The National Audit Office estimated poor data quality cost the NHS around £1 billion a year. The figure has been public since 2021. It is referenced in business cases, strategy documents, and the occasional Public Accounts Committee session. It is rarely the headline issue on any board agenda.
Every regulated organisation I have worked with has an equivalent number. Most boards have never seen theirs. Data quality fails slowly. It fails invisibly. By the time it appears on the executive radar, the cost is already locked in and the remediation budget is being negotiated under pressure rather than chosen on merit.
Why the number does not land
A billion pounds is a startling figure. The shape of the failure is why it does not behave like one.
A cyber breach announces itself within days. A safety incident produces a coroner's report. A regulatory fine arrives in writing with a return address. Each of these has the property which makes it boardroom-legible. There is a single event, a clear date, a named cause, and a measurable cost.
Poor data quality has none of these. The cost emerges across thousands of small decisions made on incomplete or incorrect information. It accumulates in duplicated records, missed billing, failed linkages, and reports which look clean but tell a misleading story. No single decision is ever obviously wrong. The aggregate is catastrophic.
Boards are trained to respond to events. The data quality problem is a condition. The two require different responses, and most governance frameworks are built for the first.
What gets reported instead
What boards do see, in the place where data quality should be, is a maturity index, a compliance scorecard, or a project status report. The NHS Data Quality Maturity Index published a national average of 91.1% for Q3 2022-23. The number sounds reassuring. It is also, in isolation, almost meaningless.
A 91.1% score does not tell a board which decisions were made on the missing nine per cent. It does not tell them whether the missing data was random or systematic. It does not tell them whether the affected records are the operational long tail or the highest-risk customers in the system. The score is a reporting object, not a decision-support tool.
This is the trap. A board which receives a maturity index quarterly believes the topic is being managed. The evidence of management has replaced the evidence of impact. Asking the right question becomes harder, not easier, because the wrong answer is already on the table.
The same pattern plays out across financial services consumer duty reporting, energy sector market data submissions, manufacturing supply chain records, and any sector where regulators have asked for an attestation rather than an investigation. The metric becomes the management response. The underlying condition is unaddressed.
The NAO figure for the NHS represents the visible portion of the cost. Operational teams burn capacity working around bad data. Decisions get re-made because the first version turned out to be based on incorrect inputs. Strategic plans built on flawed segmentation produce results which look like execution failure but are something else entirely. Regulators raise concerns which trigger expensive investigation cycles.
Each of these has a cost. None of them appears on a finance system as data quality. They appear as overtime, as project overruns, as unexplained variances, as compliance spend. The number on the board pack is small because the accounting categories were not built to surface the real one.
A BMJ Health and Care Informatics study found automated data quality checks reduced data errors by up to 70% in the systems studied. Read the figure for what it implies about the starting position. If automation reduces errors by seventy per cent, the underlying error rate was substantial. Most boards do not know what their starting figure is.
What a board should ask
Three questions move this from a reporting topic to a governance topic.
The first is what decisions of consequence depend on which datasets, and what the failure mode looks like when the data is wrong. Not the structural failure mode. The decision-level failure mode. If customer segmentation is incorrect, what does the marketing spend look like next quarter. If safety reporting is incomplete, what does the regulator see.
The second is who owns the data quality of the inputs to those decisions, and whether they have the authority to refuse a downstream use of data which is not fit for purpose. The honest answer in most organisations is no one, and no.
The third is how the cost of poor data quality is currently classified in the management accounts. If the answer is it is not classified, the £1 billion figure is not surprising. The cost is invisible because the accounting system was not designed to make it visible.
These three questions take a discussion which currently lives in the technology directorate and place it in the boardroom. The technology directorate cannot answer them on its own. The board has to decide it wants the answers.
I have spent twenty-five years working in environments where the data quality problem is well understood by the people closest to it and almost invisible to the people accountable for the consequences. The pattern is consistent. The fix is governance, not technology. Putting the cost on the board agenda, owning the question, and treating data quality as a board-level topic rather than an operational one is what changes the picture.
The NAO figure was published in 2021. It has not changed the conversation in the way it should have, because the figure alone is not enough. What changes the conversation is a board which decides to ask what the equivalent figure is for their own organisation, and refuses to accept a maturity index in place of an answer.
If the question has not been asked at your board, the silence is the issue.