AI for Public Works Call Centers: Automating Pothole, Work Order, and Road Hazard Intake

The Public Works Call Center Problem

Most public works departments operate with a call center budget that was set in 1998 and a call volume that has tripled since. A single mid-sized city fields tens of thousands of resident-initiated reports every year — potholes, water main breaks, streetlight outages, sidewalk damage, graffiti, illegal dumping, downed signs, snow complaints — and the people answering the phones are usually two or three clerks who are also managing permit inquiries, contractor coordination, and internal staff requests at the same time.

The result is predictable. Calls go to voicemail. Voicemails pile up. Work orders get created in batches at 9 a.m. the next morning, which means a pothole reported at 6 p.m. Tuesday does not become a ticket until Wednesday — and does not get a crew assigned until Thursday. Residents call back. Some of them call their city council member instead. Now a clerk is answering the same pothole complaint three times across three different channels.

AI voice agents solve this the way they solve every other high-volume intake problem: by answering every call, capturing structured data on the first try, and writing that data directly into the system that dispatches crews.

How AI Handles a Public Works Call

The workflow is designed to match what a good human intake clerk does — just faster, 24 hours a day, and without hold queues.

1. The resident calls the public works line. The AI answers within one ring and identifies itself plainly: "You've reached the City of Example Public Works line. What can I help you report today?"
2. The AI classifies the request. Pothole? Water leak? Streetlight out? Downed sign? The AI recognizes hundreds of public works call types from natural language — "there's a huge hole on Main Street that almost swallowed my tire" is classified as a pothole report without the caller ever using the word "pothole."
3. Location is captured and verified. The AI asks for the address, nearest intersection, or landmark. It geocodes the location in real time and reads it back: "Just to confirm — that's the 400 block of West Elm Street, near the intersection with 5th Avenue. Is that right?"
4. Severity and safety are assessed. Is the pothole causing vehicle damage? Is the water leak flooding a street? Is the downed sign blocking traffic? These questions drive priority in the CMMS.
5. A work order is created in the CMMS. The AI calls the city's work order API, creates the ticket with all required fields populated, and receives a work order number back.
6. The resident gets a confirmation. "You're all set. Your work order number is 82-14-7731. You'll get a text update when a crew is assigned. Is there anything else you'd like to report?"

Total time: 60 to 120 seconds. Dropped calls: zero. Voicemails: zero. Next-morning ticket backlog: zero.

Call Types AI Handles End-to-End

Public works intake is one of the cleanest use cases for AI voice agents in local government because the call types are well-defined, the data fields are standardized, and the decisions are rule-based. Here are the categories AI handles without any human in the loop:

Pothole and Road Surface Reports

The single highest-volume public works call in most cities. AI captures the street, block, lane, approximate size, and whether the pothole has caused vehicle damage. Work order is created and routed to street maintenance with severity scoring.

Water Leaks, Main Breaks, and Hydrant Issues

A resident sees water bubbling up through the pavement or a leaking hydrant. AI captures the location, asks about the volume and duration of the leak, flags any reports of water outage or discoloration, and creates a priority work order to water operations. Emergency-level leaks (flooding a street, affecting a building) escalate to a live operator.

Streetlight and Traffic Signal Outages

AI records the pole number if available (residents surprisingly often read it off), the intersection, and the type of outage (dark, flickering, all-red flash). Traffic signal issues with safety implications are flagged for immediate crew dispatch; routine streetlight outages queue into the normal maintenance schedule.

Downed or Damaged Signs

Stop signs, yield signs, street name signs, speed limit signs. AI captures location and sign type. Stop signs automatically get a priority flag because of the safety implication.

Sidewalk, Curb, and ADA Issues

Cracked sidewalks, trip hazards, curb ramp damage, missing tree-well grates. AI captures location, measurements if the resident can provide them, and any ADA-specific implications.

Storm Drain and Flooding Complaints

Clogged catch basins, standing water, flooded intersections. AI captures location, flagging weather-related urgency when applicable, and routes to stormwater operations.

Graffiti, Illegal Dumping, and Debris

AI takes the report, asks if the property is public or private, and routes to code enforcement or public works cleanup crews depending on the city's workflow.

Snow and Ice Complaints

During winter weather events, snow complaint volume can overwhelm call centers in hours. AI absorbs the spike — capturing street, block, and specific issue (not plowed, icy, berm blocking driveway) — and aggregates the reports so operations can see which routes need attention first.

Tree Issues and Storm Damage

Fallen branches, leaning trees, storm-related tree damage. Urgent (blocking a road, on a car, on a house) escalates; routine (dead tree in park) queues into urban forestry.

