06 · One number everybody agrees on
If marketing and finance are reading different numbers, one of them is making decisions with fiction.
- 1
- Source of truth, agreed by marketing and finance
- 3 tiers
- Server-side capture, mix model, live experiments
- Reconciled
- Tied to bank-deposited revenue and real margin
The problem
Post-ATT attribution is a consensus hallucination. Platforms over-claim, last-click under-credits brand demand, and the dashboard everyone stares at each Monday agrees with no one. The fix is not a better attribution tool. It is triangulation, and the discipline to trust the experiment over the dashboard.
We build measurement in three tiers that check each other. Tier one is clean data capture: server-side events, consent-aware identity resolution and a warehouse that stores raw events rather than a vendor's interpretation of them. Tier two is a media mix model: the strategic read, telling you how budget should be split across channels over a quarter. Tier three is experimentation: geo-holdouts and lift tests that settle arguments the other two can only estimate.
Then we reconcile it to finance. Every number we report ties back to bank-deposited revenue and real gross margin, including returns, shipping and payment fees. It is slower than reading a platform dashboard. It is also the only version that survives a board meeting.
The output is deliberately boring: one dashboard, one definition per metric, one weekly rhythm. Boring is the point. When the numbers stop being contested, decisions get faster.
What we actually hand over
Built in your accounts, documented as we go, and yours to keep whether or not you renew.
- BigQuery
- GA4
- Meta CAPI
- Northbeam
- Triple Whale
- dbt
- Looker Studio
- 01
Server-side tracking via Conversions API and Google Ads Enhanced Conversions
- 02
Warehouse-native data model in BigQuery or Snowflake
- 03
Media mix modelling with quarterly re-fit
- 04
Incrementality testing programme: geo-holdout, PSA and blackout tests
- 05
Contribution-margin reporting reconciled to finance
- 06
Executive dashboard with a single agreed metric dictionary
- 07
Consent-mode and privacy-compliant identity resolution
The same discipline, run two ways
Lead generation
01High-consideration, offline close
Closed-loop by default. Lead source stitched to CRM outcome so you can see cost per booked appointment and cost per sold job, not cost per form fill. Call tracking, offline conversion import, and a lag-aware read, because the sale lands weeks after the click.
Ecommerce
02Direct-to-consumer, repeat purchase
Server-side capture, contribution-margin reporting reconciled to finance, media mix modelling above the spend threshold that justifies it, and geo-holdout testing to settle what the dashboards cannot.
What the first quarter looks like
Weeks 1–4
Audit
Tracking audit, data quality assessment, metric dictionary drafted, finance reconciliation started.
Weeks 5–12
Instrument
Server-side tracking deployed, warehouse model built, dashboard live, first holdout test running.
Month 4 onward
Model
Media mix model fitted and validated against experiments, quarterly re-fit and budget recommendation.
Questions we get about this
Only above roughly $500,000 per month in media spend and with two years of clean history. Below that, a well-run geo-holdout programme gives you a sharper read for a fraction of the cost, and we will recommend that instead.
Pairs with
Want to know what measurement & data would be worth here?
Send us your numbers. We will come back with where we think the leverage is, what it is plausibly worth, and what we would need to be true for it to work.
Engagements from $12,000 / month · 90 days, then month-to-month