Manufacturing / Sales order automation
Sales order entry automation software: email and PDF orders to your ERP
Sales order entry automation for manufacturers and distributors whose customers still order by email. Send us a batch of real orders, as PDFs, Word tables, scans or plain text, and within a week you get them back as ERP-ready order lines, matched to your customer file and catalog, with every doubtful line flagged.
In production at Biodéal since July 2025: orders written to SAP Business One through its Service Layer.
The reality
Most customers never send an EDI message. They send a purchase order exported from their own ERP, a scanned form or three lines in an email, and someone in sales administration retypes it into SAP, Sage, Odoo or Microsoft Dynamics, line by line, before the order acknowledgment can go out.
What Mirage does
Automated order processing, from inbox to ERP
Mirage reads the purchase order the way your team does, applies the rules your team applies, then writes the sales order through an interface your ERP already has.
Reads every order, whatever the format
The email and its attachment: a PDF from the customer's ERP, a Word or spreadsheet table, a scanned form or plain text. Anything that is not an order, such as a delivery confirmation or a stock-out notice, is recognised and left out.
Matches it to your own master data
Customer references, units and prices are matched to your item master and customer file: exact, then close, then by delivery address. A unit missing from the purchase order is inferred from the price.
Writes the order, sends the acknowledgment
The sales order is created in SAP, Sage, Odoo or Dynamics through its API, EDI or a flat file import. The order acknowledgment can then go back to the customer automatically.
WHAT ARRIVES, AND WHAT HAPPENS TO IT
Which order formats can be automated, and what each one needs
Every format a customer uses to order, what the agent reads from it, and the rule that decides whether the line is written or held. The formats are the ones B2B order desks actually receive; the rules are the ones your team applies today, written down.
| How the order arrives | What is read | Written when | Held when |
|---|---|---|---|
| PDF purchase order exported from the customer's ERP | Header, ship-to, lines with the customer's references, units, prices | Every reference resolves through the customer's cross-reference and the price is within tolerance | A reference has two candidates, or the price is outside the agreement |
| Spreadsheet or Word table attached to an email | The table's columns, mapped once per customer | Columns map cleanly and quantities fit the pack sizes | A unit is missing and the price points to more than one pack |
| Scanned or photographed form | The form after OCR, with low-confidence fields flagged | Every field the rules need is read with confidence | A quantity or a reference is unreadable |
| Three lines in the body of an email | Customer from the sender, items from the text, quantities as written | The customer is known and each item resolves to one SKU | The ship-to is not among the customer's registered addresses |
| EDI message (ORDERS, 850) | Not handled here: it stays on your EDI connection | n/a | n/a |
| A resend of an order already received | PO number, customer, lines, dates | Never: it is linked to the first order | Always flagged as a duplicate, with the link |
What is not an order, a delivery confirmation, a stock query, a quote request, is classified and routed to the right queue with a log entry, so nothing waits in the order inbox and nothing is entered by mistake.
The output
Email orders to ERP, line by line
| Purchase order line | As received | Written to the ERP | Status |
|---|---|---|---|
| PO 4471, header | Deliver to our Lyon depot | Ship-to 2 of customer C1045 | Matched by address |
| PO 4471, line 1 | Hex bolt M12x40 zinc, 500 | Item 30412, 500 pcs | Written |
| PO 4471, line 2↳ The purchase order gave no unit. At €21.00 the price only matches a box of 100 on your price list, not a single washer, so the line is written in boxes and flagged for confirmation. | Washer M12, 5 at €21.00 | Item 30577, 5 boxes of 100 | Unit inferred |
| PO 4471, line 3↳ The customer's own reference has no cross-reference in your item master. The line goes to your team with the two closest items suggested; the rest of the order is not held back. | Ref. 88-120-B, 40 | No match in the item master | Sent to a person |
Proof
PDF order entry software, in production
Biodéal, an organic dairy distributor supplying retail chains, bakeries and wholesalers, runs this pipeline on SAP Business One.
Biodéal
48 seconds median from order email to order ready in SAP
2,943 orders processed automatically since go-live on 22 July 2025
About 7% of orders flagged for a human check, none lost silently
Reduce order entry errors on your own orders
Send us a batch of real customer orders, as they arrived. Within a week you see each one as it would be written to your ERP, with every line we would not trust flagged.
How a deployment runs: Observe, Map, Build, Run, Expand. See the method
FAQ
Frequently asked questions
Connect the inbox orders arrive in, give the system your item master, customer file and the rules your team applies, then let it write orders to the ERP through its API, EDI or a flat file. The first two weeks run in parallel with your team, so every automated order is compared with the one keyed in by hand before the switch.
Software that takes an incoming customer order, in whatever format it arrives, and turns it into a sales order in your ERP without anyone retyping it. The hard part is not reading the document but applying your rules: which item a customer's reference means, which unit, which warehouse, which discount.
It depends on your volume, the lines per order and the rework that entry errors cause further down: credit notes, re-deliveries, disputes. The honest way to measure it is to count the orders keyed in each day, the time each one takes and the corrections traced back to an entry error. We do that count with you before quoting anything.
Take the retyping out, and check every line against your own master data before it is written: item, unit, price and delivery address. A line the system cannot confirm goes to a person with the reason, instead of being guessed.
Any ERP that accepts orders through an API, EDI or a flat file import, which covers SAP, Sage, Odoo and Microsoft Dynamics. The production deployment at Biodéal writes to SAP Business One through the Service Layer.
It uses large language models to read an order that no two customers format the same way, and rules and master data to decide what to do with each line. The reading is generative AI; the writing into the ERP is a standard interface call, and nothing is written that failed a check. What matters to an order desk is the second half.
Test it on a week of your own orders, including the awkward ones, and ask four things: which formats it reads without a template per customer, what it checks against your master data before writing, where a held line goes and with what evidence, and which named deployment it can show with the definition of each figure. The sales order exception handling guide on this site lists the vendor questions in full.
No. Keep EDI with the customers who already send EDI orders. This handles everyone else: the customers who send a PDF, a spreadsheet or an email.
Most deployments go live in 4 to 6 weeks, then run in parallel for the first two weeks to confirm accuracy before full cutover.
A fixed-price pilot on a batch of your real orders first, so you see the result before committing. After that, an annual subscription priced by volume, quoted in the first conversation.