Orders into SAP, untouched: AI order management for a food distributor
A food importer and distributor in Sri Lanka supplies hotels, restaurants, retail chains and institutions, including same-day orders. Every purchase order used to be read by a person and typed into SAP.
Orders arrive any way the customer likes
A PDF by email, a photo of a handwritten list on WhatsApp, an Excel sheet, several orders in one message, one order split across outlets. The stock is frozen, chilled and ambient, so an order entered late or wrongly becomes a missed delivery or wasted product. And goods are sold on credit, so the invoice and the follow-up matter as much as the order.
| Manual step | What went wrong |
|---|---|
| Typing each order into SAP | Orders waited for a person; scanned and handwritten ones were re-keyed with mistakes |
| Matching items and prices by eye | Wrong item, price or unit; one unclear line held up the order |
| Approving | Depended on who looked and how closely |
| Chasing payment by phone | Promises not recorded; missed dates noticed late |
Four stages, with SAP left exactly where it was
1. Purchase order to sales order
Orders ingested from email, WhatsApp, SMS and API. Customer, address and lines extracted and matched against SAP. A partial order is created with matched items; unmatched ones are highlighted with suggestions.
2. Approval by risk
Each order gets a high, medium or low risk level from thresholds and tolerances. Validated and approved orders are marked ready for invoicing.
3. Automatic invoicing
A setting turns automatic invoice creation on or off. Stock is checked and allocated from the warehouse before every invoice.
4. Commitments and follow-up
Reminders by WhatsApp and email from the due date. Responses and promised dates are recorded; missed dates trigger further reminders and an internal alert.
People handle exceptions, not transcription
- Orders from any channel reach SAP without being re-typed
- Customer, item, price and unit matched to SAP master data
- Every order carries a risk level
- Invoices raised as soon as an order is approved
- Follow-up with recorded payment dates continues until payment
Time, accuracy, cash and control
| Lever | How to size it |
|---|---|
| Order-entry hours | Orders a month x minutes per order x share that goes straight through |
| Fewer errors | Errors avoided x cost of one correction |
| Earlier cash | Annual credit sales / 365 x days of collection saved |
For every 100 units of annual credit sales, each day of collection saved releases about 0.27 units of working capital. In the Gulf the same approach adds Arabic purchase orders and local e-invoicing rules.
Questions we are asked
Which order formats can it read?
PDF, images, Excel, HTML, JSON, scanned PDFs and handwritten images, including several purchase orders in one email or one PDF.
Does one unclear line hold up the order?
No. Matched lines become a partial sales order in SAP straight away; the unclear line is highlighted with suggestions and added once confirmed.
Who decides what is approved automatically?
You do. Each order gets a high, medium or low risk level from thresholds you set, and automatic invoicing can be switched on or off.
How is payment followed up?
By WhatsApp and email from the due date, with promised payment dates recorded and the internal team alerted if one is missed.
Have a problem you want measured?
Tell us what you want to count, check or protect. An engineer will reply, not a sales script.