Most attribution problems aren't attribution problems. They're data-quality problems. This scorecard shows you what to audit, how to score it, and what to fix first so your reporting stops lying.
The fastest way to waste a year is to buy a new attribution tool on top of broken inputs. When identity is fragmented, events are inconsistent, and the CRM is rotting, the best dashboard in the world just visualizes nonsense with confidence.
What is marketing data quality?
Marketing data quality is the degree to which your customer journey data is complete, consistent, matchable, and tied to downstream outcomes. In practice, it means your ad platform events can be reconciled to your CRM and revenue with minimal manual cleanup. Good data quality is not a vibe. It is a set of checks you can run every week.
- Identity spine
- An identity spine is the small set of stable IDs you use to connect web events, lead records, and revenue across systems. Typical spine IDs are email (hashed), phone (hashed), CRM contact ID, and an internal person_id. If you do not have an identity spine, you do not have attribution. You have guesses.
- Reconciliation rate
- Reconciliation rate is the percentage of “conversion” events reported by ad platforms that can be matched to a real person and a real outcome in your CRM or order system. It is the simplest truth metric you can track. If reconciliation is low, optimization is training on partial data and finance will not trust marketing numbers.
The 2026 reality: signal loss is permanent, not a phase
The cookie timeline changed, but the reliability problem didn't. On July 22, 2024, Google said it was proposing an updated approach that elevates user choice, and that instead of deprecating third-party cookies it would introduce a new experience in Chrome that lets people make an informed choice across their browsing. The point is not Chrome's decision. The point is that the web's direction is still toward fewer matchable signals.
Publisher economics show why this is hard to unwind. A study summarized by MediaCat reported that removing cookies entirely reduces publisher revenue by 29.1% relative to the status quo, while publisher revenue in the Privacy Sandbox condition fell 27.9% relative to the status quo. Whether you like those numbers or not, they explain why the industry is stuck in half-measures.
So you do not win by waiting. Per IAB's State of Data 2024, 71% of brands, agencies, and publishers were increasing their first-party datasets. This is no longer a niche best practice. It is the default direction of the market.
The scorecard: four layers, 20 points
This is the scorecard we use to diagnose why reporting doesn't reconcile. It is intentionally boring. That is the point. Score each line 0 to 5. Then fix the lowest line item first.
| Layer | What it tests | How to score 0 | How to score 5 |
|---|---|---|---|
| Identity | Do you have a stable identity spine from click to CRM? | No universal IDs. Email exists but is not normalized. No event_id strategy. | Hashed email/phone + CRM IDs are consistent. event_id is stable. Cross-system joins work. |
| Event design | Are events consistent, deduped, and mapped to real outcomes? | Different names for the same event. Duplicate fires. Value fields are wrong or hardcoded. | Clear naming. Deduped on event_id. Parameters are validated. Value is real and reconciles. |
| Transport | Do events arrive reliably across browsers and consent states? | Browser-only pixels. Missing events on Safari/iOS. No server-side redundancy. | Server-side delivery exists (CAPI / sGTM). Redundant paths. Monitoring on drops and lag. |
| CRM hygiene | Can you trust stages, lead source, and revenue fields? | Stages are subjective. Duplicates everywhere. Lead source is overwritten or blank. | Strict stage definitions. Dedupe rules. Required fields enforced. Source is immutable. |
Identity: fix the spine before you fix attribution
If you have to pick one thing, pick identity. Every other improvement depends on it. Your goal is a minimal set of IDs that can follow a person across web, CRM, and revenue. That means you need an explicit policy for normalization, hashing, and ID precedence.
- Normalize email and phone the same way everywhere (lowercase, trim whitespace, E.164).
- Choose a single internal person_id and write it to every system that can store it.
- Use a stable event_id for dedupe (order ID, form submission ID, or session-scoped UUID).
- Decide precedence (CRM contact ID beats email match; email beats anonymous cookie).
Event design: your taxonomy is your measurement model
Most teams do not have an event taxonomy. They have a pile of tags. A useful taxonomy is small: 10 to 20 events that map cleanly to funnel progress and revenue. Everything else is noise.
Transport: redundancy beats perfection
In 2026, relying on a browser-only pixel is optional. But it's the wrong option. You want two independent paths for critical events: browser and server. When one leaks, the other still trains the algorithm.
If you run Meta, you already have a diagnostic: Event Match Quality. It is not a vanity metric. It is a proxy for whether your identity spine is showing up in the event payload.
CRM hygiene: a dirty CRM makes MMM and MTA both fail
Your CRM is the source of truth for outcomes. If stages are inconsistent and records are duplicated, every model downstream gets trained on garbage. Fixing CRM hygiene is not a RevOps project. It is a measurement project.
- Lock stage definitions. Make each stage testable with criteria, not vibes.
- Make lead source immutable after first touch. Store “first touch” and “last touch” separately.
- Deduplicate on email and phone, with rules for which record wins.
- Make revenue fields mandatory for closed-won, with validation.
Frequently asked
Is this a replacement for an attribution tool?
No. It's a prerequisite. If your inputs do not reconcile, an attribution tool gives you a prettier lie. Fix the inputs first. Then decide whether you still need the tool.
What score is “good enough”?
Aim for 16 out of 20. Below 12 means you will argue about numbers in every meeting. Above 16 means you can start optimizing budgets with confidence.
Does this matter if we only run one channel?
Yes. Even single-channel teams need clean identity and CRM hygiene to know what actually converted, and to train channel algorithms on true outcomes.
How often should we run the scorecard?
Monthly for most teams, weekly during major tracking changes. Data quality drifts, so your audit has to be recurring, not a one-off cleanup.
What does Moonshot do differently here?
We treat measurement like a product. We define the identity spine, the event contract, the CRM contract, and the monitoring. Then we wire it through the stack so you can scale without your numbers falling apart.