What problem does it solve?
Local government runs dozens of services, and residents contact it about all of them through the same few channels: a 311 or general number and a website. Most contacts are routine (collection days, opening hours, a pothole, a noisy neighbour), but each one needs a person to listen, find the right department, type the location and create a case. At the same time police and 911 centres spend dispatcher time on non emergency calls to their ten digit lines, pulling attention from real emergencies.
Menus and web forms do not fix this: residents do not know which department owns their problem, and forms lose the detail (exact location, photos) that crews need.
How does it work?
- Listen and classify. The agent asks what the resident needs and classifies it into the city's service catalogue, or detects that it is actually an emergency and transfers immediately.
- Answer information questions. Collection days, bylaws or opening hours are answered from the city's content and data, including address specific answers from GIS.
- Capture the request. For a service request, the agent collects the location (address, map pin or photo), description and contact details, and checks for duplicates nearby.
- Create and route the case. It creates the case in the work order or CRM system with the right category and priority and tells the resident the reference number.
- Keep the resident updated. Status updates go back on the same channel until the case closes.
- Hand over. Complex, sensitive or vulnerable cases go to a contact centre agent with the conversation attached.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Phone and voice, Web chat, WhatsApp, Mobile app, SMS
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Interactions handled | Not pooled | 20,000 to 30,000 | 2 | 2 vendor |
| Accuracy | Too few to pool | 80% | 1 | 1 vendor |
| Contact deflection | Too few to pool | 73% | 1 | 1 vendor |
| Customer satisfaction | Too few to pool | 50% | 1 | 1 vendor |
Value drivers: Lower cost to serve, Customer experience, Speed and cycle time, Inclusion and access.
Indicative value
A city of 500,000 residents with 400,000 contacts a year to its 311 service
USD 200,000 to USD 980,000
Agent handled contact cost avoided per year
How this is calculated
Formula: contacts * automatable * containment * costPerContact. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| 311 contacts per year across phone, chat and messaging contacts, contacts per year | 400,000 | 400,000 | The reference city. |
| Share of contacts that are routine questions or simple service requests automatable, fraction of contacts | 0.5 | 0.7 | Editorial assumption. Prepared's Galt case study says more than 73% of calls to that police department are non emergency, which is a different mix; replace with your own data. |
| Share of those the agent completes without an agent containment, fraction of automatable contacts | 0.25 | 0.5 | Editorial assumption, conservative against the contact deflection benchmark on this page. |
| Cost of an agent handled contact costPerContact, USD per contact | 4 | 7 | Editorial assumption. Replace with your own fully loaded cost. |
What it leaves out: Gross contact cost only. It leaves out better case data for crews, fewer duplicate reports, dispatcher time protected on police lines and the cost of integration and running the agent.
Who already uses it?
7 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Newcastle City Council
United Kingdom · Government and public sector · 2026
Newcastle City Council replaced legacy telephony with Amazon Connect, which routes resident calls and chats to the right team; Amazon Lex may also power menus for self service and triage. During live contacts, Amazon Q in Connect suggests approved knowledge and responses to agents; Contact Lens transcribes calls and surfaces topics and quality signals for supervisors. The council states that all outputs are advisory and do not decide eligibility, enforcement or case outcomes.
No outcome disclosed.
London Borough of Barnet
United Kingdom · Government and public sector · 2025
Barnet Council piloted Ami on its website to signpost residents 24/7 to the right page or service in a few areas: council tax, housing benefits, waste and recycling, schools and pest control. Ami only uses council approved content, can move the resident straight to the relevant page, watches for signs that a resident needs more support and, in office hours, connects them to an agent in the Amazon Connect contact centre that Capita runs for the council. The pilot expected about 30,000 chats over six months; no decisions are made about residents.
No outcome disclosed.
Abu Dhabi Government (TAMM)
United Arab Emirates · Government and public sector · 2025
TAMM is Abu Dhabi's single platform for about 950 government services from many entities, from car registration and visa renewals to traffic fines. The platform, including its AI assistant, is powered by Azure OpenAI Service and G42 Compass 2.0 (which also gives access to the Arabic JAIS model). The assistant answers questions about processes, shows the status of a user's requests and speaks several languages. A photo reporting feature lets residents photograph a problem such as a pothole or a broken traffic light; the assistant helps fill in the report and updates the reporter on the repair.
No outcome disclosed.
Rio de Janeiro City Data Office (Escritório de Dados)
Brazil · Government and public sector · 2025
Rio de Janeiro's City Data Office (Escritório de Dados) uses Dialogflow to run the chatbot of 1746, the city's citizen service, which handles urban maintenance requests and municipal enquiries, the kind of work a 311 line does elsewhere. Google reports that it cut the citizen response time from 30 minutes to 5 minutes. The only public source found is a one paragraph entry in Google's customer list.
- Interactions handled: at least 30,000, conversations per month
"The conversational AI reduces citizen response time from 30 minutes to 5 minutes across more than 30,000 monthly conversations."
Claimed by: vendor
Galt Police Department
United States · Government and public sector · 2024
Galt Police Department serves 26,000 residents with eight dispatch staff, who handle nearly 30,000 calls a year, more than 73% of them non emergency. Since 2024 an AI agent from Prepared answers the ten digit non emergency line, works out what the caller needs, resolves it or routes it to the right resource, and transfers any genuine emergency immediately. On 911 calls, dispatchers get a live transcript, an AI summary and key details highlighted as the call happens. During a shooting in Galt, the agent handled the incoming non emergency calls in the background while dispatchers managed the response.
- Contact deflection: 73%, share of call volume handled before reaching a dispatcher
"With 73% of call volume now handled before it reaches a dispatcher's headset, the calls that do come through are the ones that genuinely need a human."
Claimed by: vendor
Montgomery County Government
United States · Government and public sector · 2024
Montgomery County first launched Monty to relieve its 311 hotline during the pandemic, with 20 topics, and retired it when demand fell. Monty 2.0, built with Zammo.ai on Azure OpenAI Service and Azure AI Search, answers questions on more than 3,000 topics, with automatic translation into 140 languages, from the county's own knowledge base, and uses the county's geographic data to give address specific answers such as trash pickup days. It went through a seven month beta with a constituent focus group before the full launch in late 2024.
- Interactions handled: at least 20,000, since the beta deployment
"Since its beta deployment, Monty 2.0 has facilitated more than 20,000 constituent conversations, achieving a 50% customer satisfaction rate and reducing unanswered queries from 35%–45% to just 10%–15%."
Claimed by: vendor - Customer satisfaction: 50%, since the beta deployment
"Since its beta deployment, Monty 2.0 has facilitated more than 20,000 constituent conversations, achieving a 50% customer satisfaction rate and reducing unanswered queries from 35%–45% to just 10%–15%."
Claimed by: vendor
City of Kelowna
Canada · Government and public sector · 2023
Kelowna, a city of nearly 150,000 in British Columbia, uses Zammo.ai on Azure to answer calls and chats to its 311 non emergency line about property taxes, landfill rules, utilities and snowplowing, the last using live GPS data from the plows. With a provincial housing grant it also built a quick reference tool that returns the zoning bylaws and documents for a given address, and plans to extend it so the assistant walks applicants through the permit process up to the point where a city planner takes over. The city is creating an online registry of the data sources the system uses.
- Accuracy: 80%, snowplow schedule calls
"In 80 percent of cases, AI has already proven that it can deliver correct responses to people who call in about the snow."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- The city's service request catalogue with categories, owners and priorities
- Municipal content and address level data (collection schedules, zoning, service areas)
- Historic 311 cases to train and test classification
Systems to integrate
- 311 CRM or work order system (case creation and status)
- GIS and address lookup
- Telephony, web chat and messaging channels
- Contact centre platform for handover and emergency transfer
Complexity: Medium
Answering questions is simple; creating good cases needs integration with the work order or CRM system, a clean service catalogue, location capture and duplicate checks, and a safe transfer path for emergencies.
- 1
Clean the service catalogue first
Agree the categories, required fields and owning team for each request type; the agent can only route as well as the catalogue allows.
- 2
Start with the top ten requests
Launch with the highest volume information questions and two or three simple request types, such as missed collections and potholes.
