CDP greenfield: migration from a legacy data warehouse in 14 months

Context

The operator started the CDP initiative 18 months ago inside the existing legacy DWH. Data engineering tried to turn a batch warehouse into a real-time profile store. Latency would not drop below 6 hours. Channel teams did not use it — too slow.

CDO admits: “we picked the wrong architectural foundation”.

Diagnostic (3 weeks)

• Legacy DWH: optimised for analytical queries, not for low-latency lookups.
• Identity resolution rudimentary — 65% match rate.
• Schema bound to a 2014 biller dump structure, change cycle six months.
• Data engineering team capable, but stuck in legacy patterns.

Recommendation: greenfield CDP with migration of legacy use cases.

Approach (14 months)

Months 1-2. Architecture design. Profile store on cloud-native (Cassandra or managed equivalent), event ingestion via Kafka, identity resolution rebuilt with probabilistic matching.

Months 2-6. Foundation build. Legacy DWH still works in parallel. Greenfield CDP has no users yet.

Months 6-9. Pilot use case (CVM marketing). 3-5 marketing scenarios on greenfield. Latency <500ms, identity match rate 91%. Marketing team validates.

Months 9-12. Migration of remaining use cases (NBA, churn modelling, contact centre personalisation).

Months 12-14. Decommission of legacy CDP module in the DWH. DWH stays for pure analytics.

Results

• CDP latency p95: 6 hours → 280ms.
• Identity match rate: 65% → 91%.
• Channel adoption: 2 channels → 9 channels using profile.
• Time to launch a new use case: 3-4 months → 2-3 weeks.
• Cost: greenfield was 30% more expensive than the legacy approach in year one, but continuing the legacy approach would have been tribal — would never have reached target.

What is critical

Decision to greenfield, not sunk-cost fallacy. 18 months of investment in the legacy approach were expensive, but another 18 months would be disaster.

Parallel operation of legacy and greenfield. A big-bang switch would have broken marketing operations.

Strong CDO sponsorship. CFO needed to understand “greenfield not waste, but reset”.

What would not have worked

Continuing the legacy approach. 36 months later the same situation, $M wasted.

Outsourcing to a managed CDP vendor without an internal foundation. Vendor lock-in + identity resolution gaps.

“Hybrid”: writing data into both stores. Operational chaos.

How SamaraliSoft engages

An engagement of this class — 12-18 months. Diagnostic 3-4 weeks, foundation build 6 months, migration use case by use case.

Related

• /en/architecture/telecom-cdp-architecture/ — CDP architecture
• /en/decisions/telecom-build-vs-buy-cdp/ — build vs buy
• /en/architecture/telecom-event-bus-architecture/ — event bus
• /en/architecture/telecom-master-data-management/ — MDM