ICC telephony — 22 dispatcher commands collapsed to one (US patent)
Integrated 911 telephony with the Spillman Flex CAD across a CJIS-mandated network air-gap. Twenty-two manual dispatcher commands collapsed to a single action — pick up the phone, the system does the rest. Patented; ~5 seconds saved per call, ~10M phone minutes saved per year, across 12% of US 911 traffic.
- Dispatcher commands
- 22 → 1
- Saved per 911 call
- ~5s
- Phone minutes saved / yr
- ~10M
- of US 911 calls
- 12%
Problem
Integrating telephony with the Spillman Flex CAD required dispatchers to execute up to 22 sequential commands per call — accept, look up the caller, query state and federal databases, dispatch, hold, transfer, log, release. Every step was manual; every second was a measurable cost. In a mission-critical environment where five to six screens and two air-gapped PCs were already on the desk, the operator was paying that cost on every call, all shift.
When COVID hit, the Flex CAD was lagging on new features and competitors were closing in. The bet: leapfrog them by collapsing the call-taking workflow itself, rather than catching up feature-by-feature.
Approach
I started by riding along on 911 calls and running design-thinking sessions with first responders before writing a line of spec. The point of field work wasn’t to validate a hypothesis — it was to feel the cognitive load of running 22 commands while a caller is bleeding on the other end of the line. After that, the design choice made itself.
- Field work before spec.
Ride-alongs on live 911 calls + design-thinking workshops with dispatchers and call-takers. Mapped the end-to-end journey across the five-to-six-screen workstation. The 22-command sequence wasn’t the workflow — it was the symptom of two systems unable to talk to each other.
- Strip to single user value.
The biggest value for a 911 operator is the ability to pick up the phone. Everything else — caller location, gun permits, prior incidents, active alarms — is the system’s job, not the operator’s. Re-scoped the project to that one sentence and held the line on it through every review.
- WebSocket Secure as the air-gap bridge.
Hardware approaches and embedded UI approaches had both failed CJIS review. The breakthrough was treating telephony as a network service: WSS endpoints on the telephony side exposing six actions — accept, hold, pause, transfer, release, dial — that the CAD side could call without sharing memory, disk, or a network segment.
- Auto-trigger the lookups, not the operator.
On call-accept, the CAD side fired parallel queries against local, state, and federal databases — caller location (~10 cm accuracy via NG911), gun licenses, alarm permits, past incidents. The dispatcher saw a populated incident screen the moment the line opened, instead of typing for 30 seconds.
- Two telephony platforms, one integration.
Wrote the integration against an abstracted action surface so the same CAD client worked with both leading telephony platforms (ECW and VESTA). Customers picked their telephony vendor; we shipped one CAD build.
I decided, okay, let’s just strip it down from everything. What is the biggest value for a 911 operator? Well, that is its ability to pick up the phone — and the system does the rest. That’s it.
What shipped
- Six-action telephony API over WSS.
Accept, hold, pause, transfer, release, dial — exposed as WebSocket Secure endpoints on the telephony side, callable from CAD without crossing the CJIS air-gap.
- Keyboard-first control strip in CAD.
A minimal embedded telephony strip in the incident screen, driven entirely by keyboard shortcuts so the operator never has to leave the keyboard during a call.
- Parallel database fan-out on accept.
Caller location, gun licenses, alarms, prior incidents, permits — fetched in parallel on the moment the call connects, displayed without a single dispatcher action.
- ECW + VESTA, one CAD build.
Same CAD client integrates with both leading 911 telephony platforms, removing a forced vendor decision from the customer’s buying motion.
- US patent.
The WSS-bridged, action-abstracted integration was filed as a US patent and granted — converting the engineering breakthrough into a defensible moat.
Outcome
Dispatcher load
22 commands → 1
Call-taking collapsed to a single action — accept the call. The remaining 21 commands either disappeared or moved into automated flows triggered by the accept event itself.
Time saved
~5 seconds per call
Five seconds compounds when you ride 240M 911 calls a year. Net: roughly 10 million phone minutes saved annually across the deployed footprint.
Reach
12% of US 911 traffic
Now part of the only end-to-end vertically integrated NG911 stack on the US market. Customer base grew 50%+ in the five years following launch.
Business
Most profitable year in 4 decades
The product line posted its strongest financial year in 40+ years on the back of this integration and the customer-base expansion it unlocked. Patent issued.
What I’d do differently
- Instrument the dispatcher keyboard from day one. Shipped without per-shortcut telemetry, which cost a full quarter of dashboard insight on which actions actually mattered most under pressure. That data would have shaped V2 priorities far more sharply than the customer interviews we ran instead.
- Lead with the patent narrative externally. The WSS-bridge approach was patent-worthy and defensible; we under-sold it in customer conversations and let competitors come closer than they should have. A clearer "why this can’t be copied" story belonged in every NG911 RFP we replied to.
- Push field work even earlier in the process funnel. Ride-alongs happened after the first hardware-integration attempt failed CJIS review. They should have happened before — that’s the cheapest place to learn that the problem isn’t the integration, it’s the workflow on top of it.
Related work
