Operating ProcedureExecution layer

CDP Migration Runbook (Delivery)

A sequence with 7 steps: Map the client's current data sources and destinations before touching the CDP.

By InnovaAI ResearchPublished

What are the steps?

sequence

CDP Migration Runbook (Delivery)

  1. 01

    Map the client's current data sources and destinations before touching the CDP

    Inventory every system that generates customer data, from CRM and email platforms to ad managers and the data warehouse. Confirm which sources are live, which are stale, and which destinations the client actually needs to activate.

  2. 02

    Define the unified profile schema with the client's team

    Agree on the identity resolution keys, the fields that must be populated for every profile, and the data retention rules. This schema becomes the contract that every source and destination must honor.

  3. 03

    Run a pilot sync with a single high-value source

    Pick one source with clean data, such as the CRM or a primary email platform, and sync it into the CDP. Validate that profiles resolve correctly and that the data lands in the warehouse or destination without errors.

  4. 04

    Validate identity resolution against a known segment

    Create a test segment of 100 known customers and confirm the CDP merges their records across devices and channels. If the match rate is below 95%, revisit the identity rules before scaling.

  5. 05

    Activate the pilot segment to one destination

    Send the test segment to a single ad platform or email tool, such as Meta Ads or a marketing automation platform. Verify that the audience counts match and that the data is usable for targeting.

  6. 06

    Document the data flow and governance rules for the client

    Produce a one-page diagram showing how data moves from source to profile to destination, and note any privacy or consent requirements. This document becomes the reference for future changes.

  7. 07

    Schedule a full rollout with a rollback plan

    Plan the migration of remaining sources in waves, with a checkpoint after each wave to review data quality. If a wave fails validation, roll back to the previous state and fix the issue before continuing.