Agentic Location Intelligence for Emergency Response

By Sandi C. Stroud, ENP · Published September 25, 2026

The address is only the beginning

The July 20, 2012 shooting at Aurora’s Century 16 theater exposes the difference between locating an incident and navigating a response. The city’s after-action report describes ambulance access obstructed by parked vehicles and curbs. Police knew of a southern approach, but access instructions were not passed to fire and EMS. The report also documents coordination failures; a better map alone would not have resolved them. Aurora after-action report, p. 18.

References such as the west parking lot, the south side, and a former business name carry local knowledge. An arriving mutual-aid crew may not share it. The operational question is whether those references can be resolved into usable, current context without adding another round of radio traffic.

Assemble the context at the call

Preplanning and site surveys remain valuable. The question is what can supplement them when a detailed operational plan is unavailable or incomplete.

Here, an agent means software that can query several approved data sources, compare what comes back, and organize the results for a person to review. It is a proposed support capability, not an autonomous incident commander.

Parcel boundaries and building footprints may help identify a lot or structure. Road attributes may describe access restrictions. Terrain can reveal changes in elevation. Points of interest and former tenant records may connect a remembered business name to a place. Imagery adds context, but its capture date matters: a recent image is not a live view.

Those are candidate sources. Their existence does not establish that they contain the right attributes, at the right accuracy, for a particular incident. A parcel boundary cannot establish a surface’s load capacity, and a mapped approach cannot establish that it is clear now.

The first test is whether suitable records were available for the site and time being examined. The second is whether a system can retrieve and present them within an operationally useful time. Neither should be assumed in advance.

Build on the mapping work already done

School mapping programs offer a place to investigate the relationship between a map deliverable and reusable operational data. Where a site has already been surveyed, the useful questions extend beyond whether a file opens in public safety software.

• What was verified on site, and when?
• Which features have stable identifiers and usable attributes?
• Who owns updates, and how are changes validated?
• Can authorized systems retrieve those records while preserving access controls and source information?

A dated inspection can strengthen confidence in a record. It does not guarantee that the record is current, machine-readable, or suitable for every operational use. The opportunity is to connect mapping, data stewardship, and system delivery so that existing work remains useful after the map is handed over.

Add live context only when it can be trusted

Radio transcription and unit-location feeds could eventually supplement reference data. A report of casualties and a reliable location for the reporting unit might help resolve an ambiguous phrase such as “over here.”

That depends on the actual equipment, integration, permissions, timing, and quality of both feeds. A radio may not provide an accessible or reliable position. A transcript can misidentify a place or speaker. A confidence score alone does not establish correctness.

For that reason, I would keep these feeds outside the initial reference-data pilot. They introduce a separate set of questions about identity, latency, uncertainty, and the handling of sensitive operational information.

What this means for agencies and vendors

For agencies, the starting point is the operational burden: what information do your people repeatedly have to recover during a call, and which sources could support it? Requirements should address data quality, human review, update responsibility, and measurable performance alongside the interface.

For vendors, the opportunity is to make existing data usable in a demanding workflow. That means showing which records support an answer, where the answer stops, and how performance changes when data is missing, stale, or contradictory.

Ground Truth Advisors brings the public safety and location-data perspective to that conversation. My experience includes leading a statewide 9-1-1 program, building GIS validation and aggregation programs, and contributing to standards and procurement work. The aim is to connect a credible technical capability to a real agency need.

Three steps toward a practical pilot

1. Start with incidents that have published findings.

   Use after-action reports to identify information needs and test cases. Match the data to the incident date where possible; today’s records cannot establish what a system could have known then. Reports provide evidence for evaluation, not a complete operational ground truth.

2. Measure supported answers and appropriate refusals.

   Assess correctness, retrieval time, source coverage, and whether the system declines unsupported conclusions. Include missing records, conflicting data, and stale information. Have practitioners evaluate whether the output is usable.

3. Publish the method and its limits.

   Share the evaluation approach with agencies, vendors, and standards communities. A useful result includes what failed, what required human verification, and which data improvements would change the outcome.

The question is not simply whether an agent can return an answer. It is whether it can return a defensible answer quickly enough to support the people responsible for the response—and make the limits unmistakable.

Sources & exhibit notes

1. Aurora Century 16 Theater Shooting After Action Report. TriData for the City of Aurora, April 2014. Printed page 18 discusses ambulance access and interagency coordination.
2. Critical Incident Review: Active Shooter at Robb Elementary School. U.S. Department of Justice, January 2024. See the emergency medical response findings. DOJ review resource page.
3. Original business-case exhibit. Sandi Stroud, ENP, September 11, 2026. Source of the supplied recording and selected transcript. This website edition revises the framing and distinguishes proposals from demonstrated results.

The map in the title panel is decorative fictional geography, not a map of either incident. The workflow illustrates a proposed approach, not a deployed capability.

← All insights