Park and Facility Maintenance

Broken playground equipment, vandalism, damaged benches, unsafe field conditions. AI captures facility name, specific equipment, and severity.

How AI Captures an Accurate Location

The single biggest failure mode of paper-and-voicemail public works intake is bad location data. A crew gets dispatched to "the pothole on Main Street" and spends 45 minutes driving the corridor trying to find it. AI voice agents are specifically designed to fix that.

• Full address geocoding. If the caller provides a street number and name, the AI geocodes it in real time against the city's address authority file, confirms it is a valid location, and reads it back for verification.
• Intersection parsing. For callers who only know the nearest cross street, the AI parses "Main and 5th" or "Elm at Oak" and converts it into a coordinate.
• Landmark recognition. "Across from the library," "next to the elementary school," "behind the Kroger on Route 50" — the AI matches named landmarks against the city's POI database and resolves to a location.
• Block-level estimation. When the caller only knows "somewhere on the 400 block of West Elm," the AI creates the work order with a block-level geometry rather than a specific point, and flags it for field verification.
• SMS follow-up for GPS and photos. After the call, the AI can send the caller a text message with a secure link that captures GPS coordinates from their phone and lets them upload photos of the issue. Optional, opt-in, and massively improves location accuracy.
• Read-back confirmation every time. The AI never finalizes a work order without reading the location back to the caller and getting an explicit confirmation. This single step is what drives accuracy from ~85% (human intake) to 97%+ (AI intake).

CMMS and Work Order Integration

An AI voice agent that does not write work orders directly into your system is a glorified answering machine. The entire value proposition depends on clean, bi-directional integration with the city's CMMS or 311 platform.

BetaQuick's AI voice agents integrate with the major platforms public works departments actually use:

• Cityworks — AMS work orders and service requests created via REST API with full field mapping.
• Cartegraph OMS — work requests and inspections created directly, including asset linking.
• Lucity — work order creation with task and asset association.
• Accela — service requests created via Accela Civic Platform API.
• Maintenance Connection, Brightly (formerly Dude Solutions) — work order creation via documented APIs.
• Salesforce Public Sector Solutions — case creation with custom object mapping for 311 deployments.
• Custom / legacy systems — REST, SOAP, ODBC, or flat-file integration when that is what the city is running.

Integration goes both ways. The AI can also query the CMMS to answer status questions: "I reported a pothole last week, what's the status?" The AI looks up the work order by phone number or address and answers: "That work order is currently assigned to Street Maintenance Crew 3 and is scheduled for repair on Thursday, April 16."

What Agencies Are Measuring

Public works call center automation is easy to measure because the baselines are painfully obvious. Here are the metrics that show up in pilot reports most consistently:

Two numbers in that table matter more than the rest. The first is after-hours coverage — public works calls don't stop at 5 p.m., but staffing usually does. AI closes that gap completely. The second is duplicate work orders — when AI deduplicates incoming reports against existing open tickets ("I see we already have a pothole reported at that location — your report will be added as a confirmation"), crews stop driving to the same location twice.

How to Deploy AI in Your Public Works Call Center

Public works is one of the fastest government departments to deploy AI in, because the procurement, integration, and risk profile are all more forgiving than public safety. Here is the path most cities follow:

Step 1: Pull a Year of Call Type Data

Export the last 12 months of service requests from your CMMS, grouped by type. In almost every city, 8 to 12 call types represent 80 percent of volume. Those are your Phase 1 targets.

Step 2: Confirm CMMS API Access

Talk to your CMMS vendor or IT department. Confirm that the work order creation API is available, who owns the credentials, and whether there are any field-level validations the AI will need to satisfy.

Step 3: Write the Routing Rules

Which call types go to which crews? What severity levels escalate to a supervisor? What hours trigger different workflows? Document these rules — the AI is configured from this document, not from tribal knowledge.

Step 4: Use Cooperative Purchasing

Skip the RFP. Texas DIR contract DIR-CPO-6057 (held by BetaQuick), NASPO ValuePoint, and state-level cooperative vehicles all allow direct procurement for AI voice agents at pre-negotiated terms. This removes 3 to 9 months from the timeline.

Step 5: Pilot with One Call Type

Launch with pothole reports only. Run AI alongside your existing intake for 2 weeks. Audit every work order created for accuracy. Tune the prompts. Then add the next call type.

Step 6: Add Status Lookups and Notifications

Once intake is solid, layer on status lookup ("what's the status of my report?") and outbound notifications ("your pothole repair is scheduled for tomorrow"). This is where resident satisfaction scores start climbing fast.

Frequently Asked Questions