Most mid-size law enforcement agencies have been buying technology for fifteen years. The specifications read like a catalog of two decades of procurement cycles: body-worn camera programs authorized after a critical incident. License plate reader networks funded through a state grant. A records management system selected by committee in a process that took eleven months. A computer-aided dispatch platform replaced once, then upgraded, then patched into a state CJIS bridge. A radio infrastructure refresh that arrived with new handhelds and a dispatch console that the communications unit is still learning. Surveillance camera coverage that grew one camera at a time across parks, government facilities, transit corridors, and schools.
Each of these systems was procured separately. Each was evaluated against its own RFP. Each was selected by a different committee in a different budget year under a different set of constraints. Each arrived with its own vendor relationship, its own training program, its own support contract, and its own user interface. Each solved a defined problem — a problem the chief or the commander or the city manager could point to when the purchase was approved. And each one did what it was bought to do.
None of them was designed to work with the others.
This is not a criticism of any individual decision. Each of those procurements was rational in its context. The LPR vendor was not expected to integrate with the RMS vendor. The body-camera contract was not scoped to include CAD bridging. The radio refresh was a communications project, not a data project. The surveillance buildout was a capital improvement line item, not an intelligence investment. Every system was purchased to solve a specific problem, and every system did. What no one purchased — because no vendor was selling it, and no RFP asked for it — was the thing that would make all of them work together.
What the data environment looks like from inside an agency
Walk through a typical operational day inside a mid-size agency and look only at what the technology produces, not what the officers do. In eight hours of shift activity, the CAD system generates thousands of events — dispatches, status changes, timestamps, geocodes, call classifications, unit assignments. Each event is a structured data record that lives inside the CAD. Most of it is never queried by anything outside the CAD.
In that same eight hours, the LPR network produces hundreds of thousands of plate reads. Every fixed camera at every intersection covered by the program reads every plate that passes. Every mobile unit equipped with an LPR reads every plate it passes on patrol. The reads are timestamped, geocoded, and stored. Most are stored and never looked at again. The ones that match a hot list get an alert. The rest sit in a database that, in most agencies, is not connected to the CAD, not connected to the RMS, and not searchable by anyone who is not an LPR analyst.
Body-worn cameras are recording continuously in pre-event buffer mode and actively whenever an officer activates the camera. A shift's worth of footage across a patrol division is measured in terabytes. It uploads to evidence management. It is catalogued by case number if an officer tagged it correctly. It is almost never transcribed. It is almost never searchable by content. The intelligence inside the video — what was said, who was seen, what happened in frame — lives as pixels in a file that will only be pulled if someone remembers that a specific officer was at a specific scene at a specific time.
The radio is talking the entire shift. Every transmission from every unit on every channel. Observations, descriptions, location updates, BOLO calls, requests for backup, outcomes of contacts. Radio traffic is the most information-dense real-time feed in the agency — and it is the least connected. It is captured on a recording log that almost no system in the stack is watching.
The RMS is receiving reports written after the fact. Surveillance footage is accumulating at whatever retention policy the agency can afford. And every one of these systems is operating in its own silo, producing its own data, answering its own questions, and speaking to no one.
The cost of disconnection
The cost of this disconnection is measured in what gets missed. Not in spectacular failures — agencies generally do not miss the big call. The cost is measured in the small, everyday intelligence that existed in the data but never reached the person who needed it.
Consider a composite scenario. A patrol unit is dispatched to a reported disturbance at a commercial address. The CAD holds a record of two prior calls at the same address in the past ninety days — one for a welfare check, one for a domestic dispute. The LPR network read a vehicle registered to a subject listed in the domestic report passing through an intersection six minutes ago, moving in the direction of the address. The radio channel has a transmission from two hours earlier — an officer relayed a plate he had stopped, a plate that also appears on the LPR read. None of these three systems is aware of the other two. The officer dispatched to the scene receives what CAD shows — the call type and the address. The rest of the intelligence existed in the data environment. None of it arrived before the officer did.
