The short answer
Automating order processing with AI means that orders arriving by email, PDF, Excel or WhatsApp are read automatically, matched line by line to your own products and placed as a draft order in your ERP or accounting package. Price, stock and customer agreements come from your own system, never from the AI. An employee only reviews the doubtful lines and releases the order. At Tamboer AI this one process costs 1,500 euro excluding VAT and is in place in ten working days.
Which orders you can automate with AI
AI helps mostly with orders that do not arrive structured. An order through your web shop or EDI is already in the right format. An email saying "same as last time, but 24 units" is not. That is where the retyping, the searching and the errors sit.
| How the order arrives | What AI prepares | What an employee does |
|---|---|---|
| Email with free text | Customer, products, quantities, delivery date and reference taken from the text | Resolve doubtful lines and release |
| PDF purchase order in the customer's format | Lines read out, even when every customer sends a different format | Check what differs from the customer agreement |
| Excel or CSV order list | Columns recognised and matched to product code, unit and pack size | Assign unknown products |
| WhatsApp or a phone note | A draft order with a list of what is still missing | Ask for the missing details |
| Web shop, customer portal or EDI | Nothing: these orders are already structured | No extra step needed |
How order entry with AI runs, step by step
- Order arrives. In the shared order inbox, as an email, attachment or message.
- Customer recognised. From your own customer data, with the reference or project number attached.
- Lines matched. Each line to a product from your own product list. Product code, unit and pack size must match; a similar description is not enough.
- Price, stock and agreements fetched. From your ERP or accounting package, never invented by the AI.
- A verdict per line. Certain, doubtful or no match, with a pointer to the place in the original email or PDF.
- Draft order and draft confirmation. Ready in your own system or in a review screen alongside it.
- An employee releases. Only the doubtful lines need attention. Nothing goes to the customer without approval.
Illustrative example, not a client figure: a PDF order of twenty lines. Seventeen lines are matched with certainty, two have two possible products and one has no reliable match. The employee looks at three lines instead of twenty.
Three ways to automate order processing
| Route | Works well when | Limit |
|---|---|---|
| Web shop, customer portal or EDI | Large, regular customers are willing to order through your channel | Smaller customers keep emailing; you cannot force them |
| PDF recognition with a template per customer | A few customers always send the same PDF format | Every new or changed format needs a new template; free text does not work |
| AI workflow with human review | Orders arrive in changing forms: email, PDF, Excel and free text | Needs a product list in good order and an employee who reviews the doubtful lines |
If nearly every large customer already orders through your web shop or EDI, AI is not the first step here. I will say so. Often it is a mix: EDI for the large customers, the AI workflow for everything that arrives outside it.
Which ERP or accounting package it works with
I build on what you already have first. Approved lines go to your own system through an API, an import file or a review screen alongside the package. Which route works depends on your package and edition, and we check that with your administrator before building. If you use Exact Online, see how I prepare orders for Exact Online. For other packages, the approach is on the software page.
If your package can already pull orders from PDFs, switch that on first. I do not rebuild what your software already does.
What I built around orders and enquiries in practice
Three examples from the projects on this site. At Tandtechnisch Laboratorium Beijn, Power Automate links form, scan file, delivery date and practitioner to one order, and Copilot checks the job against the requirements per product type: 50 percent fewer status questions and 7 hours a week back. At Team for Logistics, Claude reads the whole email thread and produces one transport brief, with missing or conflicting details flagged: 9 hours a week back. At Van den Bosch en Fikkert, Copilot pulls the specifications from request, attachment and previous order: 55 percent fewer correction rounds.
For technical wholesale there is a fully worked example on order processing for wholesale.
What AI does not do in your order processing
- Set a price, discount or delivery time by itself.
- Release an order or send a confirmation without a person.
- Replace your ERP, web shop or EDI.
- Touch orders that already arrive through EDI, web shop or portal.
- Promise a link with your package before we have checked what your edition allows.
When it fits and when it does not
Probably worth it when
- a noticeable share of your orders still arrives by email, PDF, Excel or free text;
- your package has an API, import or export, or a review screen alongside it is already a gain;
- your order desk cannot keep up with order volume, or errors only show at delivery.
Probably not the first step when
- nearly all orders already arrive through web shop, portal or EDI;
- your product data is not in order: duplicate codes, old products, no units;
- you only enter a few orders a week by hand. You would never earn the build back.
What it costs and how to start
You start with a free AI plan. I look at your order flow and the rest of your business, and within a week of our call you get a plan, three tailored examples and a demonstration. If AI does not fit here, you will hear that too.
After that you can have this one process automated with AI: 1,500 euro excluding VAT, reviewed, built and tested in ten working days. You pay half at the start and half at delivery. If it does not work as agreed, I fix it or you get your money back. The choice is yours. You pay licences directly to the supplier. All prices are on the pricing page.
Frequently asked questions
Can AI enter orders from email and PDF automatically?
Yes, as preparation. The AI reads the email or PDF, matches each line to your own product list and places a draft order in your ERP or in a review screen. Price and stock come from your own system. An employee reviews the doubtful lines and releases the order.
What does it cost to automate order processing?
At Tamboer AI one process costs 1,500 euro excluding VAT, in ten working days, with half at the start and half at delivery. You start with a free AI plan. You pay licences and usage of the AI services directly to the supplier; you see them up front.
Does AI order processing work with Exact Online or another ERP?
Usually yes, through an API, an import file or a review screen alongside the package. Before building, we check with your administrator whether your edition supports the required order fields and a suitable import or API route.
What is the difference between AI order processing and PDF recognition?
PDF recognition works with a template per customer format and gets stuck on a new format or free text. An AI workflow reads changing forms, such as email, PDF and Excel, and shows per line how certain the match is. That is why an employee keeps reviewing the doubtful lines.
What happens if the AI picks the wrong product?
The workflow only matches with certainty when product code, unit and pack size match. Otherwise the line stays open as doubtful or no match, with a pointer to the source. An employee releases every order, so a doubtful line never reaches the customer unseen.
How long before it works?
Ten working days from the agreed start, once the examples and access to your systems are in place. That covers the review, the build, testing on real orders and the walkthrough for your order desk.
About the author
I am Joost, AI specialist and founder of Tamboer AI. I have worked with AI daily since 2022. I tested and developed applications at one of the largest banks in the Netherlands and now build AI workflows for smaller businesses, with a person who checks every outcome. More about me.
