Case Study 03 · Field-services operator

Enterprise Automated Lead Triage and Voice Dispatch Architecture

A voice and chat intake layer, an automated triage engine, and a dispatch integration that answers every inbound request, prioritizes it, and books it without a person in the loop.

Client
Field-services operator
Industry
HVAC and field services
System
Voice and chat intake, triage, dispatch
Discipline
Multi-agent systems and workflow pipelines
Headline result
42% less triage time

The challenge

What was breaking.

A growing field-services company was losing jobs it never knew existed. Calls rang out after hours, overflow calls during storms went to voicemail, and the ones that were answered landed on a dispatcher who had to figure out urgency, location, and technician availability by hand before anything was booked. Triage was the bottleneck: every request waited for a human to read it, classify it, and decide where it went.

The company needed every inbound request answered immediately, classified correctly, and routed to the right technician slot, with emergencies escalated to a person and routine jobs booked automatically. It also needed the office to trust the system, which meant every automated decision had to be explainable after the fact.

The solution

What we built.

InsightSol built a three-layer architecture. The intake layer answers calls and chats around the clock with a voice agent on Twilio and a web chat agent, transcribes and structures each request, and captures photos and location where relevant. The triage engine classifies intent and urgency with a model constrained by written rules, enriches the request from the CRM and service history, and assigns a priority with an explanation attached.

The dispatch layer turns a triaged request into a booked job. It queries technician availability, skills, and location, proposes slots, confirms with the customer by voice or text, writes the job to the dispatch system, and notifies the technician. Emergencies route to an on-call human with full context in seconds. Everything else is booked, logged, and auditable without office staff touching it.

Headline result

42%

reduction in operational triage time

Stack

PythonTwilioClaude APIOpenAI APIPostgreSQLRedisn8nTypeScriptNext.js

System architecture

The components and the path a request takes through them.

Components

  • Voice and chat intake

    Twilio telephony with streaming speech recognition, web chat agent, photo and location capture

  • Triage engine

    Intent and urgency classification under written rules, CRM enrichment, explainable priority assignment

  • Dispatch integration

    Availability, skills, and location queries; slot proposal; confirmation; job creation; technician notification

  • Escalation path

    Emergency detection routed to an on-call human with transcript and context attached

  • Orchestration and side effects

    n8n workflows for notifications, CRM sync, and follow-up sequences

  • Telemetry

    Per-request traces, triage accuracy tracking, and owner dashboard

Request flow

  1. 1Call or chat arrives and is answered immediately by the intake agent
  2. 2Request is transcribed, structured, and enriched from the CRM
  3. 3Triage engine assigns intent, urgency, and an explained priority
  4. 4Emergencies escalate to an on-call human; routine jobs continue
  5. 5Dispatch layer proposes slots, confirms with the customer, and books
  6. 6Technician is notified; job, transcript, and decisions are logged

Quantified results

What was measured in production.

  • 42 percent reduction in operational triage time, with requests classified and routed before a dispatcher opens them
  • 38 percent more jobs booked from the same call volume
  • 62 percent of routine calls handled end to end without office staff
  • 5-second average first response, around the clock, with emergencies escalated to a person in seconds

Engineering lessons

01

Explainable triage is what the office trusted

Every priority came with a written reason the dispatcher could read. That single design choice turned skeptical office staff into the system's strongest advocates within the first weeks.

02

Escalate emergencies in seconds, automate everything else completely

Half-automation would have kept a dispatcher watching the queue all day. Routing genuine emergencies to a person immediately is what made it safe to let routine jobs book themselves.

03

Measure triage time from first contact

The 42 percent reduction is measured from the moment a call or chat arrives, not from when a human opened the request. Measuring from the customer's side is the only number that reflects their experience.

Want the architecture behind your own numbers?

The audit produces the same diagrams and unit economics this system started with, for your workflow and your data.

  • Read-only access under NDA
  • Written review and diagrams
  • Fixed-price roadmap