Automating Order Entry With AI: When Your Customers Still Order by Email, PDF and Excel
It is 7:40 in the morning at a building-materials wholesaler near Stuttgart. The order desk has not opened yet, but the inbox already holds nineteen messages. One is a PDF on a roofing contractor's letterhead. Two are forwarded Excel sheets with last quarter's article numbers still in them. Several are plain emails that read like a text message: "Morning, send me 40 bags of the usual cement and 6 rolls of the membrane, same site as Tuesday." By nine, someone will have retyped every one of them into the ERP, line by line, and phoned two customers to ask what "the usual" means. This is the part of B2B no glossy demo ever shows, and it is exactly where AI-assisted order entry earns its keep.
Why the orders still arrive the old way
You can build the cleanest self-service portal in the DACH region and a stubborn share of your customers will still email. A site foreman ordering from a phone will not log in. A purchasing clerk at an old-school contractor has an Excel template he has used for eleven years and will not give up. None of these people are wrong - they are simply ordering the way that fits their day. The cost lands on you: an order desk that spends its mornings as a human OCR machine, copying numbers between two screens and praying nobody fat-fingers a quantity. The honest answer is not to force everyone onto a portal. It is to let the email keep coming and have software read it. That is what Selldi's AI order entry does: it takes the message, the PDF or the spreadsheet, works out what was actually ordered, and prepares it for your ERP, without the clerk typing it twice.
What the AI actually does with an email
When an order lands, the AI pulls it apart the way an experienced clerk would. It identifies the customer from the sender and the letterhead, separates the real order lines from the greeting and the signature, and maps each line to a product in your catalogue - matching on article number, EAN or GTIN, or the free-text description when that is all the customer gave. It reads quantities and units, so "6 rolls" and "40 bags" come through against the right items. The result is a draft order, on the right account, with the right prices already applied - because Selldi knows that customer's individual price list and group discounts, and does not quote list price by accident.
Where the human stays in the loop
This is the part to be plain about: the AI proposes, a person confirms. It is not a black box that books orders into your ERP behind your back. A clean line - clear article number, exact match, quantity in stock - sails through. Anything uncertain gets flagged rather than guessed: a customer wrote "membrane" and you stock four, or a quantity looks like a typo because nobody orders 400 of those. The clerk reviews the flagged lines, fixes them in a moment, and approves. The job shifts from typing every order to checking the handful the AI was unsure about - a far better use of an experienced person than copying digits. Once approved, the order flows into your ERP through the API, whatever system you already run, with stock, prices and orders syncing both ways every 15 to 60 minutes. No separate data-entry step, no second source of truth.
What it catches, and what it cannot
It is strong at the repetitive grind: pulling line items out of a tidy PDF, matching codes, normalising units, applying the correct per-customer price, turning a fifty-row spreadsheet into a draft in seconds. What it cannot do is read minds. If a customer writes "send the small ones" with no further clue, the AI flags it rather than inventing an answer - and that is the correct behaviour. It will not reliably parse a blurry photo of a handwritten note, and it will not resolve a request so ambiguous your own staff would have to phone about it. The point is not to kill every phone call. It is to cut the morning's twenty manual entries down to the two or three that genuinely need a human.
When this is not worth it
Be honest before you buy anything. If you take a handful of orders a day and the desk is never under pressure, this is a solution looking for a problem. If nearly all your customers already order through a portal or EDI, there is little email left to read. And if your product data is a mess - duplicate codes, missing EANs, descriptions that do not match the shelf - the AI inherits that mess, because matching is only as good as the catalogue behind it. Clean the data first, or as part of the project, but do not expect software to paper over it.
Where it pays off is the wholesaler drowning in inbound emails, PDFs and spreadsheets, with staff retyping instead of selling. Selldi runs B2B and B2C on one back end, sells across languages and currencies for export, and typically goes live in around two weeks - an implementation fee plus a subscription, no commission on your sales. See it working at demo.selldi.pl/showcase, or write to helpdesk@selldi.pl and describe the kind of orders your inbox actually gets.