ING and UniCredit bring AI to the heart of the capital markets


Peter Lee
Published on:

European banks are investing in and using the tools of an AI startup applying deep learning to syndicated loans, asset management and, soon to come, primary bond markets.

It’s not the biggest fintech investment Euromoney has reported on, but the participation of ING Ventures and UniCredit in a €1.3 million series-A funding round for Italian startup Axyon AI might be a signpost to the future for wholesale banking and finance.

Artificial intelligence is most known for being the power behind the ‘if you bought that, surely you’ll want to buy this’ follow-on recommendations on Amazon, adverts on Google suggested by previous search history and fraud detection by retail banks spotting purchases or transfers that don’t fit a customer’s previous habits or locations.

Axyon AI, by contrast, aims to apply AI to the core of the capital markets.


Frank Abbenhuis,
Axyon AI

“There are many AI startups active in the finance vertical,” Frank Abbenhuis, vice-president of strategic alliances at Axyon AI, tells Euromoney. “But when we started, there weren’t many focusing purely on building a solid technology platform to apply deep learning to time-series forecasting of financial data.”

Deep learning is a subset of machine learning where an algorithm is ