Emergency response

Police control rooms decide in seconds what a call is, how urgent it is and what to send. Under heavy demand, they have become an early target for automation.

Guide

What it is

  • Transcription automatic transcripts of calls, the most established and least contested use.
  • Triage support a suggested incident category and grade, offered to the call handler as a recommendation.
  • Demand forecasting predicting call volumes and patterns to plan shifts and position resources.
  • Automated non-emergency handling taking initial reports on non-emergency lines without a person.

How it is used

Forces use these tools to relieve pressure from high call volumes and staff shortages, while handlers keep the final grading on emergency calls. Inspectorate reports on call handling are the main external scrutiny. The record tracks control room systems and what is known of their performance.

Key tensions

Miscategorisation

The grade a call receives decides the response, and errors at first contact cannot be undone. A system keyed on what callers say risks under-grading calls where the caller cannot speak freely, such as domestic abuse.

The non-emergency boundary

Callers unsure how serious their situation is often choose the non-emergency line. If that route is automated, the way out of it to a person matters more than the automation itself.

Measuring the right thing

The useful question is whether a force measures grading accuracy by type of call, and would notice if one category got worse while overall figures held steady.

View all emergency response records Latest coverage

In depth

The full guide, about 3 minutes to read.

The control room as a decision point

Control rooms receive comparatively little public attention relative to their importance. The decisions made there, what a call is, how urgently to respond, what to send, shape almost everything that follows, and they are made in seconds with incomplete information by someone who cannot see the scene.

They also carry sustained demand pressure. Call volumes are high, a substantial proportion of contact is not police business at all but arrives because other services are unavailable, and staffing is a persistent difficulty. That combination is why control rooms have become an early target for automation: the pressure is real and the efficiency case is not manufactured.

What is actually deployed

Transcription is the most established application and the least contested. Automatic transcription of calls produces a searchable record and reduces the note-taking burden on a handler during a live call.

Triage support suggests an incident category and grading from the call content, presented to the handler as a recommendation rather than applied automatically. Related to this is surfacing relevant history, for example flagging that an address has previous incidents, which is genuinely useful and carries its own risk of anchoring a handler's judgement before they have heard the caller out.

Demand forecasting predicts call volumes and incident patterns to inform shift planning and resource positioning. This is the application closest to predictive policing, and inherits the same question about whether historic demand data reflects underlying need or historic response patterns.

Automated non-emergency handling takes initial reports on non-emergency lines without human involvement. This is where automation goes furthest, and where the boundary problem below is sharpest.

Miscategorisation is the central risk

Everything about control room automation comes back to one thing: the grading a call receives determines the response it gets, and getting that wrong at the first point of contact is not recoverable later in the way most errors are.

The risk is not random error, which is manageable, but systematic error against a particular kind of call. Consider calls where the caller cannot speak freely, because the person they are calling about is present. These are precisely the calls where accurate grading matters most, and precisely the calls least likely to contain the explicit language a system trained on typical calls would key on. A system that grades on content risks systematically under-grading exactly the contacts that most need urgency.

Domestic abuse is the clearest instance, and it is an area where control room grading has already been the subject of serious findings by inspectorates and inquiries independent of any technology. Automating a judgement that has a documented history of going wrong does not automatically make it worse, but it does risk making the failure consistent rather than occasional, and harder to see.

Where the boundary sits

The distinction between emergency and non-emergency contact carries a great deal of weight in how these systems are deployed, and it is less solid than it appears.

A caller does not always know which category their situation falls into. Someone uncertain whether their circumstances are serious enough may well choose the non-emergency route, and that hesitancy is common in exactly the situations where it is least warranted. If the non-emergency route is substantially automated and the emergency route is not, then the quality of first assessment depends on a judgement the caller made before anyone assessed anything.

This is not an argument against automating non-emergency contact, which relieves genuine pressure. It is an argument that the escalation path out of an automated flow matters more than the automated flow itself, and that it should be evaluated on how reliably it catches the calls that were placed in the wrong channel.

Where the rules sit

There is no dedicated framework for control room automation. Data protection law applies to call recording and processing, and automated decision-making provisions are relevant where a decision materially affects someone, which a grading decision plausibly does. Inspectorate reports on force call handling and response times are the most substantive external scrutiny in practice, and they assess outcomes rather than the systems producing them.

The most useful oversight question here is not whether a force uses automation in its control room, but whether it measures grading accuracy by call type, and whether it would notice if performance degraded for one category while overall figures held steady.

Records on the tracker

Open the full tracker

26 records on emergency response, most recently checked first: 23 in use now, 0 withdrawn and 3 on the horizon. Each is one police force or public body using one system.

  1. Italy2025

    Polizia Locale di Busto Arsizio — AI video analytics in the control room

    Lombardy (Varese)

    OperationalChecked
  2. United States

    Federal Bureau of Investigation — AI triage of public tips (TIPS)

    Nationwide; National Threat Operations Center, West Virginia (per the FBI chief AI officer)

    OperationalChecked
  3. United KingdomSalesforce (platform)2026

    Staffordshire Police — AI Emergency and Non-Emergency Call Triage

    Staffordshire

    ProposedChecked
  4. United Kingdom

    Merseyside Police — Bobbi AI virtual assistant (non-emergency enquiries)

    Merseyside

    OperationalChecked
  5. United KingdomTTEC Digital and AWS, with West Midlands Police (per Police Science magazine)2024

    West Midlands Police — Andi-Esra AI voice assistant for 101 calls

    West Midlands

    OperationalChecked
  6. United Kingdom2026

    South Yorkshire Police — Bobbi AI chatbot (non-emergency enquiries)

    South Yorkshire

    OperationalChecked
  7. United KingdomHome Office and NPCC, with Vodafone Business

    Nottinghamshire Police — AI 101 call routing (pilot)

    Nottinghamshire

    TrialChecked
  8. France

    Gendarmerie nationale — I-Accueil chatbot for brigade reception staff

    TrialChecked
  9. United StatesCarbyne (acquired by Axon, closed Q1 2026)2023

    Orleans Parish Communication District — AI emergency call triage

    New Orleans, Louisiana

    OperationalChecked
  10. United StatesCorti (Denmark)2023

    Seattle Fire Department — AI emergency medical call analysis

    Seattle, Washington

    OperationalChecked
  11. United KingdomSalesforce (Agentforce 360 platform), according to Salesforce2025

    Thames Valley Police — AI Emergency and Non-Emergency Call Triage

    Thames Valley

    OperationalChecked
  12. United KingdomSalesforce (Agentforce 360 platform), according to Salesforce2025

    Hampshire and Isle of Wight Constabulary — AI Emergency and Non-Emergency Call Triage

    Hampshire and Isle of Wight

    OperationalChecked
  13. United StatesAxon (Prepared)2026

    Colorado Springs Police Department — AI Emergency and Non-Emergency Call Triage

    Colorado

    OperationalChecked
  14. United StatesAxon (reported by the Daily Camera, 7 October 2026; the city page does not name the supplier)2026

    Boulder Police and Fire Communications — AI Emergency and Non-Emergency Call Triage

    Colorado

    TrialChecked
  15. United StatesAmazon (named as the technology provider by WTKR; product not named)2026

    Norfolk Police Department (Virginia) — AI Emergency and Non-Emergency Call Triage

    Virginia

    OperationalChecked
  16. United StatesHyper (Motorola Solutions)2026

    Lompoc Police Department — AI Emergency and Non-Emergency Call Triage

    California

    OperationalChecked
  17. United Kingdom

    Cheshire Constabulary — AI stalking-behaviour flagging tool

    Cheshire

    In developmentChecked
  18. United KingdomHome Office and NPCC, with Vodafone Business2026

    Cumbria Constabulary — AI 101 call routing

    Cumbria

    OperationalChecked
  19. United KingdomSalesforce (Agentforce platform)2025

    Humberside Police — Bobbi AI virtual assistant

    Humberside

    TrialChecked
  20. United Kingdom2026

    Cambridgeshire Constabulary — Citizen Assist AI online crime reporting

    Cambridgeshire

    TrialChecked
  21. United KingdomHome Office and NPCC with Vodafone Business2026

    Home Office — AI 101 call routing and triage

    England and Wales

    TrialChecked
  22. United StatesAxon (Prepared)2026

    Fort Worth Police Department — AI Emergency and Non-Emergency Call Triage

    Texas

    OperationalChecked
  23. United StatesAxon (Prepared)2026

    Antioch Police Department — AI Emergency and Non-Emergency Call Triage

    California

    ProposedChecked
  24. United KingdomHome Office and NPCC, with Vodafone Business2026

    South Yorkshire Police — AI Emergency and Non-Emergency Call Triage

    South Yorkshire

    OperationalChecked
  25. United StatesHyper (Motorola Solutions)2026

    Santa Maria Police Department — AI Emergency and Non-Emergency Call Triage

    California

    OperationalChecked
  26. United KingdomBuilt in-house with the NPCC Digital Public Contact programme2026

    West Yorkshire Police — Post Call Analysis

    West Yorkshire

    OperationalChecked

Latest coverage

All 75 stories