Inspiration

Ambulance dispatch-to-on-scene times vary wildly across city zones, but the 911 center only looks at monthly averages. An average can look fine while one patient waits far too long.

Take ten response times: 4, 5, 6, 7, 8, 9, 10, 11, 12, and 20 minutes. The mean is 9.2 minutes, close to a 9-minute target. But only 6 of the 10 calls were on time, and one patient waited 20 minutes.

We wanted a tool that makes that patient visible: where it happens, why, and since when.

What it does

ERGent is a Shiny for Python dashboard for the 911 EMS dispatch center of Ithaca, NY, with two roles.

The dispatcher: in real time

Logs each call: The five timestamps—received, dispatched, en route, on scene, and cleared—are recorded through button presses, so nobody types a time.
Tracks running clocks: Clocks turn amber at 80% of a target and red past it.
Captures reasons for delays: A reason code is required only when a stage runs late (poka-yoke).
Shows a live map: Each Ithaca zone is colored by its worst open call.
Reviews shift history: A History tab shows the current and previous shift as a box chart by zone, with suggestions such as “Dispatch now,” “Rebalance,” or “Route.”
Demonstrates the workflow: A “Simulate a shift” replay shows all of this in a minute.

The supervisor: at the start of every shift

Zone health: An interactive map of Ithaca’s six zones, colored green, amber, or red. Hover over a zone for the reasons; click it to drill in.
Control charts: A daily X̄-S control chart per zone on log response times, with Western Electric rules, plus a chart of every single call.
Patient-centered metrics: 90th and 95th percentiles and percentage on time, instead of only an average.
Delay by stage: Is the delay in dispatch processing, the crew getting out the door (turnout), or the drive (travel)?
Zone comparison and capability: One-way ANOVA on log times, with Ppk translated into statements such as “About 89 in 100 calls within 9 minutes.”
Actions and follow-up: Rule-based suggestions with an impact estimate, an action log, a before/after Welch test to check whether a fix worked, and slip-back alerts if the zone gets worse again.

The tool flags and suggests. Qualified staff decide.

How we built it

Six Sigma first: We wrote the Define and Measure work—charter, CTQ tree, SIPOC, process map, and data-quality plan—before any code. Targets follow NFPA 1710: 9 minutes for life-threatening calls at the 90th percentile, a 64-second processing target, and a 60-second