News

March 13, 2020

How artificial intelligence is changing the game for banks

Eigen Technologies, which uses small datasets to train its AI, has bagged deals with ING and Goldman Sachs.


Maija Palmer

5 min read

Lewis Z Lu, founder of Eigen Technologies

When the fintech sector moves to using artificial intelligence on an industrial scale, the winners are unlikely to be today’s big tech companies. Instead, think niche. Think small data sets. Think startups that you’ve probably not heard of before.

Or at least that’s according to Eigen Technologies — the UK-based natural language processing startup, which today announced ING as a second, strategic financial services investor, joining Goldman Sachs.

The amount of money in the deal is pretty modest, $5m, rounding up Eigen’s Series B funding round to $42m.

Advertisement
The big tech behemoths don’t care about the Dodd-Frank Act or European banking liquidity. They are solving for a talking kitchen appliance.

But the deal gives an insight on how the finance industry is planning to use technologies like natural language processing, the branch of artificial intelligence (AI) that allows machines to analyse written and spoken speech.

So far, AI has mostly been a big tech, big data game. Algorithms have been taught to deal with unprocessed information in text, images, audio and video by training them on vast amounts of data. The more data you can access, the better your AI, making this a game dominated by the likes of Google, IBM, Facebook and Amazon.

But a handful of small, nippy startups are finding ways to train AI with small data sets, an approach that works better for specialist and regulated businesses.

“The traditional machine learning approach is to throw a lot of data at the problem, to give the programme 100,000+ examples. But in some cases, you may only have a much smaller number of examples of something — for example, a trader with a set of 300 collateralised loan obligations — and it is not enough to train on,” says Lewis Z Liu, cofounder of Eigen Technologies.

Eigen says it only needs two to 50 examples to train its natural language processing algorithm. (No, that’s not a typo — it can work with as little as two). This is partly because Eigen has trained its NLP algorithms specifically around legal and financial documents — it spent most of the £13m it raised in its Series A round on buying 22m financial services documents to train its algorithm on.

“The big tech behemoths [developing general AI] don’t care about the Dodd-Frank Act or European banking liquidity. They are solving for a talking kitchen appliance,” says Liu.

Big opportunities with law firms — even bigger ones with banks

Eigen started out working with legal clients, such as Linklaters, helping the law firm parse legal documents like contracts. It then realised that the need in financial services to automate document reading was several orders of magnitude greater

“With a law firm you might be helping them automate 300 documents. With financial services they might need to be reading 1m,” says Liu.

There is no human way of completing the Libor transition by 2021 without use of machine learning.

One big source of potential business comes from the financial services sector’s shift away from the London Interbank Offer Rate, or Libor, a benchmark lending rate that underpins some $350tn worth of financial contracts worldwide. The interest rates on credit cards, car loans and mortgages are linked to Libor, and all these will need to be re-written as the financial services sector moves away from using this benchmark next year.

“There is no human way of completing the Libor transition by 2021 without the use of machine learning,” says Liu.

Advertisement

Eigen is one of a handful of specialist AI companies starting to come to prominence. Luminance, a UK startup backed by Mike Lynch’s Invoke Capital, is doing similar work with law firms and raised $10m in last year from investors including legal firm Slaughter and May. Seal Software, a Californian startup that helps companies such as Nokia, PayPal and Dell analyse contracts, was bought for $188m by DocuSign last month.

The closest “big tech” competition probably comes from IBM Watson, although much of Watson’s recent work with banks has centred around understanding customer queries.

How machine learning will change banks

The partnership with ING has been several years in the making — Liu chuckles a little wryly as he tries to estimate the time the bank’s due diligence took.

But now, Benoit Legrand, ING’s chief innovation officer, says the bank is investing, in part, to signal that it is backing the technology.

“We are signalling that we are committed for the long term and that we have validated the technology,” he says.

It would help with crisis management. It can show you things such as if one part of the bank goes bankrupt, how quickly would that crisis spread.

There is no plan to buy the whole company, he adds. The $5m investment is part of a multi-pronged approach to innovation at the bank. ING builds some of its own startups, with 27 initiatives started so far in internal labs and has partnered with some 200 fintech companies on projects. The €300m corporate venture fund has so far invested in 27 startups including Eigen.

Legrand says the technology has the potential to be transformational.

“Natural language processing will dramatically change the way we will operate. There is a tremendous amount of hidden knowledge locked away at a bank — we’re sitting on a goldmine. This will give us a way to access it,” he says. “We will be able to make faster, better decisions on everything from mortgages to calculating how much collateral the bank holds.”

Legrand’s mention of collateral begs a big question: would AI — with all its better, faster ways of handling data — have been able to sound an early alarm on a big systemic problem like the financial crisis?

Liu, for all his ambition, baulks at making quite such a big claim.

“I don’t think it would avert a crisis because there are so many different elements that come into it. But it would help with crisis management. [This technology] can help show you things such as if one part of the bank goes bankrupt, how quickly would that crisis spread.”

With the world economy rocked by COVID-19, he adds, banks are again coming under pressure to spot problems with liabilities and non-performing loans as quickly as possible. Eigen may very soon get a chance to demonstrate what it can do.