The 9-1-1 Call Is Over. The Work Isn't.

How agentic AI is transforming one of the most overlooked workloads inside the Emergency Communications Center — from days of manual research to minutes of intelligent, human-reviewed results.

‍

After the Incident Ends, the Work Begins

When people think about ECC workload, they think about what happens when the phone rings. But a second, equally demanding workload begins after the incident is over.

‍Prosecutors & Investigators Need 9-1-1 calls, radio traffic, and CAD events tied to arrests or complaints.

Media & Attorneys Request audio, timestamps, dispatch records, and radio transmissions across incidents.

Citizens & Agencies Submit information requests describing events — often without incident numbers.

‍

One Request. Many Systems.

Information is scattered across logging recorders, CAD, radio systems, transcripts, notes, email, and multiple talkgroups. Every request becomes a scavenger hunt.

"Provide the 9-1-1 calls and radio traffic for the rollover near Main and 5th around 5 p.m. last Tuesday." — No CAD number. No case number. Approximate time. A person must interpret, locate, retrieve, organize, and redact.

‍

Agencies Already Charge for This Labor

Weld County, CO charges for cumulative staff time to retrieve, redact, and disseminate dispatch records.

Oregon State Police distinguishes requests requiring more than one hour of research or audio redaction.

‍

An Industry That Cannot Afford Another Manual Workload

82% — Hiring Difficulty. Of 9-1-1 centers able to hire were having difficulty filling vacancies (NENA, 2024).

600+ — Requests Analyzed. Information requests across the first five DISCOVERY customer sites.

7min — Median Close Time. Median time to close an information request with DISCOVERY.

84.9% — Closed Within 1 Hour. Of requests closed within one hour using AI-assisted research.

‍

APCO continues to identify ECC staffing shortages as a major industry challenge. The question is becoming obvious: Why are experienced professionals spending hours on work software could do in minutes?

‍

Not Search. Agentic AI.

Traditional Search waits for someone to tell it exactly what to look for. Requires the human to already know which system, which talkgroup, which timeframe.

‍CommsCoach DISCOVERY receives the request itself — in natural language — and deploys specialized AI agents that work together through the same research process an experienced person would perform.

Think of it as a digital research team: multiple agents working in parallel across CAD, 9-1-1 audio, radio traffic, transcripts, and talkgroups — then correlating results before a human ever opens a file.

‍

A Digital Research Team, Working in Parallel

Intake Agent interprets the request. Extracts dates, times, addresses, names, event numbers, and incident descriptions.

‍Research Agents work in parallel across CAD events, 9-1-1 audio, radio traffic, talkgroups, units, and transcripts.

‍Correlation Agent connects results. Validates whether calls, CAD events, talkgroups, and timestamps support the match.

‍Prepared Package assembles responsive material with confidence scores and plain-language explanations for human review.

‍

Confidence Matters as Much as Speed

Every potential match receives a confidence factor with a plain-language explanation of why the AI believes the information is responsive. The system surfaces uncertainty rather than hiding it.

‍High Confidence address, incident time, ANI/ALI data, and CAD association all correspond with the request. Minimal human review required.

‍Lower Confidence radio traffic occurred during the correct time period and involved a responding unit, but its relationship to the incident is less direct. Human judgment is flagged.

‍

Even the AI Should Have Someone Checking Its Work

1. Agents: Finds and Prepares 4 Independent AI agents find all data that matches, assigns a confidence score, assembles and redacts.
2. AI Judge: Independently Reviews A separate agent challenges the first four: Was all information found? Is the redaction overbroad? Was anything missed?
3. Human: Retains Final Authority The agency controls approval and release. AI performs the research and people make the decisions.

This human-in-the-loop philosophy aligns with NENA's Ethical Use of AI in 9-1-1 working group, which identifies human oversight and transparency as critical safeguards for responsible AI adoption.

‍

Automate the Work. Preserve the Judgment.

An information request has six basic steps. The opportunity for AI is to take responsibility for the first four — moving people to the part of the process where they add the most value.

‍Receive — AI-Owned 

Research — AI-Owned 

Gather — AI-Owned 

Prepare — AI-Owned 

Review — Human-Owned 

Release — Human-Owned

‍

Release requirements vary by state and local law. AI accelerates finding and preparing information — but should never pretend every disclosure decision is the same.

‍

What Happens When You Give Those Hours Back?

7min — Median Close Time. Across nearly 600 requests at the first five DISCOVERY customer sites.

84.9% — Closed Within 1 Hour. Requests resolved in under an hour — work that previously took days.

91.8% — Closed Within 24 Hours. Near-complete resolution within a single business day.

‍

The goal of AI in the ECC should not simply be doing something impressive. It should be giving time back to the people who never had enough of it in the first place.

‍