A retailer we worked with had a go-live date that had moved three times, and each slip traced back to the same place. The technology was configured. The processes were agreed. But every time the team loaded data into the new system and tested it, the numbers did not hold up — duplicate customers, suppliers under several spellings, opening balances that did not agree with the ledger they came from. Each failed load chipped away at confidence in the whole programme.
The honest diagnosis was uncomfortable but useful: this was not really an ERP project with a data problem. It was a data project with an ERP attached. So we reorganised it around that truth.
The challenges we had to solve
- Master data spread across several legacy systems, with no agreement on which source was authoritative for any given record.
- Opening balances that had to reconcile exactly to the closing position of the old system, or the first reports would be worthless.
- Years of dead records — inactive customers, obsolete items — that nobody wanted to carry forward but nobody had time to weed.
- A team treating migration as a final-week task, when it needed to be the spine of the whole plan.
How we approached it
We made data migration its own workstream with its own owner, rather than a step at the end of someone else’s plan. The first job was to decide, for each kind of record, which system was the source of truth — and to stop pretending several conflicting copies could all be right. From there we cleaned at the source where we could, profiled what was left, and agreed plainly what would be migrated, what would be archived, and what would simply be left behind.
For balances, we built a reconciliation that ran with every trial load: the system would not be considered ready until the migrated opening position tied back, account by account, to the old ledger’s close. We ran that loop repeatedly, and each pass the gap shrank and the team’s trust grew, until a load finally reconciled clean.
Where it stands
The retailer went live on data that reconciled to the old system on day one, which is the difference between a go-live people trust and one they quietly second-guess for months. Just as valuable, they came out of it with master data they had actually agreed on, and a clear record of where each kind of record now lives.