Tech and AI

AI routing goes live on the police non-emergency line, sending calls meant for councils and health services elsewhere before the caller joins the queue

By Staff Writer  |  20 August 2026

The Home Office and the National Police Chiefs' Council put a language model to work on 101 on 18 August, backed by 1.4 million pounds and delivered with Vodafone Business. The line takes 20 million calls a year, about four million of which are for other organisations. The estimated saving is up to 8.5 million pounds a year.

The problem the system is aimed at is not fraud and not crime. It is misdirection. The police non-emergency number receives 20 million calls a year, and around 20 per cent of them, four million, are meant for bodies such as local councils, the health advice line or the power cut helpline. Until now those calls were identified only after somebody had spoken to a handler, which wasted the caller's time and lengthened the queue behind them.

How it works

The model matches the reason a person gives for calling against a set of likely reasons, and redirects the ones that are not police business to the right agency before the caller enters the queue. It was developed by the Home Office with the National Police Chiefs' Council and the delivery partner, and it was tested in 15 forces before going into service. In those pilots it diverted damage-only road traffic incidents, fly tipping and noise complaints.

Three safeguards sit around the routing decision. The caller is told how the system has understood the reason for the call and must accept or reject that reading. Where the reason falls into a broad category, additional safeguarding questions are asked. If the system is in any doubt, the call goes straight to the police without delay, and a caller can always ask to speak to a handler.

The department also says the model was trained and tested to make sure its performance is not affected by a caller's accent or phrasing, which is the fault this class of system is most often criticised for. The emergency number is untouched, and the extent to which each force uses the new routing will depend on the telephone menus it already runs.

Police call handlers are wasting hours fielding calls meant for other organisations. Every minute spent on these calls is time taken away from helping people who genuinely need their support.

Sarah Jones, Crime and Policing Minister

The money

The build cost 1.4 million pounds of government investment. The estimated return is up to 8.5 million pounds a year released back to forces. It sits inside a wider 16.5 million pound programme to modernise how the police and the public deal with one another, which also covers automatic transcription of 999 and 101 calls and the linking of crime reports to identify patterns in demand.

The chief constable leading on contact operations for the chiefs' council described the deployment as a way of managing demand more effectively and freeing handlers to concentrate on people who genuinely need police support, and thanked the department and the delivery partner for the work.

Why a construction or legal reader should care

Two reasons, and neither is about policing. The first is the procurement shape: a government department, a representative body for the sector and a commercial delivery partner, piloted across 15 organisations before national deployment, with a stated cost and a stated saving. That is the model public bodies are now using for AI, and the contractual questions it raises about accuracy warranties, liability for a wrong routing decision and ownership of the trained model are the same questions in every sector.

The second is the design of the human override. The system does not decide alone: it states its reading of the call, the caller can reject it, ambiguity resolves towards the person rather than the machine, and a route to a human is always open. Anyone drafting or reviewing an AI clause for a client will recognise that structure, because it is what a defensible automated decision looks like when somebody has thought about who carries the risk of being wrong.