INSIGHT · ALASTAIR MAJURY

Migration exceptions: give every unresolved record a business decision

A practical way to turn migration exceptions into owned decisions before cutover, with clear treatment, evidence and review dates.

By Alastair Majury · Published · 5 minute read

A migration exception report can look reassuringly precise. It lists rejected records, validation failures and reconciliation differences. Yet a detailed report does not tell a business sponsor what should happen next. A rejected record might need correction, an agreed temporary treatment or a decision to delay part of the move. Those choices have different consequences for customers, operations and the programme.

The useful question is therefore broader than how many exceptions remain. For each material group, can the team explain the business consequence, the proposed treatment and who has authority to accept it? The following approach is a practical consulting method, rather than a claim that one template will make every migration safe.

Start with the decision the exception prevents

Describe the business activity affected before describing the technical error. An invalid reference value might prevent a transaction from being allocated. A missing relationship might leave a customer record without the information an operator needs. These are illustrative possibilities; their importance depends on the actual service, data and controls.

Ask the receiving team to explain the consequence in plain language. Can the process continue? Would someone have to intervene? Could the problem remain hidden until a later event? This conversation helps distinguish inconvenient formatting problems from failures that undermine a critical business outcome. Avoid assigning a severity solely because a validation rule happens to be labelled mandatory.

Define the population and its limits

A decision needs a bounded population. Record which extract, migration cycle, rule version and selection criteria produced the exception group. Include the number of affected records, the relevant denominator and any excluded population. If the report covers only a sample, say so explicitly.

Separate what the evidence demonstrates from what remains unknown. A clean result for one product or period does not establish readiness for every product or period. Where the source changes between rehearsals, explain whether the previous analysis still applies. A sponsor should be able to understand the reach of the finding without opening the query that generated it.

Offer treatments that operations can actually use

Possible treatments include correcting the source, changing an agreed transformation, holding records outside the migration, or operating a temporary controlled process. None is automatically appropriate. Each needs an assessment of business impact, dependencies and the cost of keeping the treatment in place.

Consider a fictional example: a small group of customer records lacks an attribute required by the receiving workflow. A proposed manual queue sounds workable until the team asks who will process it, how cases enter the queue and what happens if volumes increase. A useful treatment describes those details, the evidence needed to close a case and the point at which the arrangement must be reviewed.

Name the decision owner and the evidence owner

The person who investigates an exception may not have authority to accept its consequences. Make that distinction visible. Identify who supplies the analysis, who recommends the treatment and who can authorise the relevant business risk. Where several functions are involved, record the decision route rather than collecting ambiguous acknowledgements.

Do not let a delivery deadline become an implied approval. A decision record should show what was accepted, for which population, on what evidence and subject to which conditions. If an owner cannot decide because information is missing, capture the missing information as an action with a named provider and a date.

Keep acceptance tied to a specific migration cycle

A temporary acceptance should not silently become permanent. State the review trigger: a date, a volume threshold, a successful correction or a particular release. Preserve the rationale so a later team can understand why the original choice was reasonable and whether its assumptions still hold.

At the next rehearsal, test the proposed treatment alongside the data movement. Check that records reach the expected destination, that operational users can identify them and that the exception evidence can be reconciled back to the population. A treatment that exists only in a meeting note is still an untested dependency.

Use the register to support the readiness conversation

The Government Data Quality Framework describes quality in relation to intended use and encourages evidence-based improvement. It is government guidance, not a certification of a commercial migration. The decision register described here applies that broad idea to a specific delivery conversation; it is an original recommended approach.

Before cutover, review unresolved groups with the people who will own the service afterwards. Ask whether the population is understood, the treatment is executable and the acceptance conditions remain true. If those answers are incomplete, record the gap clearly. The value of an exception register is the quality of the decisions it supports, including a well-supported decision to wait.

Related reading: Data migration readiness.

Source and scope

Government Data Quality Framework. Published 3 December 2020. Checked 20 November 2026.

Back to all insights