An NHS trust receives thousands of patient complaints a year. Before Quail, each one was read by a member of staff, categorised by hand, and typed into a case system. The reading alone consumed a large share of the complaints team’s week — before any of the actual resolution work began.
Quail does the reading. It extracts what the complaint is about, who it involves, how severe it is, and drafts the categorisation. What it deliberately does not do is file anything on its own.
Verify, don’t transcribe
The design principle that made the system land with staff: the human’s job changed from transcription to verification. A person still sees every complaint — but instead of reading a two-page letter and typing up a summary, they review a pre-filled record against the source and correct what’s wrong. Minutes become seconds, and the judgment stays human.
In healthcare, “the AI got it mostly right” is not a defensible position. “A person confirmed every record, ten times faster” is.
Where the checkpoint goes matters
We put the human checkpoint at the point of highest consequence — the moment a complaint’s severity and category are committed, because that drives statutory reporting. Lower-stakes fields flow through automatically. Putting a checkpoint everywhere would have recreated the original workload with extra steps; putting it nowhere would have been unshippable.
That placement question — not the model, not the prompts — is where most of the design effort went, and it’s the question I now start with on every document-intake system.