What it is
An internal tool for The Naturally Clean Co that turns a ten-minute on-site audio walk-through into a written Standard Operating Procedure plus a job-specific checklist. The cleaner records themselves talking through a client's job, uploads photos, and the tool produces:
- An SOP framed against TNCC's standard template
- A checklist generated from the SOP that the team uses on subsequent visits
- A completed checklist that flows back to management when the job's done
Not client-facing. The output that goes to clients is just the completed checklist with anything sensitive scrubbed.
How it started
Writing an SOP for a client used to mean shuffling through emails, text messages, customer notes, and the cleaner's memory of what they did last time. For a business with dozens of regular clients and complicated houses, that's a lot of half-remembered detail living in scattered places.
I wanted the team to be able to capture the knowledge in one go, on the day, in the format their iPhones already make easy: by talking.
How a recording becomes a procedure
The cleaner opens the app and hits record. They walk the house and narrate what they're doing. Which rooms in what order, what the client cares about, which products go on which surfaces, where parking and access work, the quirks. Photos go in alongside the audio. Ten-ish minutes per job.
The audio gets transcribed. AI takes the transcript and structures it against TNCC's SOP template, so the resulting policy follows a consistent shape every time regardless of how the cleaner organised their walk-through. The checklist falls out of the SOP automatically.
Next visit, the team opens the checklist in the app, ticks items off as they go, and the completed version flows back to management. We forward the relevant parts to the client.
What I'd change before unpausing it
It's currently shelved while I work on three things:
- The UI is ugly. It works, but it's not something I'd want to put in front of a contractor without apologising first. Whole-pass restyle.
- The AI takes liberties. When the model converts audio to SOP, it occasionally invents structure that wasn't in the transcript, or smooths over an awkward bit instead of capturing it honestly. I need tighter prompts and probably a model swap for the structuring step.
- The raw transcript needs to be saved alongside the SOP. Right now we only keep the AI-generated output. If the AI got something wrong, there's nothing to fall back on. Keeping the transcript means we can re-run the structuring step with better prompts, or audit what was actually said.
None of those are hard. They've just been competing with paid work.
Status
Paused. The end-to-end flow works but isn't in active use until the rework lands.