- 3
Build the emergency exit
Define phrases and signals that trigger an immediate transfer to 911 or the dispatcher, and test them with real transcripts, as Galt's agent transfers genuine emergencies at once.
- 4
Capture location well
Use address validation, map pins or photos (TAMM's assistant helps fill in a report from a photo) so crews can find the problem the first time.
- 5
Close the loop
Send status updates and closure notices on the channel the resident used, and measure repeat reports.
Guardrails
- Immediate transfer on any sign of emergency, with a tested phrase list and a low threshold
- Case creation only in defined categories with required fields validated
- Answers only from city content and data, with refusal outside scope
- Personal data minimised and masked in logs
- Clear AI disclosure and a way to reach a person during office hours
KPIs to instrument
- Containment per request type and share of calls transferred as emergencies
- Routing accuracy (cases moved to another department after creation)
- Share of cases with a valid location and complete required fields
- Time from report to case creation and to resolution
- Duplicate reports and repeat contacts on the same issue
Human in the loop
Contact centre agents take handovers, complaints and sensitive cases; supervisors review samples of contained conversations and misrouted cases every week; department owners approve changes to categories and priorities.
Common failure modes
- A missed emergency
- A caller on a non emergency line describes an emergency in vague words. Keep the transfer threshold low and review every transferred and non transferred call with emergency keywords.
- Wrong department, lost case
- Misrouted cases bounce between teams. Measure reassignment and fix the catalogue.
- Accuracy only where content exists
- Barnet's transparency record gives the Ami model an average of 90% correct answers, but only for queries it has content for (a model figure, not a measured pilot result); questions outside that content fail. Track unanswered questions, not only accuracy.
- Reports without follow up
- Residents report, hear nothing and call again. Send status updates.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A 311 assistant must disclose that it is AI (Article 50). It is not high risk while it only informs and creates service cases. If it evaluates or classifies emergency calls or sets dispatch priority for police, fire or medical services, it falls under Annex III point 5(d) and becomes high risk.
Rules that apply
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Residents must be informed that they are interacting with an AI system.
- Annex III, high risk AI systems referred to in Article 6(2) (European Union, Europe). Point 5(d) covers classification of emergency calls and dispatch priority, the boundary a non emergency line must not cross silently.
- Algorithmic Transparency Recording Standard Hub (Government Digital Service, Europe). Recommended for local government in the UK; Barnet and Newcastle councils publish records.
Controls to put in place
- AI disclosure and a published description of what the agent can and cannot do
- Tested emergency transfer path, reviewed after every change
- Access control and retention limits on conversation logs and location data
- Weekly review of misrouted cases with department owners
- Accessibility testing for voice and chat channels
When it went wrong elsewhere
- NYC's AI chatbot tells businesses to break the law. In tests by The Markup, New York City's business chatbot told landlords and business owners that illegal practices, such as refusing housing vouchers or taking workers' tips, were allowed; the same risk applies to municipal information on 311 channels.
Frequently asked questions
- How much of a non emergency line can AI handle?
- It varies by call mix. Prepared, the vendor, reports that 73% of Galt Police Department's call volume is now handled before it reaches a dispatcher, in a department where more than 73% of calls are non emergency. Microsoft reports that Kelowna's assistant answers 80% of snowplow calls correctly, so measure by request type rather than for the line as a whole.
- Does AI on a non emergency line fall under the EU AI Act high risk rules?
- Not while it only answers questions and creates service cases; the Article 50 duty to disclose that residents are talking to AI still applies. It becomes high risk under Annex III point 5(d) if it evaluates or classifies emergency calls or sets dispatch priority for emergency services, so keep emergency detection as a simple transfer to a human dispatcher who decides, and document that design choice.
- What volumes do city assistants handle?
- Google reports more than 30,000 conversations a month on Rio de Janeiro's 1746 chatbot, and Microsoft reports more than 20,000 conversations for Montgomery County's Monty 2.0 since its beta. Start small and scale with the catalogue.
How to cite this page
Blits.ai AI Use Case Library, "AI for non emergency service requests and 311 routing", last verified 26 September 2026, https://www.blits.ai/ai-use-cases/non-emergency-service-request-routing. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published