Financial documents

Evolution AI extracts data from financial documents more accurately than a person

The challenge

Precise up-to-date financial data on prospects, suppliers and clients is crucial for businesses to make informed decisions.

Obtaining the data requires delving into financial documents - such as balance sheets, cash flow and income statements. To automate this extraction process requires technology with the ability to read complex financial tables.

Our solution

Our solution extracts data from financial documents with 99.5% accuracy, higher than human performance on this task.

Our algorithms have the flexibility to understand when a field is named in multiple different ways. For example, the field ‘tangible assets’ can appear as machinery, plant, property, inventory, or many other variations, but will still be extracted correctly.

The Evolution AI platform understands complex financial tables because it ‘sees’ the visual clues, like a human. Shown here, our platform correctly extracts ‘87,272’ (blue) and ‘53,662’ (orange), even though one is off-set.

The results

These are the results we deliver for one of our clients:

  1. A completely automated process
  2. Accuracy: 99.5%
  3. Volume: 1,465,000 financial documents

Our partners


Evolution AI conducts research with the Machine Learning Department at UCL and the Computer Science Department at the University of Southampton. We are funded by one of the largest grants ever awarded by the UK Government for AI R&D.

Our partners and clients include RBS, RELX Group, Blue Prism, The Economist, Sky, Dun & Bradstreet, UiPath, the Office for National Statistics, Accenture, Natwest, Automation Anywhere, and Hitachi Consulting.