An n8n automation that reads receipt photos sent over WhatsApp with AI vision, validates the data and records the transaction, with a guided chat and a manual-entry link when a receipt can't be read.
This is a WhatsApp-first expense capture automation built on n8n. Instead of opening an app and typing in every field, a user simply sends a photo of a receipt to a WhatsApp number and the system does the rest.
Logging expenses is a chore people skip. Receipts get lost, details are typed in days later, and the resulting records are incomplete. Forcing people into a separate app adds friction, while a plain chat bot breaks the moment a receipt is blurry, crumpled or in a format the AI can't read.
We built a resilient, multi-layer pipeline. Receipt images are analysed by a primary AI vision model, backed up by a second model, and, when neither can read the receipt, the bot switches to a short guided conversation. As a last resort it sends a secure link for manual entry. At every stage the user gets a clear WhatsApp reply, and no expense is silently lost.
A webhook receives every WhatsApp event. Status receipts are filtered out, and the sender's conversation state is loaded to decide whether this is a new receipt or an answer to an earlier question.
The workflow checks the server is reachable, picks the largest photo, downloads it as Base64 and sends it to Groq Vision. If that fails, Cloudflare vision runs as a fallback and its response is normalised.
The parsed result is checked for readability. Readable receipts go straight to recording. Unreadable ones start a guided entry that collects payee, amount and purpose, validating the amount along the way.
A payload is built and inserted through the API. The user receives a WhatsApp confirmation with a magic link to add a note, or a clear message if they aren't registered.
The automation turns expense logging into a single action: send a photo. It handles the messy reality of real receipts through layered fallbacks (a second AI model, a guided chat, then a manual-entry link), so users always have a way to finish. Every failure path ends in a clear WhatsApp message instead of a dropped expense, and the modular n8n design makes it easy to swap vision models, add steps or connect other systems.
Expense logged from one photo
Two AI vision models for resilience
Guided chat fallback
No expense silently lost
From receipt capture to lead intake and customer support, we build n8n and AI automations that meet your team where they already work. Tell us what's slowing you down.