AI Walkthrough lets roofing technicians document a job by talking while they film or photograph it. Zuper AI turns the narration into a clean job note.
It launched at Zuper in 2025 and covers the two places technicians document most: the camera and the checklist.
From the shipped product
Documentation happens on a ladder.
Typing on site is slow and unsafe
- Checklists, QA notes, and photos were mandatory, and slow to complete.
- Technicians reported juggling ladders, tools, and phones.
- Incomplete documentation frustrated the back office.
The back office reported notes that were too ambiguous or incomplete to act on.
One job, many hand-offs
Jobs
inspection, roof repair, emergency service, solar installation.
Documented at
inspection, material procurement, repair, quality assurance, job completion.
What the notes were missing
Connection
notes and photos didn’t line up.
Continuity
notes from different technicians on the same job rarely linked together.
Accessibility
notes in a technician’s own language or shorthand were hard for the office to read.
The bet: capture by voice.
We couldn’t ask technicians to type more on a roof. The solution had to save time and reduce risk while still giving managers complete documentation.
Three principles
Talk, not type
capture observations in the moment.
AI does the tidying
summarising and translation.
Build a foundation
a base for more AI in field service.
Beta feedback changed the capture.
We explored directions for both the camera and the checklist, then ran an early beta with a select group of customers and tracked usage with Mixpanel.
What the beta showed
We started from photos. The beta moved us to video.
- Video over photos: technicians expected to record video rather than tap for each photo, and some still wanted photos during a recording, so we kept that.
- Dual camera: managers asked to see both the technician and the work, for accountability and proof of presence.
What shipped
Hands-free documentation
Technicians talk while filming or photographing, and AI turns the narration into a note.
Checklists you can answer by voice
Long checklist answers can be spoken instead of typed. AI transcribes and summarises each one.
What we reported
These are the figures reported at the time. Hundreds of notes were standardised, summarised, and translated by Zuper AI.
- Documentation time
- 30 → 10minutes
- Reported.
- Note quality
- 40%better
- Reported; measurement method not documented.
- Reported adoption
- 100%of the reported group
- The field technicians in the reported group moved their documentation to Walkthrough mode. This is not a figure for every Zuper customer.