Ask a team how data migration is going and you'll usually hear a status: "We're exporting customers this week." Ask who decided which customers are worth moving, and the room goes quiet. That gap is where most ERP timelines quietly bleed out.
Migration gets treated as a technical chore — extract, transform, load, done. In practice it's a long series of business decisions disguised as a spreadsheet. Which of the four spellings of this vendor is real? Does a 2019 quote that never closed need to exist in the new system? Who owns the pricing on an item nobody has sold in three years? Engineers can move anything. They cannot decide what's true. When that decision-making isn't assigned, the project defaults to "bring it all over," and you pay for that choice every day after go-live.
Sort Everything Into Four Buckets
Before anyone writes an export script, put every data set your current system holds into one of four categories. This single exercise does more to shrink a migration than any tool.
- Master data — customers, vendors, items, chart of accounts, BOMs, price lists. This moves, and it must be clean. It's the foundation everything else attaches to.
- Open transactions — unpaid invoices, open POs and sales orders, work in progress, on-hand inventory. This moves as balances and open documents, not as full history.
- History — closed orders, old invoices, past journal entries. This is where teams over-commit. Decide how many years you genuinely need live, and be honest that "in case someone asks" is not a requirement.
- Archive — everything else. It stays accessible in a read-only copy of the old system or a data warehouse, not in your new ERP.
The archive bucket is the one people resist and the one that saves the project. A legacy database kept read-only for a defined period costs very little and removes the emotional argument for dragging a decade of noise into a clean system.
Clean Data Is a Business Job, Not an IT Job
Deduplication, naming standards, item categorization, customer hierarchies — none of this can be resolved by the person running the load. It requires someone in sales to say which account is the parent, someone in operations to confirm the unit of measure, someone in finance to approve the mapped chart of accounts.
Name those owners explicitly, by data domain, with dates. Give each one a working file and a deadline that lands well before the first dry run. If you can't staff that, your migration scope is too big — shrink the history, not the cleanup.
There's also a quiet benefit here. Cleanup forces the conversations nobody has had in years: why three price lists exist, why two departments categorize the same product differently. Those answers become the configuration decisions in your new system. Skipping cleanup means importing the disagreements along with the records.
Dry Runs Are the Project, Not a Phase
One migration rehearsal is a demo. Three or four is a process. Each cycle should load into a full test environment, get reviewed by the people who own the data, and produce a defect list that shrinks each round.
What you're building across those cycles is a runbook: the exact sequence, the timing of each load, who checks what, and how long the whole thing takes end to end. By the final rehearsal, cutover weekend should be boring — a script someone has already executed, not an experiment.
Also measure the clock. A load that takes eleven hours when you budgeted four changes your entire cutover plan, and you want to learn that in a rehearsal rather than on a Saturday night.
Define What Has to Tie Out
Agree in advance on the reconciliation checks that decide whether a load passed. Typically that means the trial balance, AR and AP aging by bucket, inventory quantity and valuation, open order counts and totals, and record counts by object. Write down the tolerance for each — some rounding is acceptable, a missing invoice is not.
Then name who signs. The controller signs the financial reconciliation. Operations signs inventory. Sales signs the customer list. A migration without named signoffs produces a go-live where nobody is sure the numbers are right, and that uncertainty follows you into month-end close.
The Cost of Getting This Wrong Is Delayed, Not Immediate
Bad migrations rarely blow up on day one. They surface six weeks later when a report doesn't match, when a salesperson finds duplicate accounts, when finance can't explain a variance. By then the system is live, the team is busy, and fixing data in production is far harder than fixing it in a staging environment.
That's the real argument for treating migration as a decision project. The decisions are the hard part, they belong to your business, and they're much cheaper to make before the data moves.
If you're planning an Odoo or ERP rollout and the migration plan is still a vague line item, we'd be glad to talk through scope, bucket strategy, and a rehearsal cadence that fits your team. Reach out to Infraxio and let's map it before anyone writes an export.
