How Woodvision transformed its order intake with agentic AI

200 orders a day, nine steps each

Woodvision is a Dutch wholesaler in garden timber, and as soon as spring arrives, their order department goes into overdrive. On peak days, up to 200 orders come in, via email with PDFs, WhatsApp, phone calls, or through sales reps.

Each order went through an average of nine steps before reaching the ERP system. That took about four minutes per order, with roughly four order lines each, resulting in a lot of manual input. Repetitive work, little variation, and plenty of room for error. And every mistake meant a return shipment: costly, time-consuming, and entirely avoidable.

The portal didn’t solve it

Woodvision had already invested in a dealer portal to let customers place orders themselves. The goal was to process over 80% of orders digitally. Spoiler: it didn’t work.

Many customers preferred to stick to their own ways of working, systems, and habits. Others simply lacked the IT knowledge or budget to change. Increasing that percentage turned out to be a dead end. The bottleneck wasn’t the portal; it was the sheer diversity in how customers wanted to communicate.

The solution had to adapt to Woodvision’s reality, not force change on their customers.

Why traditional IDP wasn’t an option

Traditional Intelligent Document Processing (IDP) relies on mappings. For every document type from every customer, you build a template that tells the system where to find relevant data.

With hundreds of customers and formats, that means hundreds of mappings to maintain. And when a customer changes their format without notice (which they will), the mapping breaks, and so does your process.

There’s another issue: customers don’t always speak your system’s language. They send product descriptions that don’t exactly match internal product IDs. A human back-office employee understands “pine beam 4m” instantly. A traditional IDP system doesn’t. That kind of contextual knowledge is nearly impossible to capture in static mappings.

So that approach was off the table.

Interpreting instead of mapping

Together with Incentro, Woodvision took a different route: an Order Intake Agent that automatically reads incoming emails with attachments, regardless of format or structure.

No per-customer mapping needed. The agent understands and interprets information, rather than matching it against predefined templates.

Here’s how it works:

• An email with an attachment comes in
• The agent retrieves the PDF and converts it into structured data using Amazon Textract
• An LLM identifies relevant information such as product IDs, quantities, and delivery dates
• Products are matched to Woodvision’s assortment via direct ID matching or semantic search
• The order is prepared for review—or sent directly to SAP

Fields with lower confidence are visually flagged in a review interface, so employees know exactly where to focus. Everything else can be skipped. Nine steps become three.

Less than a minute per order

Processing time dropped from around four minutes per order to under a minute. Errors decreased by roughly 35%, meaning fewer return shipments, less waste, and less frustration on both sides.

For regular customers with consistently accurate orders, the review step is already becoming optional.

Antoine de Kleine, e-commerce manager at Woodvision, highlights what surprised him most:

“The only thing we do when an order isn’t processed correctly is report it to Incentro. They adjust the prompt, and the next time it works. I barely have to change anything on my side.”

Low maintenance, high scalability: mission accomplished.

What’s next?

The next step is expanding the agent to process emails without attachments—orders typed directly into the email body.

Woodvision also plans to teach the agent to handle industry-specific calculations. For example, if a customer orders 21 meters of timber in 3-meter pieces, the agent should automatically calculate that as seven units.

Another improvement under consideration is suggesting replacement products when discontinued items are ordered - removing the need for manual lookups.

As the agent processes more orders, it keeps getting smarter. And Woodvision scales along with it.