What is sales order automation, and how is it different from OCR?

What is sales order automation?

Sales order automation is software that turns a customer's purchase order into a sales order in your ERP. It reads the order, extracts the header and every line item, and matches each line to your product catalogue. It checks quantities, units, pack sizes and prices, then creates the sales order.

The distinction between the two documents matters. Your customer sends you their purchase order. Once it is captured in your system, it becomes your sales order. Sales order automation is the step in between, done by software instead of by someone re-typing it.

That step is harder than it looks. Orders arrive as PDFs, Excel sheets, CSV exports, scans, phone photos, handwritten notes, WhatsApp messages and lines typed into the body of an email. Each customer has its own layout, its own product codes and its own way of writing a quantity.

What does OCR actually do?

OCR, or optical character recognition, converts the pixels of a scanned page or photo into characters. It is mature and dependable on clean print. Its output is text, sometimes with the position of each word on the page.

OCR does not know what the text means. It cannot tell which number is a quantity and which is a price. It cannot tell a customer's product code from yours. It has no idea whether '4 x Cola 500 12s' means four cases or four bottles.

Most OCR-based order capture adds templates on top: zones drawn over a known layout that say 'the PO number lives here'. Templates work until a customer changes their layout, sends a second page, or forwards the order from a different system. Every new customer needs a new template.

Where does OCR stop on a real order desk?

Picture the orders that reach a beverage and dry goods distributor on an ordinary Monday. OCR can turn every one of them into text. None of them is ready for the ERP after that.

• A PDF from Ridgeback Foods prints the customer's own code, RB-10442, and no code of yours.
• A spreadsheet from a hospitality group has a quantity column and an empty price column.
• An email says 'same as last week but double the rice', with last week's order three replies down the thread.
• A photo of an order book page is taken at an angle, in poor light, in two handwriting styles.
• A forty-page PDF carries allocations for six stores, each of which should be its own order.
• The total at the bottom of a scanned order does not equal the sum of its lines.

Each of these needs a judgement: which product, which unit, which price, which order. That judgement is where an order desk's time goes, and OCR leaves all of it with the order desk.

What does extraction with confidence scores add?

Extraction reads the document as a structured record rather than as a page of text. A vision-capable AI model reads native PDFs and images; spreadsheets are read as data. The output is a fixed set of header fields and line fields, whatever the layout.

At the header, that means customer, PO number, dates, currency, totals, shipping address and payment terms. On every line, it means SKU, description, unit of measure (UoM), pack, quantity, price, line total and due date.

Each field carries a confidence score. You set the threshold. A field above it flows through. A field below it is shown to a person, highlighted, with the source document alongside.

Good extraction also follows rules a template never could. A document holding six store allocations becomes six orders. An allocation table with one product and a column of stores becomes one order per store. A missing subtotal is computed and labelled as computed. A total that does not reconcile is flagged, never silently overwritten.

How are order lines matched to your catalogue?

Reading a line correctly is half the job. The other half is deciding which of your products it refers to. Catalogue matching in Weeva uses a 13-step matching algorithm that combines deterministic and statistical methods with frontier LLMs and reasoning models to match the customer's product description to your exact product code. In practice:

• A code the customer prints, whether your SKU or a code you already know, resolves to your exact product code.
• A description in the customer's own words still finds its product, so 'Cola 500 x12' and 'Cola 500ml PET case' can meet.
• Match Memory learns how this customer has ordered before and remembers the amendments your team has made, per customer.
• A line that is still uncertain is flagged for a person to confirm.

The buyer is matched separately, against your customer list. When a person confirms a match, Weeva remembers it for that customer. The next order with the same description resolves without review. Corrected once, then automatic.

How are units, pack sizes and prices checked?

'1 x case' from one customer and '12' from another can be the same order. Weeva learns each customer's convention per SKU and converts the ordered quantity to the shipped quantity, keeping both on the order. This is pack size calibration.

Units of measure are normalised the same way. Cases, cartons, trays, singles, kilos, litres and metres are converted to your catalogue's units on write. A quantity that does not divide into the pack is flagged, not rounded.

Prices come from each customer's own price list. Orders that arrive with no prices are priced from it, which is price-less matching. A printed price that differs from the list is flagged for review, never overwritten. The VAT basis is set per order or per customer, and totals are reconciled against it.

What happens when the software is unsure?

Uncertainty is normal. What matters is that it is visible and routed. Weeva places every order in an inbox with one of three statuses: ready, needs review or synced.

Review reasons are written in plain words. Some are hard stops, such as a customer not found in your list or a SKU not in your catalogue. Some are soft, such as a low-confidence field or totals that do not add up. A likely duplicate order, with the same PO number, customer or basket as a recent one, is held for a person to confirm or dismiss.

Every decision a person makes in human review is captured as training signal. The same customer, format or description is handled automatically next time.

How does the order reach the ERP?

Approved orders are pushed to your ERP by API, with the customer and SKUs already resolved. Pushes are idempotent, so a retry never creates a duplicate sales order. Every edit, approval and sync is recorded in the audit trail with the user and the time.

Teams without a connector can download an Excel workbook, copy rows into their own layout, or track orders on an in-platform board. Weeva connects to Sage, SAP, Syspro, Xero, QuickBooks, NetSuite and others; the integrations page lists each one.

What should you look for when comparing tools?

• It reads your worst documents, not only your cleanest, with no template per customer.
• It scores confidence per field, not per document.
• It matches lines to your catalogue and remembers confirmed matches per customer.
• It handles units, pack sizes and per-customer price lists, including orders with no prices.
• It never invents a number, and says when a figure was computed.
• It records who changed what, before and after.
• It writes to your ERP without creating duplicates on retry.
• It can run alongside your current process before anything is switched over.

How do you measure it on your own orders?

Headline accuracy figures say little about your business, because accuracy depends on your documents, your catalogue and your customers. The only useful measure is one taken on your own orders.

1. 01Collect a sample of real orders, including the awkward ones your team dreads.
2. 02Run them through the software in parallel with your current process.
3. 03For each order, count the fields corrected, the lines matched without review, and the orders that went through untouched.
4. 04Repeat weekly. Corrections should compound, so the review queue should shrink.
5. 05Compare the time spent per order before and after, for the same customers.

This is how Weeva is rolled out: on your orders, beside your existing process, with the parallel run measured on your own documents. The 90-day plan sets out each stage. Bring a few of your own orders to a demo and we will show you the extraction and matching live.