A Deep Learning experience was the spark for Evolution AI: Fintech interview 27
Evolution AI is creating machines that can read documents in any language so humans don’t have to and is taking some decision making out of their hands as well.
The company is one of the increasing number of startups that target the banking sector as a main customer base, but are not actually providing financial services.
There is a bit of a blurring of lines between fintechs, insurtechs and regtechs. Evolution AI has mainly financial services companies as customers, hence the fintech alignment but as its work often involves compliance for banks it also fits snuggly into the Regtech sector.
For example the company has done work for Royal Bank of Scotland (RBS) to automate its Know Your Customer (KYC) process. When potential customers sign up to RBS products and services all their documentation is read and cross checked automatically by Evolution AI software, which can then automatically make a decision on whether the customer qualifies. In the past people would read all documents and then make decisions.
Martin Goodson, who founded Evolution AI and is now CEO, was a machine learning scientist and had run data science teams at some London based startups. He them met Rafal Kwasny who had been working at investment banks in IT departments with a focus on big data and analytics.
It was 2015 when they kicked off the company. They had the tech expertise but having not worked in the business side of finance the two techies had to initially spend time talking to potential customers to help them understand the business needs.
Goodson said: “We needed to try and understand the marketplace and what where the very document based processes in banks. We knew the technology really well and wanted to understand how we could bring this to the enterprise world.”
The interest began much earlier. At an event in 2011 Goodson did a workshop in Deep Learning, before it was famous. “It was a small workshop and no one had really heard of it but it was really exciting. We saw what natural language processing, which teaches computers to read or understand human language, could do.”
So it wasn’t frustration in the workplace that triggered the founding of the startup but the realisation that a technology could revolutionise the way businesses work.
“We have an assumption that if we can teach computers to read documents and make decisions based on this there are a hell of a lot of business processes that could be automated,” added Goodson.
It was 2015 when Goodson and his co-founder built the first version of the platform combining their combined expertise in big data and machine learning. It could handle millions of documents, read them and make decisions based on that.
It got its first customer in 2016 when commercial data, analytics, and insights for businesses Dun and Bradstreet (D&B) signed up. In finance D&B provides services such as credit scoring and provides data to support KYC compliance. “It is an important part of the financial services ecosystem,” said Goodson. Dun and Bradstreet still uses the software.
KYC is the ideal home for Evolution AI’s software with its document heavy, labour intensive and repetitive nature.
In 2017 RBS Evolution AI’s first banking customer. The companies are working together on several projects, including RBS’s KYC project, but many have not been made public yet, said Goodson. “Other than UK government departments and Dun & Bradstreet all our customers are in the banking sector.”
For example Goodson said the company is working with another UK bank on reading invoices. “Most invoices contain similar information, but traditionally it has been difficult to automate the extraction of data because they come in many different varieties and formats. As a result businesses use humans to process them.”
For example Optical Character Recognition technology works well if the reader knows where to look on the page, but with invoices you can’t do that because the information could be anywhere.
“But our technology has a lot of cognitive flexibility and can understand overall document structure just by looking at the visual element on the page, much like a human does.”
Evolution AI has ten members of staff in London where its research and development is done, so most are engineers. I also has a small team in Poland.
The company received a large AI grant, about £800,000, from the UK government through Innovate UK.
“We demonstrated why our technology was more advanced than anything in the market and were successful,” explained Goodson. It has used the money to expand its capabilities to be able to read text in images. For example text in PDFs.
“It can read text in any format. Machine readable or not, we don’t care.”
How accurate is Evolution AI technology? Well the company has SLAs with customers today that mean accuracy has to be 99.5%. “The system knows when it doesn’t know and will pass to a human to check when this happens. It also learns from any corrections and improves.
The company is aiming to double its revenues every year. “We now need to get out there and communicate to the world what the technology can do.”
It recently launched fully self-service platform which allows customers to train the AI themseves.