AI

AI waybill recognition in Telegram for a distributor

All incoming waybills from 40 suppliers now arrive via a Telegram bot. Photo → 30 seconds → ready-to-review draft in the warehouse system with 92% of line items auto-matched.

Client A climate-equipment distributor, Belarus
Industry B2B wholesale · climate equipment
Duration 2 weeks
Type AI document recognition
30 sec per waybill recognition
100% of waybills via the bot
92% line items auto-matched

The problem

A distributor takes in goods from 40 suppliers. Every incoming waybill was keyed into the system by hand — supplier, line items, quantities, VAT-inclusive prices. Dozens of waybills a day, 5–10 minutes each. Item-numbering errors surfaced only at bank reconciliation — two to three weeks after the fact. Warehouse staff put goods on the shelves and then waited for the accountant to enter the document.

Before

  • Each waybill — 5–10 minutes of manual entry: look up the supplier, transcribe line items, double-check VAT-inclusive prices.
  • Discrepancies between «waybill in the system» and reality only surfaced at bank reconciliation — 2–3 weeks later.
  • Warehouse staff waited for the goods to be entered. Items sat on the floor while the system said «not in stock».

What we built

  • An internal Telegram bot for receiving waybills: an employee photographs the document → 30 seconds later gets a link to a ready-to-review draft in the warehouse system.
  • The bot identifies the supplier from the 40 it already knows — even if the waybill says «LLC Acme Corporation» and the system has just «Acme Corp», it still matches correctly.
  • The bot matches line items from the waybill against the catalog — 92% of rows land automatically. The remaining 8% are flagged: «these need a human».
  • The bot understands Belarusian waybill structure: it picks the VAT-inclusive price, not the «ex-VAT» one — no more manual recalculation.
  • If something other than a waybill arrives (a meme, a screenshot, a selfie) — the bot doesn't stay quiet. It talks back to the sender in its own distinctive style and asks for a proper document. Staff send waybills more eagerly — partly, it seems, for the bot's reply.

Results

  • 100% of waybills (hundreds per month) go through the bot. Manual data entry, as a process, is gone.
  • 30 seconds to recognize a waybill, versus 5–10 minutes of manual entry.
  • «Waybill vs reality» discrepancies surface instantly on upload — not three weeks later.
  • Warehouse staff register goods immediately — no more «waiting for the document» delay.
  • The accountant handles the edge cases (the 8% of items that didn't auto-match), not the hundreds of routine lines.

The client's name is withheld under NDA. Real names and details are shared at the brief.

A similar task?

Describe it briefly — within 1–2 days we'll come back with an analysis and an hourly estimate. If a packaged solution fits, we'll say so honestly.

✈ Telegram