Rossum Aurora 1.5 and Copilot: Instant Learning for 276 Languages

Nazar Labunets October 29, 2024

Some of Rossum’s largest enterprise customers process 10 million documents a year from as many as 7000 suppliers in 18 languages.

This is the complex environment in which many large organizations operate. They require heavy customization, flexibility in processing documents in multiple languages, and configuring data transformations to prepare the data for downstream systems.

This is a challenging task.

But what if you could process 99% of all transactional documents worldwide in all common languages and easily configure any data transformations that you need without coding?

Today, we’re turning that “what if” into reality with the launch of Rossum Aurora 1.5 & Copilot.

This groundbreaking update is designed to tackle the three biggest headaches in enterprise document processing:

• The inability to process documents in all the languages in which you conduct business and receive documents in
• Slow processing times for 100+ page documents like customs clearance document bundles
• Lack of IT resources available to support automation projects for back office and customer-facing teams

A look back at Rossum Aurora 1.0

In February we introduced Rossum Aurora, our next-generation AI for transactional documents powered by our proprietary Transactional Large Language Model. That initial release brought you:

• Instant learning: Seamless human-AI collaboration that automatically trains the model without IT support. Plus, 10x fewer documents are needed to reach the desired accuracy.
• 37.6% reduction in error rates.
• Trustworthy and safe AI: 0 hallucinated values unlike the LLMs you already know and confidence scores behind every data point.
• And much more.

Let’s dive into the highlights of Rossum Aurora 1.5 & Copilot

Rossum Aurora 1.5: Instant Learning for 276 languages

Enterprises that operate around the world work with documents in multiple languages, with up to 30% containing handwriting. Most solutions on the market today would require you to train AI models for each new language. This means involving IT or data scientists, as well as an annotator who speaks the language. Training requires up to 5000 documents to achieve acceptable accuracy levels. This is too slow and expensive.

Rossum Aurora handles documents in 276 languages and self-improves with every annotation.

With Rossum Aurora 1.5, our T-LLM has been trained to capture data in 276 languages, with Instant Learning support—the ability to learn instantly from documents annotated by users and use gathered knowledge to predict values on documents in the future. With this update, Rossum Aurora can also capture handwriting in 30 languages.

What this ultimately means is that you have one solution that:

• Handles all transactional documents and languages
• Requires no IT involvement for training and deploying models

4X faster processing of documents with 100+ pages and other under-the-hood updates

Many industries face strict SLAs and can’t afford delays in processing large files containing multiple different document types and up to 500 hundred pages.

For example:

• Logistics back office teams process customs clearance documents on behalf of their clients. Document packs often contain a commercial invoice, proof of origin, packing list, waybill, and bill of lading, all in one PDF.
• AP teams process a batch of invoices scanned into a single PDF.

For such teams and others, missing SLAs could mean government fines, unhappy customers, or missed discount opportunities.

Rossum Aurora 1.5 brings significant under-the-hood updates:

• Large documents (100+ pages) are now processed 4X faster than before.
• Our AI gets constantly better with all the new data it sees. Since the release of Rossum Aurora 1.0, we have reduced the error rate by another 25%.

66% less time configuring Formula Fields with Copilot

Businesses require clean data in their downstream system of record. But often, the data is missing from documents or it’s in the wrong output format, which leads to additional effort and time spent processing these documents, lowering automation.

For example, AP and customer service teams often need to:

• Calculate the price after a discount for orders above a certain amount
• Ensure the Document ID is 10 digits by adding leading zeros
• Make sure BIC/SWIFT codes are uppercase
• Remove spaces from IBAN
• Etc.

When done manually, all of these significantly increase document processing time and are prone to errors.

That’s why we are excited to introduce Formula Fields with Copilot, which allow non-technical users to define complex data transformations using natural language commands.

If you are an existing customer, Formula Fields replace 5 functions in the Rossum platform, making the configuration much easier for you.

Key advantages:

• 66% reduction in configuration time for data transformations
• Natural language interface for defining complex calculations
• Preview feature to easily validate formulas

Delivering end-to-end document automation on a global scale

Rossum Aurora 1.5 addresses the evolving needs of our customers:

• Handling documents in multiple languages
• Extracting data from documents with handwriting
• Processing large documents
• Enabling simple configuration of complex data transformations.

Whether you’re in finance and accounting, logistics, or managing a shared service center, Rossum Aurora 1.5 & Copilot offers the tools you need to streamline your document processing workflows, increase accuracy, and drive operational efficiency.

For Rossum, it’s another step toward our mission: Enable one person to effortlessly process one million transactions a year.

Our roadmap for 2025 continues in this vein with a major update to our Transactional Large Language Model. We are also working hard on enabling you to digitize your business processes, not just capture data on documents.

Double dare you. Break our AI.

Put our AI engine to the test with your documents. Different document types and formats. Multiple languages. Handwriting. Do your worst.