Anoka County Tests AI to Filter Non-Emergency 911 Calls

Anoka County is piloting an artificial intelligence system designed to intercept and triage non-emergency calls before they reach human dispatchers. The program, currently in testing, represents a $60,000 investment aimed at addressing a chronic bottleneck in emergency communications infrastructure.

The math is stark. Call-takers like Samantha Gust handle up to 150 calls during a 12-hour shift, with approximately two-thirds classified as non-emergencies. That translates to roughly 733 non-urgent calls daily flooding the system. Noise complaints, barking dogs, and fireworks reports consume line capacity that could otherwise route life-threatening emergencies.

Anoka County Emergency Communications Director Kari Morrissey outlined the objective clearly: "Have AI screen those non-emergency calls, so we can answer those 911 calls quicker." The county's official announcement confirms the system is available for public testing at a demo line, with full activation planned for mid-May. The county's CivicAlerts page provides the demo number and testing parameters.

During a test call, the AI responds with a synthetic voice: "Police and fire, this is your AI system, Erik. How may I help you?" When a caller describes a potential heart attack, the system recognizes the emergency keywords and transfers to a human agent. In the testing phase, calls disconnect after transfer, but the live system will route directly to a Public Safety telecommunicator with a transcript of the conversation.

The physical experience matters here. Callers hear brief clicks between prompts, wait for the AI to process speech, and then either receive automated routing or connect to a human. There's a distinct texture to this interaction—the synthetic voice, the pause for processing, the mechanical transfer. It's less seamless than a human conversation, but the trade-off is intentional.

KSTP reported that the county's goal is answering every 911 call in under 10 seconds. When dispatchers are tied up on non-emergency lines, that target becomes impossible. The AI system won't handle 911 calls directly, but it will filter the non-emergency overflow that currently clogs the queue. The 5 EYEWITNESS NEWS investigation documented the demo call and confirmed the transcript feature for emergency escalations.

This approach mirrors broader trends in public safety communications. Hennepin County faces similar pressures, with July Fourth fireworks complaints alone spiking call volumes to over 2,700 in a 24-hour period. CBS News Minnesota noted that Anoka County's system will be tested during the holiday weekend before full deployment, with officials estimating savings of two to three minutes per non-emergency call.

The technology isn't replacing humans—it's attempting to create breathing room. Gust noted the mental benefit: "It's good for us mentally as well, to be able to recover from some of those high-priority calls we're taking." Dispatchers dealing with trauma, violence, and medical crises need cognitive bandwidth. (Nobody wants to explain to a caller why their barking dog complaint is taking priority over a cardiac arrest.)

Cost-benefit analysis remains murky. The $60,000 investment covers the AI program, but ongoing maintenance, updates, and potential liability for misrouted calls aren't detailed in public documents. If the system fails to recognize an emergency keyword, or if callers describe situations in unexpected ways, the consequences could be severe.

Privacy questions also surface. The AI transcribes conversations and passes them to human agents. What happens to that data? How long is it stored? Who has access? The county's announcement doesn't address these concerns, focusing instead on operational efficiency.

Implementation timing is aggressive. Mid-May activation leaves limited time for stress-testing edge cases. What happens when a caller speaks with an accent the AI doesn't recognize? What if background noise interferes with speech recognition? These aren't theoretical problems—voice AI systems have documented failure modes in noisy environments or with non-standard speech patterns.

The non-emergency number for Anoka County is 763-427-1212. Residents are encouraged to use this line for non-urgent matters, though the AI system will handle initial screening. Minneapolis residents have a 311 option for noise complaints, but Anoka County lacks that infrastructure.

Whether this system actually reduces 911 response times depends on execution. The AI must accurately distinguish emergencies from non-emergencies, transfer calls without delay, and provide useful transcripts to human agents. Any failure in this chain could negate the time savings.

Public testing invites community feedback, but the demo line disconnects after transfers. Real-world performance under actual call volume remains unproven. The county will know more after mid-May, when the system goes live and faces the full weight of daily call traffic.

For now, the technology represents a pragmatic attempt to solve a resource allocation problem. Whether it succeeds without introducing new risks is the question taxpayers will be asking in six months.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn