Category: Technology | Published: 2026-08-27
When someone calls 101, they may expect to speak to the police about a crime, concern, or local incident. But not every call belongs with a police call handler. Some are intended for another public service, while others are so inappropriate that they consume valuable time before anyone can deal with a genuine policing matter.
A new artificial intelligence system is being used to help sort those calls at the beginning of the process. The aim is not to replace the person who deals with a serious enquiry. It is to understand the reason for a call quickly, direct straightforward cases to the right organisation, and allow human operators to concentrate on people who genuinely need police assistance.
This is the practical idea behind an AI police call centre approach now being introduced in the UK. It offers a useful example of where automation can support public services without taking difficult decisions away from people.
What Is Changing for Calling 101?
A £1.4 million natural language processing system has gone live with South Yorkshire Police following trials involving 15 police forces. Rather than placing every caller into the same queue, the software listens to the explanation and identifies the likely purpose of the call.
If the request is not a police matter, the caller can be directed towards a more suitable organisation. Depending on the issue, that might be a local council, NHS 111, or the 105 electricity helpline for power cuts.
The system is designed to deal with the first stage of the conversation. People who need police help should still be able to reach a human call handler, while uncertain situations can be passed on rather than being automatically rejected.
That balance is important. The technology is being used to improve routing, not to decide whether someone’s experience is serious enough to matter.
Why the Police Are Looking at AI Triage
The scale of the 101 service helps explain the investment. Around 20 million calls are made nationally each year, and approximately four million of them are thought to be intended for another organisation. That represents roughly one in five calls entering a service that is already handling demand from victims, witnesses, and people concerned about safety.
Some examples of inappropriate enquiries are almost absurd. Police call handlers have reportedly received complaints about a late pizza delivery, slow service in a pub, or requests for a lift. These calls may be unusual, but even a short conversation takes time away from the control room.
An automated first step could identify those requests before they join the main police queue. That would help reduce avoidable pressure without asking the public to learn a complicated telephone menu or choose the correct department in advance.
The potential financial saving is also significant. The government estimates that the system could save policing as much as £8.5 million each year if it performs as expected. The wider benefit, however, is measured in attention. A call handler who is not explaining which organisation deals with a power cut can spend more time listening to a crime report or safeguarding concern.
How an AI Police Call Centre System Understands People
The technology uses natural language processing, commonly shortened to NLP. This allows software to analyse ordinary spoken language and identify what a caller is trying to achieve.
A caller does not need to use an official phrase. They might describe a noisy neighbour, a lost item, a concern about a vulnerable person, or a problem with a local service in their own words. The system looks for meaning and context rather than waiting for one exact answer from a fixed menu.
That is a demanding task. People speak with different accents, use local expressions, change direction during a conversation, and explain problems in different levels of detail. South Yorkshire Police says the software has been trained and tested extensively to reduce the effect of those differences.
The technology should still be treated as an aid rather than an authority. Speech recognition can misunderstand words, and intent classification is not always straightforward. A short description may not reveal the seriousness of a situation until a trained person asks a follow-up question.
For that reason, escalation needs to be easy. A caller who says something unclear, describes a possible crime, or appears vulnerable should not be trapped in an automated route simply because the first sentence was difficult to interpret.
Human Operators Remain Central
The most reassuring part of the project is the stated role of human staff. Calls that require police help are intended to reach a person, and doubtful cases should be handled with care rather than diverted for the sake of a clean statistic.
This is a useful model for public-sector automation. An AI system can be good at repetitive sorting, but empathy, judgement, and responsibility still matter when a person is frightened, confused, or reporting something serious.
The system should therefore be judged on more than how many calls it redirects. Police forces will also need to monitor false diversions, abandoned calls, repeated attempts to get through, and feedback from people who interacted with the service. A lower queue count is not a success if the wrong callers are being sent away.
Clear information is important too. People should understand when they are speaking to an automated system, what it can do, and how to reach a human if the situation does not fit a simple category.
Part of a Wider Digital Programme
The 101 triage project is one element of a wider £16.5 million programme to modernise public contact with policing. Other planned improvements include AI transcription for 999 and 101 calls, along with tools that can connect crime reports and highlight patterns in demand.
This suggests that the bigger change is not a single AI police call centre application. Control rooms are becoming places where software can support transcription, sorting, searching, and analysis around the work of human teams.
Used well, this could reduce administration and help staff find relevant information more quickly. It may also create a clearer record of conversations and make it easier to identify recurring problems across areas or time periods.
Used carelessly, the same technology could make mistakes faster and spread them further. Automated transcripts can contain errors. A categorisation system can inherit bias from its training data. A pattern-finding tool can encourage people to see a connection that has not been properly established.
Human oversight is therefore needed at every stage, not only at the point where the original call is answered.
What Could Go Wrong?
The central risk is misclassification. A caller who wants to report a serious matter may use language that sounds routine. Someone in distress may not explain the situation in a neat, complete sentence. A person calling on behalf of someone else may have limited information.
If the AI misunderstands that call and sends it elsewhere, the consequences can be much more serious than a longer wait. This is why an AI police call centre should have conservative thresholds for escalation, especially when a situation involves violence, abuse, a missing person, immediate danger, or vulnerability.
The system also needs strong data governance. Organisations should understand what audio, transcripts, categories, and referral records are stored, who can access them, how long they are retained, and how errors can be corrected. Calls to public services can contain personal and highly sensitive information, so convenience cannot be separated from privacy and security.
Regular independent testing would help assess whether the system works equally well for different accents, speech patterns, disabilities, ages, and levels of distress. Performance should be measured across real-world groups rather than relying on an overall average.
The Lesson for Businesses
Businesses receive their own version of the same demand problem. A support line may handle sales enquiries, technical faults, account questions, delivery issues, and urgent complaints through one number. Employees can spend much of their day deciding where each request belongs before they can start solving it.
AI can help with that initial sorting. It could identify the topic of an email, call, or support ticket and suggest the right team. Straightforward requests might be answered automatically, while complex cases are sent to a specialist with the relevant context already attached.
The police example points to several sensible boundaries. The categories need to be clear, the route to a person must be obvious, and the system should escalate when it is uncertain. Automation should reduce repetitive work around skilled employees rather than pretend that every customer interaction can be handled without judgement.
Before introducing an AI call centre or triage tool, a business should test it with real examples, measure errors as well as speed, and decide which subjects must always be reviewed by a person. It should also tell customers how their information is being processed and provide a straightforward alternative when the automated route is not suitable.
A Practical Future for Calling 101
The use of AI in calling 101 is a sign that public services are beginning to apply automation to the parts of a conversation that are structured and repetitive. If the system can keep clearly misdirected calls away from busy control rooms while protecting access to human help, it could make a real difference.
The important measure will not be how futuristic the technology sounds. It will be whether people with genuine concerns get the right help more quickly, whether staff have more time for difficult cases, and whether the public can trust the process.
For organisations considering similar ways to improve customer contact, our AI Consultancy page is a useful place to start.