Consider a second. A stolen vehicle BOLO is active across the jurisdiction. A surveillance camera in a parking structure captures the vehicle entering at 14:12. The LPR at the same facility reads the plate at 14:12. The camera footage is not reviewed. The LPR read does not match against the BOLO list because the BOLO was entered into the RMS under a slightly different case number than the one the LPR hot list was built from. The vehicle leaves the structure at 16:40. Nobody in the operation ever knew it was there.
Consider a third. A person of interest is described on the radio during a suspicious-activity call — clothing, direction of travel, rough physical description. Forty minutes later, a body camera records the same description being given by a witness at an unrelated incident two blocks away. The description is identical. The connection was never made. The body-camera footage is not transcribed in real time. The radio transmission is not indexed against the body-camera content. The two systems live in different storage, with different access protocols, maintained by different units.
In each scenario, the information existed. The agency paid for the systems that produced it. The coverage was there. What was missing was the layer that could read across all of it simultaneously and surface the correlation while it still mattered.
Why ripping and replacing the stack doesn't solve it
The instinct, when confronted with the integration gap, is to imagine a unified platform — a single vendor providing CAD, RMS, LPR, evidence management, radio, and analytics in one product. Several companies have tried to sell this over the last decade. The strategy does not work, for reasons that are structural rather than technical.
Agencies do not procure on a clean schedule. The RMS was bought in 2017 with a ten-year amortization. The CAD upgrade is scheduled for the next biennium. The LPR grant cycle runs on a three-year refresh. The body-camera program is on a five-year renewal. At any given moment, any given agency has five to eight active vendor contracts with staggered renewal dates, and ripping and replacing the stack means paying termination fees on contracts that are not yet amortized, while simultaneously retraining every user on an entirely new set of interfaces during the transition. No chief has the budget for this and no city council will approve it.
And when agencies have tried replacement projects at scale, the outcome has consistently been that the new unified platform becomes an additional silo rather than a replacement for the existing ones. The old systems stay in place because they were integrated into every downstream process — court submissions, FOIA responses, grant reporting, officer evaluations — and the new unified platform gets deployed alongside them, used for the functions where it is better, and ignored for the ones where the legacy systems still do the job. The gap does not close. It widens.
What an intelligence layer actually does
The architecture that actually addresses the integration gap does not replace any of the existing systems. It sits above them. It reads from them. It does not ask agencies to change what they are already doing.
Concretely, an intelligence layer ingests data from each existing system through the system's own API — CAD events over the CAD vendor's integration interface, LPR reads from the LPR platform's data feed, RMS records from the records system's export endpoint, radio transmissions from the dispatch console or the PTT platform, body-camera metadata from evidence management, surveillance footage through RTSP streams or the VMS's event log. The existing systems continue to do what they do. The integration layer simply reads what they produce.
Above the ingestion layer, a set of agents reasons across the combined environment. An agent is not a dashboard query and not an alert rule. An agent is given a scope — what data sources it watches — and a responsibility — what kind of intelligence it is charged with surfacing. A dispatch agent watches CAD, LPR, and RMS simultaneously and surfaces prior history and active reads on any unit assignment. A surveillance agent watches camera feeds and known-offender registries and surfaces correlations between detected events and active cases. A communications agent watches radio transcription and open incidents and surfaces any transmission content that intersects an active situation.
The command view becomes the place where this correlation surfaces. Not as dashboards to be reviewed, but as attention directed by the agent layer. When something matters, it appears in the view. When nothing matters, the view is quiet. The intelligence is not produced by asking the right question at the right moment. It is produced continuously, by systems that do not sleep, across data sources that an individual human could not watch simultaneously.
The real question
The integration gap is not fundamentally a technology problem. The underlying technology — API ingestion, event processing, large language models reasoning over structured and unstructured data — exists and is already in production in other industries. The reason it has not been built for public safety is not technical difficulty. It is that the vendors currently in the stack have no commercial incentive to enable it. An LPR company does not make more money by feeding its reads into a rival's RMS. An RMS company does not benefit from surfacing data that bypasses its reporting interface. The silos exist because the business models that produced them reward keeping them in place.
The agencies that will operate differently in the next decade are the ones that recognize this, and procure the intelligence layer accordingly — not as a replacement for what they already have, but as the thing that finally makes what they already have work together. The question is not which vendor will unify the stack. The question is which agency will.