punch Lines
- “They were a Level 2 trying to run Level 5.”
- “Measurement modernization starts with the join key, not the model.”
- “A 48% match rate is almost always a data hygiene problem, not a clean room problem.”
- “The dashboard looks beautiful. The number is wrong.”
- “Everyone is deploying agents. Almost nobody is building the governance infrastructure that makes agents trustworthy.”
Ready to get clarity on your marketing data?
Start with a Marketing Data Audit — two hours, $2,000 flat.
Book a Strategy Call →