As October is slowly drawing to a close we are entering the last week of our Open Autumn Call. So far, we have an incredible response from tech startups. A record number of founders have handed in their applications to join the imec.istart.nl business accelerator program. Our call closes at November 1st, 23:59. So, if you want to join the worlds #1 university-linked business accelerator, take this opportunity now.

Book a meeting here.

Every three and a half minutes an error can creep into your business records. Business accounting is prone to human error: wrong types of declarations, duplicate invoices, and incorrect VAT codes, to name but a few pitfalls. With their intelligent algorithms, founders Dirk Westra van Holthe, Mischa Masthoff and partner Thomas Offerman of wijcontrolerenjedata.nl (In English: ‘we check your data’) help companies to detect such errors at an early stage. Eliminate unnecessary accounting risks by catching errors in time.

Before starting wijcontrolerenjedata.nl, Mischa and Dirk worked in (financial) administration. They saw departments relying entirely on sophisticated models and dashboards. But if there are mistakes in the data, you can’t rely on the insights either. They also saw that administrative sloppiness led to unnecessary costs, peak workloads and wrongly management information. They saw an opportunity.

“Financial records are often riddled with errors. If you are lucky, these errors are caught and put on a list of things to fix. But often, the process stops there. The errors just sit there”, says Mischa. The same errors will creep back into your records the following month. Wijcontrolerenjedata.nl scans records for errors and risks and helps organizations find, fix and prevent mistakes.

Multifunctional platform

Mischa and Dirk see customers using their platform in a variety of ways, for a one-off analysis or a few analyses a year. They also have customers who use the platform to continuously check their financial administration. For example, to check for duplicate invoices in the system and avoid double payments.

How does the wijcontrolerenjedata.nl platform work? Mischa and Dirk: “We start by retrieving customer data via a secure, fully GDPR-compliant connection. We can connect to their source system or data warehouse or export the data from the system. Our algorithms then detect errors and risks in your records. We can run it daily, weekly, or annually. Any time, really.”

They continue: “After the analysis, all errors and risks appear in a drop-down list. Through a workflow, we record each error or risk and who did what, and when to fix the error or risk. Ultimately, all of this helps us to make structural improvements in the financial administration. If you continuously check for errors, you will discover patterns, that will help you avoid these errors in the future.”

Predictive model with own AI model

At wijcontrolerenjedata.nl, they’ve been working with AI since before tools like ChatGPT were introduced. “It was cool to be able to demonstrate that we’ve been working with AI for a long time”, says Dirk. “We see that many current language models are too generic. Take a booking text, for example. You have 80 characters to tell your story. That’s a lot of abbreviations. Many AI tools can’t handle that very well.”

That is why wijcontrolerenjedata.nl is working hard to improve its own AI model. They train their model with booking texts (with the explicit permission of the customers concerned, of course). Mischa: “We have manually labelled over 100,000 booking texts to train our model, including words, dates, locations, and much more.”

Their efforts are already yielding great results. For example, they use the results of their language model to calculate time periods. Dirk explains: “Suppose you have two invoices. The first invoice says: ‘Activities De Vries week 1 to week 4’. The second invoice says: ‘Activities De Vries week 3 to week 5.’ You would pay twice for part of the work, namely weeks 3 and 4. Your current accounting software will never find this error. Getting the software to understand the content of an entry text is extremely complex.”

For example, in the data that wijcontrolerenjedata.nl manually tagged, they found 1,848 different ways of noting a date. Asterisks, dashes, they saw it all. And what about made-up words like ‘augseptoctober’? Once your language model understands a date, you can automatically compare time periods. “It is a hugely valuable application for many analyses.”

Improving is learning

On 27 May 2024, during the Business Angel Network Nijmegen (BANN), wijcontrolerenjedata.nl was one of the start-ups that pitched their story. “In preparation for BANN, we improved our ‘story’ a lot”, explains Mischa. “For entrepreneurs, it’s common to have blind spots in your story; some things make sense to you but not to an outsider. That’s why we went back to the core: what exactly do we do, and how do we help our customers? Through Briskr, who organised everything perfectly, we got in touch with various experts and investors. And BANN gave us a lot of valuable feedback. Our product is good, but there is still room for improvement in our sales strategy. Although no investment was made, we decided to improve our sales strategy ourselves. This insight alone made the BANN process invaluable!”

Time for action!

What is most important to Dirk and Mischa? To have a cool company with nice customers! And they are right on track. But the men also want to stress the need to focus on awareness: “We’re in a niche that a lot of people don’t know about yet. We have found that an error creeps into financial records every three and a half minutes. It’s like a leaky faucet. To err is human, but things get out of hand once you leave them and do nothing about them. So, get your financial records checked. Today, rather than tomorrow. The implications are bigger than you might think.”