Cloud

Richmond Reclaims ‘Precious Seconds’ in 911 Calls With Amazon Connect

The Virginia city has reduced emergency call wait times, saved money and lowered the burden on human dispatchers.

Twitter

Calvin Hennick is a freelance journalist who specializes in business and technology writing. He is a contributor to the CDW family of technology magazines.

When citizens in Richmond, Va., call either 911 or the non-emergency police number, they’re directed to the city’s emergency communications center. Emergency calls are prioritized, but with more than 8,500 total calls coming in each week, dispatchers are forced to balance calls about parking tickets and potholes with those about car crashes and crimes in progress.

“Our call takers have a very stressful job,” says Thomas Gann, technology manager for the Richmond Department of Emergency Communications, Preparedness and Response. “They do a fantastic job of getting the right help to callers at the right time. But that goes awry when a call taker is already on the phone assisting a citizen with a non-emergency, and 911 rings.”

Last year, Gann saw an opportunity to use artificial intelligence to reduce the burden of routine inquiries on dispatchers, freeing up more capacity for emergency calls. In July 2025, Richmond went live with Amazon Connect, a cloud-based contact center platform built on Amazon Web Services (AWS) and deployed in collaboration with CDW Government. The system uses generative AI to greet non-emergency callers, answer questions about city services and route calls to the right agencies — ideally, without sending calls to a human dispatcher.

“My goal is to have AI handle as much as possible, so that many of these calls never hit the emergency center,” Gann says. “As AI engines get better at fielding these calls, they’re getting even better at reducing that workload. It’s a complete rethinking of how we provide the best customer service to our citizens.”

Click the banner below to lean more about modernizing collaboration strategies.

xs_collaboration_animated_q424_showup_desktop

xs_collaboration_animated_q424_showup_mobile

Building a New System Focused on Training and Deployment

Gann and his team worked through a number of build sessions with CDW Government over the course of three months before going live. Since then, Gann says, CDW Government has built three different dashboards at Richmond’s request. “They’re so good about working with us to get what we want,” Gann says. “Now, I’ve got graphs and charts, and I can drill down to find out what the AI agent thinks of the call, and whether or not the caller was satisfied.”

To train the system, the team fed Amazon Connect all of the content on the city’s website, including every department page, every agency description, every FAQ list and even the city’s staff directory. The team also provided a curated list of common caller questions. The result is a system that knows who to call about a broken traffic signal, which department handles street repairs and how a caller can pay a parking ticket.

“Most of our callers don’t know what they need or how to get it,” Gann says. “The bot can provide that information immediately. It can provide phone numbers and websites, and it can either text that information to callers or read it to them.”

The new system started with some growing pains. Originally, Gann wanted the bot to simply ask, “How may I help you?” and allow citizens to explain their questions in natural language.

But Gann says that approach was “not ready for prime time” at deployment. In some cases, the bot scraped information from the wrong city department, or even hung up on callers. Within two days, the city switched to a model in which the bot gives callers a list of options, along with an invitation to ask their questions at any time. The team also put safeguards in place to make sure the system would always transfer calls to dispatchers if the bot was stumped, rather than disconnecting.

The tweaks paid off, both in lower call volumes for dispatchers and in a much more important metric: how quickly the emergency communications center picks up 911 calls.

How Richmond Measures the Benefits of the 911 Upgrade

The city is currently working with CDW Government and AWS to set up a new development environment to improve the system even more, in the hope of eventually shifting to a more open-ended model as the technology catches up with Gann’s original vision.

So far, Amazon Connect has been able to handle around 20% of citizens’ non-emergency calls autonomously. The system has also diverted another 6% of calls directly to agencies, rather than routing them through the emergency communications center. Additionally, Gann says, Amazon Connect is reducing the burden on dispatchers.

Another bonus: Richmond has spent “nowhere close” to the money the city originally budgeted for the system. “It’s not costing as much as our budgeted amount,” Gann says. “And even our budgeted amount was still far less than what it costs for humans to answer those calls.”

But to Gann, the most important outcome is the city’s improvement in 911 answer times. January 2026 statistics revealed that dispatchers answered 92.9% of 911 calls in under 15 seconds, compared with 84.9% in the same month of the previous year.

“Seconds are precious in 911,” Gann says. “If we can reduce our call answer times and get people help faster, I wouldn’t care if this cost five times as much as it does. It’s worth it.”

Brought to you by: