Data migration budgets fail in the same place almost every time: reconciliation. Writing the pipelines is the cheap part. Proving that the new warehouse returns the same revenue number finance has reported for the last nine years, then finding the undocumented rule that makes the two disagree, is what turns a three-month plan into a seven-month one.
So split the purchase. Buy discovery as a small, separate piece of work with a written output, and only then decide how to buy the build. Any data migration consulting proposal that jumps straight to a delivery price is quoting a scope neither side has seen.
Where the effort actually goes
| Phase | Share of effort | What moves it |
|---|---|---|
| Source discovery and profiling | 20–30% | Number of source systems, quality of documentation |
| Modelling and pipeline build | 25–35% | Target platform, depth of transformation |
| Reconciliation and sign-off | 20–30% | Row-level versus aggregate tolerance |
| Cutover and parallel run | 10–20% | Downtime tolerance, number of downstream consumers |
| Decommissioning and backfill | 5–10% | History depth, retention obligations |
Two lines surprise buyers. Discovery gets priced as a formality by vendors who want the build. Reconciliation gets estimated as testing, when it is closer to forensic accounting against a system nobody in the building fully understands any more.
The three ways to buy the work
Fixed-price migration practices quote the scope you can describe and raise change requests on everything you could not. That trade works when the source is a packaged product with a published schema. It works badly when half the business logic lives in spreadsheets and stored procedures written by someone who left in 2019.
Platform professional services arms move fast on their own destination and know its failure modes better than anyone. They will not tell you the destination is wrong, and they rarely price the reporting layer you have to rebuild afterwards.
Rented senior engineers on time and materials keep the scope open, which is the honest shape for work whose size is unknown on day one. The cost is that you carry the estimation risk and need someone internal running the plan. Our note on who carries estimation risk in managed services versus staff augmentation sets out that trade.
What moves the price
- Source count. Each extra source system adds discovery, its own reconciliation and its own cutover window. Five sources is not five times one source, but it is nearer that than twice.
- Undocumented logic in the old reporting layer. If the legacy BI tool holds calculations that never reached the warehouse, someone reverse-engineers them before sign-off.
- Reconciliation tolerance. Aggregate matching at month level is a week of work. Row-level matching across ten years of history is a month or more, per domain.
- Downtime tolerance. A weekend cutover is cheap. A parallel run with dual writes for two months roughly doubles the cutover line.
- Regulated or resident data. Personal data under GDPR, or contracts requiring EU-only processing, restricts who may touch the extract and where test copies live.
For a mid-sized migration we would staff one lead plus one or two data engineers, and a part-time analyst from your side who owns sign-off. That analyst is not optional. No supplier can approve a reconciliation on your behalf. Current European day rates by seniority sit on our daily rates page, and the wider practice is described on the data engineering page.
When renting engineers is the wrong call
- You have a regulatory deadline with penalties attached and want one supplier contractually on the hook for the date. Buy that as a fixed-price engagement and pay the risk premium.
- The migration is genuinely small: one source, twenty tables, no history. The target platform’s own loading tools plus a fortnight of one contractor beats any staffed team.
- Nobody internally can approve that the numbers are right. A staffing model does not fix an absent data owner. Fix that first, or the parallel run never ends.
Four questions before you sign
- What is the reconciliation standard, and who signs it?
- How many source systems have we profiled, as opposed to listed?
- What happens to the legacy reporting layer, and who rebuilds it?
- Which named engineers start, and what did they migrate last?
A supplier who answers all four without hedging has done this before. Our guide on how to evaluate an engineering vendor in one call covers the rest of that conversation.