The migration patterns enterprises use to unify legacy data estates without disrupting the business.
Many enterprises have already built a data lake once. Too often it became a data swamp: large volumes of raw data that few people trust and fewer can use. A lakehouse promises the flexibility of a lake with the reliability of a warehouse — but the technology alone does not fix the problem. How you migrate matters more than what you migrate to.
Why migrations stall
Data modernisation programmes usually stall for the same reasons: they try to move everything at once, they start from source systems rather than business needs, and they leave governance for later. The result is a new platform carrying the same old problems.
Start from decisions, not tables
Begin by listing the business decisions the data must support — pricing, replenishment, credit, claims — and work backwards to the data products those decisions need. This gives the programme a clear order of work and a way to show value early, instead of spending a year copying tables nobody uses.
Land, then refine
A layered design keeps the platform understandable:
- Raw — data landed as-is from source systems, immutable and fully traceable.
- Refined — cleaned, conformed, and de-duplicated data with consistent keys and definitions.
- Curated — business-ready data products, modelled for specific domains and owned by named teams.
Each layer has its own quality checks, and consumers only build on curated data.
Build governance in from day one
Cataloguing, lineage, access control, and data-quality rules are far easier to establish on an empty platform than to retrofit onto a busy one. Every curated data product should have an owner, a definition, quality expectations, and documented lineage back to its sources.
Migrate by domain, run in parallel
Move one business domain at a time. For each, run the legacy and new pipelines side by side, reconcile outputs until they match, and only then switch consumers over. Parallel running costs more for a short period, but it is what makes cutover a non-event for the business.
Measure progress by consumers moved
Terabytes migrated and pipelines rebuilt are activity measures. The real measure of progress is how many reports, models, and applications now run on the new platform — and how many legacy systems have been switched off.
A phased roadmap
- Foundation — platform setup, security, catalogue, and the first priority domain.
- Expansion — additional domains migrated in parallel, with reusable ingestion and quality patterns.
- Decommission — consumers switched over and legacy systems retired.
- Activation — analytics, machine learning, and AI built on curated, trusted data products.