How Adyen achieved 92.6% accuracy after only 20 documents

Adyen case study

Who are they?

Adyen is a leading financial technology firm for retailers, hospitality, travel, subscription and digital businesses. Founded in the Netherlands in 2006, Adyen has grown to employ over 4000 people in 27 offices across the globe, and processed over $970bn in payment volume in 2023.

What was the challenge?

Adyen processes over 1 million invoices and Boletos, an official Brazilian payment method, every year. The huge volume of documents meant manual processing was becoming impossible and unscalable. This triggered a search for an intelligent document processing (IDP) solution that could reliably automate the end-to-end Accounts Payable (AP) process and reduce the risk of human error.

Due to the high variability in invoice layouts, Adyen preferred an AI-based solution that had the flexibility to quickly learn new formats, including Nota Fiscals and Boletos, rather than a template-based solution requiring IT resources. Adyen also wanted to streamline data flow directly into their system of record, Workday, to eliminate the need for manual data entry from one system to another. However, it was important that the IDP solution could verify accuracy against Workday data, ensuring that only high-quality information was posted to the downstream system.

Adyen chose Rossum, the only IDP solution focused on transactional document automation, for its market-leading AI, data validation, and out-of-the-box extensions environment.

“It was becoming impossible to process the volume and complexity of invoices manually. If we wanted to continue growing sustainably as a business we needed an intelligent document automation solution. Experiences with other automation providers had left us very disappointed, we felt that Rossum offered us the AI visibility and control to help us continue to scale our operations efficiently.” 

Marco Favaretti, Financial Controller, Adyen.

What did Rossum do? End-to-end processing

Adyen required Rossum to ingest invoices as email attachments. Rossum automatically split and sorted the PDF documents contained in the emails into queues ready for data extraction by AI, which “reads” the invoices without the need for templates.

While extraction accuracy was important, to implement trustworthy end-to-end automation of document processing, Adyen needed to ensure only high-quality data entered their downstream system. Rossum solved this by integrating with Workday and deploying Rossum’s Master Data Hub extension to validate the extracted information against Workday master data. The extension was customized according to Adyen’s specific business rules to ensure it matched spend category, tax codes, cost center, and location data. All data validated in Rossum’s UI, including PO, supplier, and currency data, was sent to Workday along with the original PDF documents.

Nota Fiscals (Portuguese invoices) and Boletos (a proforma invoice) required a special workflow, different from regular invoices. Rossum’s AI was trained to recognize these documents and send them to a specific queue for processing. Where they arrived in the same PDF, the documents were split, extracted, and validated against the Workday master data. At the end of the process, they were matched together and exported as one invoice along with the PDF to Workday.

Rossum’s professional services team also configured the integration between Rossum and Workday to allow new suppliers to be created within the Rossum UI. The integration allows the system to search for suppliers within Workday and match the information, if none is found a new one can be created in Rossum and the information sent to Workday where the new supplier will be approved manually. Adyen also customized Rossum to check for duplicates based on Adyen’s pre-defined business rules. Utilizing the Duplicate Detection extension, Rossum will detect if a document is already processed by the customer and will flag it to the business-user.

The outcome of this end-to-end solution was an exceptionally high and reliable automation rate of 74.8% on average, which freed the AP from an unsustainable manual workload.

The flexibility of Rossum’s cloud-native architecture to adapt to our workflows and business rules via its customizable pre-built extensions and integration with Workday enabled us to automate our document processing end-to-end.

 Marco Favaretti.

Deploying next-generation AI – pushing accuracy levels to new heights

Recognizing that Adyen is a technology leader in the fintech space with an agile ethos, we invited them to be an early adopter of our next-generation AI, Rossum Aurora. 

Rossum Aurora uses instant learning to quickly gain an understanding of each new document it “sees”, just like a human does, and gets noticeably smarter with each one. This cutting-edge technology uses advanced AI models and is powered by Rossum’s proprietary Large Language Model (LLM), a neural network trained on millions of transactional documents. Additionally, unlike public LLMs, Rossum’s customers benefit from enhanced security and trustworthy automation that flags when confidence is low. 

Rossum deployed the next-generation AI to Adyen’s invoices in the Netherlands, Brazil, and the US and compared the results with the previous generation of Rossum’s AI in an A/B test.

While the previous generation AI performed excellently, Rossum Aurora pushed the accuracy levels even higher. The next-generation AI produced higher data quality by reducing errors by an average of 20.5%, improving extraction accuracy to 93.4% on average. 

Unmatched speed of learning

“Rossum Aurora achieved a 92.6% accuracy rate after processing only 20 documents. To see the AI learning and instantly getting smarter gives us confidence we can truly deploy AI to automate processes and scale sustainably.”

Marco Favaretti.

To estimate the speed of learning, we simulated starting from scratch and found that it would take only 20 documents for Rossum Aurora to achieve a high initial threshold of 92.6% accuracy for a new Accounts Payable use case. 

This market-leading performance has helped build trust in the AI, which along with the easy-to-use interface, has contributed to the adoption of Rossum by Adyen’s AP team.

Accuracy graph showing Rossum Aurora’s speed of learning versus the previous generation of Rossum’s AI.

Adyen instant learning accuracy graph

 

What’s next?

The successful digitization of invoice processing has encouraged Adyen to work with Rossum Aurora toward IDP automation at scale across the business. 

Saving time, improving efficiency

93.4%

Av. accuracy level

20.5%

Av. error rates reduction

92.6%

Av. accuracy threshold after 20 annotated documents

Ready to get started?

See why hundreds of companies power their document processing with Rossum.