AI expense categorisation & approval
LiveExpenseIQ
An expense system for Naira spending. Staff file a claim in minutes, AI categorises and checks it, and larger claims go to a manager queue automatically.
- Trigger01
Expense submitted
Vendor, amount, purpose, receipt
- Action02
Make webhook
Receives JSON payload
- AI step03
AI categorise
Category + review notes
- Logic04
Validate output
Reject malformed responses
- Data05
Airtable record
Reference number issued
- Logic06
Amount check
- ↳ ≥ ₦100k → manager queue
- ↳ < ₦100k → standard review
The problem
Expense claims in small businesses usually arrive as receipts in chat threads and spreadsheets nobody keeps up to date. Managers can’t see what was spent or why, and large claims get approved without a proper look.
What I built
- An AI-assisted categorisation workflow in Make that receives each claim, sends it to an AI model and writes a structured record to Airtable.
- A live front end built with Lovable: employees file without logging in and get an instant reference number, while managers sign in to an approval queue.
- Threshold routing: claims of ₦100,000 and above are routed to manager approval automatically, with approve, decline or clarify actions.
- A manager view with the full record, AI notes, spend totals and approval rates in one place.
How it handles failure
- Fixed malformed HTTP requests by tightening the JSON payload structure between the app and Make.
- Added a guard for undefined AI message content, so an empty model response cannot become a blank record.
- Resolved content-type mismatches between the webhook and the AI API so requests are parsed the same way every time.
Outcome
Every claim is captured, categorised and traceable from one place, and anything above the threshold always reaches a human before it is approved.
Stack
- Make
- Airtable
- AI model APIs
- Webhooks
- JSON
- Lovable