The hospital group where every number had three versions

The situation

A healthcare group assembled by acquisition inherits more than buildings. Each site arrives with its own patient administration system, its own coding habits, and its own quietly reasonable answers to questions like when a bed day starts. Individually none of that is wrong. Collectively it makes group reporting impossible.

This group had grown to four hospitals over eight years. Clinically it operated as one organisation. Informationally it was still four.

Under UK GDPR the group also had obligations that made ad hoc data movement uncomfortable: patient data is special category data, and a reporting process that works by emailing extracts between sites is difficult to defend to the ICO and harder to defend to a patient.

The problem

The monthly board pack took nine working days to produce. Two analysts spent most of that time not analysing anything: they were pulling extracts from four systems, reconciling them by hand in spreadsheets, and chasing site managers about discrepancies.

The discrepancies were the real problem. Occupancy at one site counted a bed day from midnight; at another from admission time. Day cases were in the numerator at two sites and excluded at the other two. Neither convention was incorrect. There were simply two of them, and the group had been averaging across them for years.

The consequence was that the board had stopped trusting the pack. When a number looked wrong, the discussion became about the number rather than about the thing the number described. A board that argues about measurement is not governing.

What we did

We started by refusing to build anything for six weeks, and instead ran a definitions exercise with the clinical and finance leads in the same room. We catalogued every metric in the board pack and every variant of it in use. There were thirty-seven definitions behind what the pack presented as eleven metrics.

This was the unpopular part. The group had approved a data platform, and we spent the first phase running workshops about vocabulary. Our argument was that a platform built on thirty-seven definitions produces thirty-seven kinds of wrong faster, and with more authority, which is worse than spreadsheets.

The moment it turned was when the finance director and the medical director discovered they had spent two years disagreeing about occupancy while measuring different quantities. Once that was on a whiteboard, the remaining definitions took four weeks rather than the six we had allowed.

We then built the platform against agreed definitions, with lineage recorded so any number on the board pack can be traced back to the source system and the rule that produced it. Site data flows in on a schedule rather than by email, with access controlled by role and every extract logged.

Governance came with it, not after it: a named owner for each metric, and a change process, so a definition cannot drift again without somebody agreeing to it.

The outcome

The board pack now takes about four hours rather than nine days, and the two analysts spend the reclaimed time on analysis.

Thirty-seven definitions became eleven, each owned and documented. When a number looks surprising now, the conversation is about why the underlying thing moved, because the definition is not in dispute.

The platform was never the hard part. Building a warehouse is well-understood work. Getting four organisations that had been merged on paper to agree what a word meant was the engagement, and the technology was what made the agreement stick. A shared dictionary is not glamorous, but an organisation without one is four organisations wearing the same logo.