Case study · AI intake
Inspectrik.
A property inspection marketplace offering chat, video, and in-person help. The intake had to route a vague problem to the right support mode, structure the summary, and keep a human on every recommendation.
- Client
- Inspectrik
- Scope
- AI-assisted intake
- Status
- Ongoing
- Modes
- Chat · video · visit
Context
Homeowners, landlords, and buyers arrive with unstructured problems: a leak, a crack, a noise, a pending closing. The platform offers three ways to get help, each with different cost and turnaround. Choosing the wrong mode wastes a visit or delays an answer.
A generic chat assistant made this worse. It answered with confidence, missed the property context, and gave no clean handoff to an inspector.
The brief
- Route each request to chat, video, or in-person based on the issue.
- Collect the context an inspector needs before the session starts.
- Produce a structured summary, not a transcript.
- Keep a human on the recommendation and the final report.
What I did
Intake flow
Guided questions that frame the issue before a mode is suggested.
Triage logic
Rules and model output combined to select the right support mode.
Summary schema
Structured findings: issue, property context, urgency, next action.
Human review
Low-confidence cases and every recommendation route to a person.
What shipped
- AI-assisted intake across all three support modes.
- Routing that accounts for issue type, urgency, and availability.
- Structured summaries carried into the client workspace.
- Human review on recommendations and written follow-up.
The platform now runs AI-assisted intake in production. The measure is not demo quality; it is whether the right request reaches the right inspector with usable context.
Stack
Marketplace platform, AI triage, structured summaries, client workspace.