Three years late, with data migration among the causes, on data the supplier itself held. I have reconciled enough opening balances to report that the data is never surprised. Only the plan is. Four beliefs keep producing that plan.
Myth one: migration is a technical task
The belief is that migration means extract, transform and load, so it belongs to the technical team. The tools are the easy part. The work is a series of business decisions about what the company is prepared to say is true: which of two records for the same customer survives, and whether a disputed invoice crosses over as open.
The National Audit Office said much the same ten years later. Its March 2026 update on government shared services, where HM Treasury has committed £1.15 billion to three of the five clusters, found that delays in adopting common standards "have affected data convergence and caused delays across some of the clusters". Most clusters reported finance the easiest function to prepare, because it had implemented its standard. Agreement came before movement.
So here is my rule: budget migration by the number of decisions, not the number of records. Clean rows load over a weekend, however many there are. Unanswered questions about ownership and valuation each wait for a meeting that nobody has called. Count the decisions, put a name and a date against each, and the estimate becomes honest.
Myth two: we will clean it in the new system
You will not. After go-live the budget closes and the people who know why the data is wrong go back to month end. The new system then enforces rules the old one forgave, so a record that was merely untidy now blocks an invoice.
Clean at source, before the first mock load, while the old system can still show how the mess was made. Cleansing is cheapest where the evidence lives.
Myth three: all the history must come across
The case for bringing everything is sentiment dressed as prudence. The real question is statutory: what must be kept, for how long, and in what form.
HMRC's VAT Notice 700/21 requires business records to be kept for at least six years. Section 36 of India's CGST Act sets seventy-two months from the due date of the annual return. The UAE's Federal Tax Authority reminded businesses in August 2025 that corporate tax records must be held for at least seven years after the tax period ends. In Saudi Arabia, ZATCA ties the storage of e-invoices to Article 66 of the VAT Implementing Regulations, and where the records sit in a cloud outside the Kingdom they must be accessible from the branch inside it.
None of these rules says the history must live in the new ERP. They say it must be retained and producible on request. That is an archive with a search screen and a named owner, and it costs far less than converting every closed purchase order into a new data model. Migrate open items and balances, plus the history your operations query every week.
What an agent will read on Monday
Until recently, bad history was merely slow. Someone ran a report, distrusted it and rang a colleague. An AI model rings nobody. A forecast trained on years of sales that include duplicate customer hierarchies learns the duplicates. An agent matching invoices reads the vendor master you chose to keep and acts on it at volume, without the raised eyebrow.
Migration decisions have therefore become AI design decisions. What you carry forward is what software will believe about your business. A small history that somebody owns and can describe is a better asset than a complete one nobody will vouch for. We expect auditors to ask which data an agent was permitted to read, and who signed for its quality.
Myth four: the cutover weekend is the risk
Cutover is only where the risk becomes visible. It was created months earlier, in the meeting where nobody agreed to own the customer master.
Picture a Saturday, the third mock cutover. The board on the wall lists every exception from the overnight load: customers failing tax-number validation, a stock variance at one depot, a supplier bank account that does not match. Beside each line is a name, and none of them belongs to the project team. The credit controller and the depot manager each hold a rule agreed in advance and the authority to apply it.
By lunchtime the finance lead has signed the opening balances against the old ledger. The agent that starts matching payments on Monday is pointed at a vendor master with a signature on it. Nobody is awake at two in the morning. That Saturday is dull. Dull is the deliverable.
Count the decisions before you set the budget
This quarter, ask your programme for one list: every data decision that must be made before cutover, with its business owner and the date it falls due. Any row without a name is an overrun you have found before paying for it.
An Dependable Core readiness review with NectarGlobal builds that list with your finance and operations leads, and tests it against the retention rules of each country you operate in.