Cut 911 dispatch time with AI

You are a forward deployed engineer at an applied AI company. You are in a 30-minute discovery meeting with the director of emergency communications for Larkfield County (fictional, United States, population 1.1 million) and her operations lead.

What they tell you:

• The centre handles about 2,300 calls a day, with 14 call-taker consoles staffed per shift. Call-takers also dispatch for fire and EMS; police has separate dispatchers.
• Median call-to-dispatch time for priority-one calls is 94 seconds. The state standard they report against is 60 seconds.
• Staff vacancy is 22% and overtime is high. Roughly 30% of calls on the 911 line are not emergencies (noise complaints, "is the road closed", pocket dials).
• About 18% of callers prefer Spanish; a phone interpreter service is used and takes 40 to 70 seconds to join.
• They run a computer-aided dispatch (CAD) system from a large vendor, upgraded two years ago. Call audio is recorded.
• The county board has approved USD 1.2 million for "AI in the 911 centre" this fiscal year and wants a visible result in six months.

The director closes with: "Everyone is pitching us a voice bot. Tell us what you would actually build."

Write how you would break this problem down in the meeting and the week after it. Treat this as a live conversation: the interviewer will assess what you ask and how you frame success before what you propose.