Before the Plug Is Pulled: A Data Governance Checklist for Retiring UK Systems
Most organisations treat the retirement of a CRM, an ERP or a reporting platform as a housekeeping task: a licence ends, a contract expires, a replacement goes live, and the old system is switched off. The data it held is often the last item considered. That ordering is a mistake. In UK organisations, the obligations attached to data do not expire when a subscription does, and the evidence a board may one day need is frequently produced by the very system being removed. Decommissioning deserves to be run as a data-strategy event, with an owner, a checklist and a record of decisions.
Begin with obligations, not with hardware
The first question is not which server to turn off but which data categories the system holds and what the organisation is required to do with each of them. Retention duties come from several directions: statutory rules for tax and employment records, sector rules for financial services and health, contractual terms with customers and partners, and the storage-limitation principle under UK GDPR. These duties differ in kind. Some require that data be kept; others require that it be available on request; others require that it be deleted once the purpose has passed.
A workable approach is to build a short table before any switch-off date is agreed. For each data category, record the retention driver, the minimum and maximum retention period, the business owner, and the intended disposition: transfer, archive or delete. Include any active legal hold, and confirm that the hold has been reviewed by whoever maintains it. This table is the difference between a controlled retirement and a hope that nothing is asked for later.
Can the number still be explained?
Analytics has a habit of outliving its source. A dashboard built on a retired system may remain in circulation as a PDF, a slide or a copied spreadsheet. When a figure is questioned eighteen months later, the organisation needs more than the number itself. It needs the definition, the calculation logic, the reference data used at the time and a record of any manual adjustments.
Lineage work is therefore part of the retirement plan, not an optional extra. Capture the data dictionary entries for the fields that matter, the transformation rules, and a snapshot of the code lists and mappings in use. Where reports are being migrated, test that the new output reconciles to the old before the old system goes dark. Where reports are being retired, decide explicitly what evidence will remain and where it will sit. A provenance note attached to the final extract is a modest artefact with a long useful life.
Dependencies are wider than the asset register
Asset registers rarely capture the full set of dependencies. A system may feed a regulatory return, a pricing model, a commission calculation or a set of regional performance reports that nobody has documented. It may also support a spreadsheet that a team relies on daily and that has never appeared in any inventory.
Before switch-off, ask the people who use the system rather than only the people who administer it. Review scheduled jobs, integration logs, API keys and service accounts. Look for embedded analytics in customer-facing products and for automated emails that carry extracts. Each dependency should be assigned a disposition: migrated, replaced, retired with the system, or retained under an exception with a named owner. Unowned dependencies are the ones that cause difficulty after the event.
Access records and the question after switch-off
Access records are often overlooked because they feel like security housekeeping rather than data governance. They matter for two reasons. First, they show who could see what, which is useful when a subject access request or an internal review arrives. Second, they reveal the accounts that must be closed: service accounts, shared mailboxes, API credentials and embedded reporting tokens.
Retain access logs in line with the retention schedule, and record the date on which access was revoked. If data is archived rather than deleted, test that the archive can be searched by the identifiers the organisation actually uses, such as customer number, employee reference or case ID. An archive that cannot be searched is not an archive; it is a cost.
The supplier exit is part of the plan
Contract exit terms determine how much of this work is possible. Check the notice period, the required format for data return, the transition assistance offered, the deletion certificate and the treatment of backups. Confirm which sub-processors hold copies and how their retention periods interact with erasure requests. Where the supplier hosts the archive, establish whether it is searchable and for how long.
These points are easier to negotiate before the relationship ends than afterwards. Where a contract is silent, record the gap and the mitigating step; a board is better served by a known weakness than by an assumption.
Archive, delete or something in between
Retirement does not have to mean deletion, but it should mean a decision. For each data category, choose between a full archive, a reduced archive of schema and reference data, an aggregated extract, or deletion with a certificate. Apply minimisation: keeping everything indefinitely is a liability, not a precaution. Record who approved the choice, particularly where the data is sensitive or the system supports a regulated process.
The one-page evidence pack
The most useful output of the whole exercise is a single page that a board or an audit committee can read. It should state what is being retired and why, the data categories involved, the retention basis for each, the disposition chosen, the dependencies cleared, the access revoked, the supplier exit confirmed, and the residual risks with owners and dates. That page turns a technical shutdown into a governed decision.
A system can be switched off in an afternoon. The obligations it carried cannot. Treating retirement as a data-strategy event is not bureaucracy; it is the cheapest insurance available against a question that arrives long after the switch has been thrown.