What the Blanks Are Telling You: A Field-Level Audit of Missing UK Customer Records

A monthly data-quality dashboard reports 94 per cent completeness across the customer master. The figure is presented at a governance forum, noted, and filed. What it conceals is that the missing six per cent is not spread evenly. It clusters in the fields that decide whether a customer can be collected from by Direct Debit, whether a vulnerability indicator is set, whether a service handoff reaches the correct team, and whether a marketing segment is lawful to contact at all. Completeness is an average. Consequence is not.

The dashboard is not wrong so much as uninformative. It answers a question nobody in operations actually asks. The useful question is narrower: which empty fields change what the business does next, and who decided that the emptiness was acceptable?

Why an aggregate score misleads

Completeness metrics are usually calculated across a table, weighted by row count, and reported as a single number. That construction has three familiar flaws.

First, it treats a missing middle name as equivalent to a missing bank sort code. The arithmetic is defensible; the operational reading is not. Second, it hides distribution. A field that is 99 per cent populated overall may be 60 per cent populated among customers onboarded through a particular channel, which is exactly the cohort an eligibility rule is designed to catch. Third, it rewards remediation that fills cells rather than remediation that makes them trustworthy, an incentive that reliably produces placeholder values.

An aggregate score also tends to sit in the data team's reporting rather than the operations team's. The people who feel the consequence of an empty field — the billing analyst, the contact centre adviser, the fulfilment coordinator — rarely see the metric that describes it, and so have no standing to challenge it.

Three kinds of absence

The audit begins by refusing to treat all nulls as one phenomenon. In most UK customer estates, empty values fall into three categories that require entirely different responses.

Legitimately absent. The field does not apply to this customer or this product. A sole trader has no company registration number. A customer paying annually by card has no Direct Debit mandate reference. These nulls are correct and should be documented as such, not chased.

Conditionally absent. The value does not yet exist because a process has not completed. An address awaiting confirmation from a third-party verification service; a consent record awaiting a customer response through a preference centre. These are expected for a defined period and become defects once that period elapses.

Operationally absent. The value should exist, was required by a process, and is missing. These are the nulls with cost attached. They are also the ones most likely to have been quietly filled with a workaround value — an email of noemail@, a date of 01/01/1900, a postcode of ZZ99, a status of unknown — which converts a visible gap into an invisible one.

The third category is the point of the exercise. The first two exist so that the third can be isolated with credibility.

Running the field-level audit

A workable method has five stages, and none of them require a new platform.

1. Enumerate the fields that carry operational weight. Work backwards from decisions rather than forwards from the schema. For each downstream rule — an eligibility check, a segmentation criterion, a billing route, a service handoff, a regulatory obligation under UK GDPR or the FCA's Consumer Duty — list the fields it reads and the conditions it tests. This produces a shorter list than the data catalogue and a far more useful one.

2. Profile each field by cohort, not in aggregate. Break null rates down by source system, onboarding channel, product, tenure and, where relevant, by the team or supplier that created the record. Distribution is the finding.

3. Detect pseudo-nulls. Search for sentinel values, default dates, placeholder strings, whitespace, single-character entries and codes that mean 'not known'. In mature estates this step frequently reclassifies a third of the apparently present data as absent.

4. Trace each gap to its consequence. For a sample of records with an operational null, follow the record through the downstream process and record what happened. In practice this yields a small number of recurring failure patterns: records that bypassed a validation rule, records merged from a legacy migration, records created by an automated interface that silently dropped a field, and records where an adviser had no option but to proceed.

5. Score by consequence, not by volume. A gap affecting 40 high-value accounts may outrank one affecting 40,000 low-risk ones. Severity should reflect the decision at stake, the regulatory exposure and the cost of the workaround currently in use.

The acceptable-null contract

The audit's durable output is not a remediation list but a set of contracts. For each operationally weighted field, the business should record: what the field means, what an empty value actually denotes, which absences are permitted and why, the maximum tolerable null rate, the accountable owner, the evidence supporting that tolerance, the downstream consumers who must be notified of change, and the monitoring threshold at which the contract is considered breached.

This is unglamorous work and it is where most programmes stop. A contract that is not registered in the data catalogue, and not linked to a monitoring rule, will be forgotten within two quarters. The contract needs a home and a mechanism, not just a signature.

It is also worth stating plainly that some tolerances will be accepted rather than fixed. That is a legitimate outcome, provided the acceptance is explicit, owned and reviewed. The failure mode is not a null rate of eight per cent; it is an unowned null rate of eight per cent that three separate teams are compensating for in three different ways.

The workaround economy

Every operational null that survives long enough attracts a workaround. A team builds a manual check before running the billing file. An adviser learns to ring the customer for a missing reference. A marketing analyst excludes a cohort defensively, losing reach rather than risk a complaint. These compensations are rarely costed and almost never documented, which is why the case for remediation so often looks weaker than it is.

Recording them is part of the audit. When the manual checks, the exception queues and the defensive exclusions are counted, the empty fields stop looking cosmetic. They start looking like an operating model.

The blanks are not silence. They are a record of decisions nobody made deliberately — and, once audited, a reasonable place to start making them.