Every Field Has a Job: The Half-Day Reckoning for Columns UK Firms Capture and Never Consult

Ask a mid-market data leader to name every platform in the estate and a tidy answer usually follows: the CRM, the finance system, the warehouse, the marketing suite, the HR tool. Ask the same person how many fields the CRM contains, which of them are mandatory, which are read by a report, and which are retained because nobody has ever dared delete them, and the answer becomes a shrug. That gap is the subject of this piece. It is not a catalogue project and it does not require a vendor. It requires half a day, a spreadsheet of field names, and the discipline to ask the same five questions of every column.

The timing matters. Field-level sprawl is the quiet precondition for most of the failures that surface later: consent records that cannot be substantiated, retention schedules that describe the database rather than the reality, model training runs that ingest columns nobody has validated, and migration costs that balloon because the target schema must accommodate two decades of accumulated guesswork. None of these problems announce themselves as a field problem. All of them are one.

Why the system inventory stopped being enough

System inventories were the right instrument when estates were smaller and integrations were fewer. They told a board what it owned and what it paid. They no longer answer the questions that come up in practice. A subject access request arrives and the team must find every field holding personal data, including the free-text notes box that was never designed for it. An AI pilot begins and the data team discovers that the candidate feature set contains eleven columns with inconsistent formats and three whose meaning has changed since the last reorganisation. A renewal negotiation approaches and nobody can say which fields are load-bearing and which are ornamental.

A field-level view answers each of these directly. It is also cheaper to produce than most governance artefacts, because it is built from evidence the organisation already holds rather than from a new taxonomy.

The shape of the half day

The audit works best as four blocks of roughly an hour each, with a fifth reserved for the write-up. Do not attempt to cover the whole estate. Choose one system or one domain, ideally the one feeding the next piece of work — a customer platform, a supplier register, a workforce dataset. Breadth can come later; credibility comes from finishing something.

Block one: extract and de-duplicate. Pull the field list from the system metadata, the data dictionary if one exists, and the reports that actually get run. Reconcile the three. Discrepancies are the first finding and are usually substantial: fields that appear in metadata but nowhere in reporting, and fields used in reporting that the dictionary never documented.

Block two: trace the question. For each column, establish where a value is captured, who or what populates it, whether it is mandatory or optional, how often it is non-null, and which decision or output consumes it. The last of these is the test that separates a field with a job from a field with a habit. "It might be useful" is not a consumer.

Block three: score. Rank the shortlist on three axes — storage and processing cost, privacy and obligations exposure, and operational drag (the manual effort, reconciliation, and downstream confusion the field generates). A column that is empty but sensitive is a liability. A column that is full but unused is a cost. A column that is misused is a defect.

Block four: decide. Produce a keep-fix-retire list, and attach a named owner to every line. Unowned decisions do not survive contact with the next quarter.

Five questions for every column

First, what is the lawful basis or contractual reason for holding this value, and is it recorded where the person who needs it can find it? Second, which named report, process or model reads it? Third, what happens downstream if the field is wrong, stale or blank? Fourth, how long is it retained, and does the schedule describe the field or merely the table? Fifth, who is accountable for its quality and its definition?

Applied honestly, these five questions dismantle most of the ambiguity in an afternoon. They also surface the awkward cases: the field with a lawful basis but no consumer, the field with a consumer but no owner, the field whose definition has drifted and whose historical values are now uninterpretable.

Sorting the findings

A practical output groups fields into four bands rather than two. Keep covers fields with a documented consumer, an owner and a retention rule. Fix covers fields that matter but are unreliable — inconsistent formats, absent validation, ambiguous definitions. Retire covers fields with no consumer and no obligation; these need a decommissioning step, not merely a decision, and in regulated contexts the removal itself should be logged. Quarantine is the fourth band, and the most useful: fields that are sensitive, possibly justified, but whose consumer cannot yet be identified. Quarantine imposes a review date and a named reviewer, which is a great deal better than indefinite silence.

The quarantine band matters because it prevents the audit from becoming an exercise in deletion for its own sake. In UK organisations subject to the UK GDPR and the Data Protection Act 2018, the principle of data minimisation is real, but so is the risk of destroying evidence of a decision that a regulator, an auditor or a court may later ask about. Quarantine buys time without buying ignorance.

Making it stick

The audit only pays if it changes something within a quarter. Three moves help. Publish the keep-fix-retire list where the engineering and analytics teams already work, not in a governance repository nobody opens. Convert the retire list into tickets with owners and dates, and treat a retired field as a delivered outcome worth reporting. Finally, add the five questions to the change process so that new fields arrive with an owner and a consumer rather than accruing silently.

There is a pleasing side effect. Teams that run this exercise once tend to run it again, because the first pass almost always finds a field that was quietly influencing a number the board has been quoting for years. That is the real argument for the half day: not tidiness, but the discovery that some of the estate has been making decisions on its own.