Orders and sales
Automatic order intake from spreadsheets, PDFs and photos with AI
Problem
Customer orders arrive as spreadsheets, PDFs and photos in messengers, and your managers spend hours retyping them into the accounting system.
Solution
A manager uploads the customer’s file, and the system builds a draft order from it, matches the items, converts the units and learns from every correction.
Who will recognise this
This module solves a problem for almost any B2B business that takes orders as assorted files.
- Distributors and wholesalers whose customers order in their own spreadsheets.
- Manufacturers (FMCG, food, household chemicals) supplying retail chains that use strict order forms of their own.
- Suppliers to HoReCa and retail, where orders often arrive as photos of handwritten lists or messenger screenshots.
- Suppliers of building materials, auto parts and pharmacy goods with large catalogues and complex SKUs.
Feel it most: Sales managers, Order entry operators, Head of sales, Business owner
The business pain
Every customer sends orders the way it suits them: a retail chain sends its own peculiar Excel form, a small shop drops a photo of a notebook page in WhatsApp, and a distributor sends a PDF from its own system.
Each customer has its own names for the same products, and quantities come in boxes, kilograms or pieces. The operator opens the file on one half of the screen and the accounting system on the other, and copies the order line by line.
What it leads to
Human error and returns
With manual entry, mistakes are inevitable: the wrong pack size, boxes mixed up with kilograms so the order is several times too small or too large, or simply a skipped line.
Company knowledge in one head
Only an experienced manager knows which “odd name in the customer’s form” is which product in your database. When they go on holiday or leave, the work stops.
A bloated team
The more customers you have, the more people do mechanical retyping instead of selling and working with customers.
What we automate
The module can run as a separate lightweight add-on on top of your current accounting system or ERP, without breaking familiar processes.
Any file format
A manager uploads a file in any format (Excel, CSV, text PDF, photo or scan), or a whole batch in an archive. Customers can also upload their own form through a personal web link.
Smart matching against the price list
The system extracts the items, quantities, order number and delivery address, for example finding the store by its number in a chain’s PDF form. Every line is checked strictly against the products this customer is allowed in their price list.
Quick review on a split screen
The operator sees the original document next to the assembled draft. Exact matches go straight through, and doubtful lines are highlighted in yellow with a list of suitable candidates: the manager only checks those.
Automatic unit conversion
If the customer ordered in kilograms or pieces and you ship in boxes, the system converts the quantity through the pack weight. If it doesn’t come out to whole boxes, the system highlights the remainder.
Learning on the fly
When a manager fixes a yellow line once and picks the right product, the system remembers the match for the customer’s whole group of companies. Next time the same form is recognised in full, with no one touching it.
Under the hood: protection from AI mistakes
The main fear when bringing AI into a business is that the model will “invent” a product that doesn’t exist or mix up pack sizes. We ruled this risk out in the architecture.
AI only reads, the algorithm checks
The model is used only to read the document and to help choose between a few candidates. The matching itself is done by a strict algorithm that follows your database rules, so the system physically cannot put in a product with another label (private label), a different weight or one outside the customer’s range.
Cleaning out the noise
Before comparing names, the algorithm strips filler words and standardises pack sizes, so two different products with the same weight never merge into one.
Protecting volumes
The system checks units against the table header in the form. If no unit is given anywhere, it defaults to boxes (the larger pack), so “silent kilograms” never shrink a customer’s order several times over.
Sanity and margin checks, optional
The system can flag at once when a customer usually takes 10 boxes but today’s form says 150 (protection from customer typos), and warn when the price in the order is below current cost.
What the business gets
Speed
The operator checks a ready draft in 30 seconds instead of 15 minutes of manual entry.
Knowledge stays in the company
The quirks of every customer’s and every chain’s forms stay in the database for good, not in one employee’s head. A newcomer can process complex chain orders on day one.
Errors stopped before shipping
Mistakes with pack sizes, units and other brands are blocked by rules before the order is saved.
Transparent economics
Every AI call is logged with its exact cost, and the number of manual corrections falls week by week.
How to check the effect in a pilot
We don’t make abstract promises in percentages. For a pilot we suggest measuring four specific indicators.
- Time from receiving the file by email or messenger to a finished document in your system.
- Share of lines the system recognises straight away, with no operator involvement.
- How corrections change: how many fewer manual fixes a manager makes in week four compared with week one (the learning effect).
- Number of item and pack-size errors that used to surface only in the warehouse during picking or at the customer’s goods receipt.
What we need to start
To deploy the module and show results on your own data, we need very little.
- An export of your product catalogue (with pack weights) and customer price lists from your accounting system or ERP, or simply in Excel.
- A list of delivery points with store numbers, if you work with retail chains.
- 10–15 real order forms, PDFs and photos from your key customers for the initial setup.
A 3-minute demo
At the presentation we take three real files from your customer: a chain’s spreadsheet with its internal names, a PDF and a photo of a crumpled form. We upload them, and a few seconds later there is a ready draft. One line is yellow (“matched loosely”), we pick the right product from the list and save. Then we upload the same file again, and the system recognises every line automatically.