Moving master and transactional data into a new system — the phase that delays more go-lives than any other.
Data migration is the extraction, cleansing, mapping, loading and reconciliation of data from legacy systems into the new ERP. Master data — customers, suppliers, items, accounts — moves first; balances and open transactions follow at cutover. Historical transactions are a decision, not an obligation.
Migration delays go-lives more than configuration does, for a reason that is structural: the cleansing is the client's work and cannot be outsourced. Only your people know whether two similar customer records are the same company. That work competes with their day jobs, which is why it has to start in the first month rather than the last.
What exists, how much of it, how dirty it is. Duplicate rates and missing mandatory fields are the useful measures.
What migrates and what does not. Carrying ten years of transactions is expensive and rarely used — decide deliberately.
Legacy to target, with transformation rules written down. Ambiguities surface here and are cheap to resolve.
Load into a test environment, read the errors, fix at source, repeat. Three or more full rehearsals is normal.
Record counts and control totals against source. A load that completed is not a load that is correct.
A business with eleven years of history in its legacy system assumes it must migrate all of it. That assumption, unexamined, is often the single largest cost in the project.
A more useful division is by what the new system actually needs to operate. Open items are mandatory: outstanding receivables and payables, open orders, current stock, active master data. Reference history is usually worth one to two years of closed transactions, so users can look up a recent invoice without leaving the system. Everything older is archive — needed for statutory retention and occasional lookup, but not needed inside the new ERP.
Keeping the legacy system available read-only for a defined period, or extracting the archive to a queryable store, satisfies retention at a fraction of the cost of migrating and reconciling eleven years of transactions into a different data model. The conversation to have early is which of the three buckets each data set falls into, because 'migrate everything' is a decision that nobody actually made — it is what happens when the decision is deferred.
| Term | What it means |
|---|---|
| Extract | Pulling data from the legacy system, in a repeatable scripted way rather than by hand. |
| Transform | Mapping and cleansing to fit the target structure. |
| Load | Importing into the new system, usually through a supported bulk mechanism. |
| Reconciliation | Proving what loaded matches what was extracted, by count and by value. |
| Cut-off | The point at which legacy data is frozen for the final extract. |
It is the single most reliable predictor of a delayed go-live.
History has a cost and usually little use. Archive it accessibly instead of loading it.
Opening balances that do not tie to the legacy system make the first close impossible to sign off.
Each pass surfaces problems the previous one masked. Plan for several.
| Metric | What it tells you | What distorts it |
|---|---|---|
| Records extracted vs loaded vs reconciled | The three-way check. Gaps between any pair mean something was lost silently. | |
| Duplicate master records | Measured before migration. The number that justifies cleansing effort. | |
| Rehearsal elapsed time | The input to the cutover plan, knowable only by rehearsing fully. | |
| Post-load reconciliation variance | Control totals against source. The only real proof it worked. |
We build and support the systems this page describes. Start with a free working proof of concept — no cost, no commitment.