AI-powered back office: scanned bills and WhatsApp orders into the books
An AI pipeline for a trading business: purchase bills are scanned, read and entered into the accounting software, and customer orders sent as photos, typed text or voice notes on WhatsApp and email are parsed, completed and posted once staff confirm them.
RESULTOrder turnaround cut from 4 hours to 30 minutes, with purchase bills and sales orders posted to the books without re-keying.
- Purchase bill AI
- Order intake AI
- Clarification bot
- Staff review queue
- Accounting posting
“Customers order however they like: a photo of a handwritten list, a WhatsApp voice note, an email. We used to read each one, call back for missing details and type it in again, and a typical order took 4 hours to confirm. Now the system reads it, asks the customer what's missing and has it ready for us to approve. We handle 20–50 orders a day, most need no correction, and turnaround is down to 30 minutes.”
- 4 h → 30 min
- Order turnaround
- 20–50 B2B orders a day
- Volume
- ~85%
- Parsed without staff fixes
- WhatsApp · email
- Order channels
- Photo · text · voice
- Input formats
- None
- Manual re-keying
Supplier bills arrived on paper and were typed into the accounting software by hand. Customers ordered however suited them: a photo of a handwritten list, a WhatsApp message, a voice note or an email. Staff read each one, chased missing details by phone and keyed the order in again, and a typical order took about 4 hours to confirm.
- Scanned purchase bills read by AI and entered into the accounting software as purchase vouchers
- Orders accepted as photos, typed text and voice recordings, on WhatsApp and email
- Every order parsed and matched against one order schema: customer, items, quantities, rates, delivery
- Missing or unclear fields trigger an automatic question back to the customer on the same channel
- Staff confirm each order in one click, and it posts straight to the accounting software
- Order turnaround down from 4 hours to 30 minutes
- 20–50 B2B orders a day; about 85% parse cleanly, and the ~15% with messy data are flagged for staff
What we built
Purchase bill capture
Bills are scanned or photographed, read by a vision model, matched to the supplier and item masters, and entered into the accounting software as purchase vouchers for review.
Multi-channel order intake
WhatsApp Business and email feed one inbox. Photos and handwritten lists go through vision, voice notes through speech-to-text, and typed messages straight to the parser.
Schema matching
The AI maps each message to the order schema and to the customer and item masters, so "2 boxes of the usual" becomes real items, quantities and rates.
Clarification loop
When a field is missing or ambiguous, the system asks the customer on the channel they used and fills the order from the reply, instead of a staff member phoning them.
Staff confirmation and posting
Staff see the original message next to the parsed order, confirm or correct it, and the confirmed order posts directly into the accounting software.
- WhatsApp Business
- Accounting software
- Vision AI (OCR)
- Speech-to-text
- Laravel
- PHP
- Queues
- LLM APIs
- WhatsApp Cloud API
Still re-keying orders, bills or chat messages?
Tell us where the typing happens today and include a sample message or bill. You get a written scope, timeline and quote within 3 business days.