The First-Party Data Playbook

Why most brands do not have a data problem

They have a data model problem. Same events named differently across surfaces. No stable customer key. Warehouses full of duplicates. Every dashboard tells a slightly different story and nobody can say which one is right, so decisions get made on intuition.

This playbook is the sequence we run when we take on a brand where the data layer has to be fixed before anything else works. It is deliberately slow at the start, because every hour spent stabilising identity saves ten hours downstream.

Phase 1. Days 0–15. Audit

Map every source, every event, every identifier. Score each source by trust and coverage. The output is a single spreadsheet the CEO can read in ten minutes. It is not glamorous, and it is the most important artefact in the entire program.

• Inventory all data sources: storefront, POS, subscription, support, ads, email, SMS.
• For each source, list every event and every identifier it emits.
• Score events on trust (does it fire when we think it does) and coverage (what share of sessions).
• Flag every place the same event is named differently.

Phase 2. Days 15–45. Rebuild capture

Unified schema. Single identity. One source of truth for orders, sessions, and consent. This is where most brands try to shortcut with a tool. The tool is not the answer. The schema is the answer, and the tool serves the schema.

• Publish a single event dictionary. Every property, every source, every consumer.
• Resolve identity: email, phone, device, order, in that order of priority.
• Route all events through one collection layer before any activation.
• Backfill 12 months of history into the new schema. Do not launch on partial data.

Phase 3. Days 45–90. Activate

Ship the first three use cases: winback, replenishment, VIP. Measure lift against a holdout from day one. If a use case cannot demonstrate incremental revenue against a 5% control after four weeks, kill it and try another. This is not sentimental work.

The seven mistakes that keep brands stuck

• Buying a CDP before defining the schema it should enforce.
• Treating identity as a project instead of a permanent function.
• Letting each team define its own version of orders and revenue.
• Running activation without a holdout, then reporting attributed revenue as incremental.
• Storing consent in the ESP instead of the source of truth.
• Assuming the warehouse is clean because it exists.
• Hiring an analyst before hiring an engineer.

What Loiale does inside this playbook

When we run this program, one senior engineer owns the 90 days end to end. We do not staff a team of five. Data work rewards depth and continuity, and the fastest programs we have shipped were owned by one person who could hold the whole system in their head.