Data Migration
Migration moves data from the system being replaced into the one replacing it, and decides what history survives. It is consistently under-estimated, consistently on the critical path, and the single most common cause of a delayed go-live.
Data in and out
- OUT Opening balances and master data to General Ledger · at cutoverWithout it: The new ledger opens on numbers nobody can reconcile to the old one.
If you change this
It implies 6 topics across 3 modules, including Integration and PMO & Programme Governance which you might not expect.
See the path ahead →Why it matters
Migration is where the quality of twenty years of data becomes everyone's problem at once. It is also where the cutover either works or does not.
- It is the critical path: Every dress rehearsal depends on it, and it is usually the last thing ready.
- It exposes quality: Data that was tolerable in the old system fails validation in the new one.
- It sets the opening position: The new ledger opens on these numbers, and they have to reconcile to the old one.
The workflow
- 01
Scope
What migrates, what is archived, and what is deliberately left behind.
- 02
Profiling
The real state of source data assessed, not assumed.
- 03
Cleansing
Issues fixed at source where possible, because fixing in transit hides the problem.
- 04
Mapping
Source fields mapped to target, with transformation rules recorded.
- 05
Trial loads
Repeated, timed, with results reconciled.
- 06
Reconciliation
Balances and counts agreed between source and target, signed off.
- 07
Cutover load
The real load, within the cutover window.
Decide what history you are not taking, early and explicitly. The instinct is to take everything; the cost of taking everything is usually discovered too late to reverse.
What you need in place
- Profiling before estimating: Scope set on what the data actually looks like.
- Cleansing owned by the business: IT cannot decide which duplicate customer is real.
- Reconciliation criteria agreed in advance: What "successfully migrated" means, numerically.
- Repeatable, timed loads: The cutover window is finite.
- A rollback position: What happens if the load fails at hour six.
Questions to ask
01History
How much history are we taking, and why that much?
The answer should be a decision, not a default.
02Profiling
Have we profiled the actual data, or assumed?
Assumption is why migration estimates are wrong.
03Cleansing
Who owns fixing data quality?
If it is IT, it will not get fixed, only transformed.
04Reconciliation
What are the sign-off criteria, and who signs?
Agree before the first trial load.
05Timing
How long does a full load take, measured?
It has to fit the window, with contingency.
Metrics and KPIs to track
Trial load success
Loads completing without manual intervention.
Load duration
Measured against the cutover window.
Reconciliation variance
Source against target, by object.
Data quality defects
Open, by severity and owner.
Records rejected
Per load, with reasons.
Alongside this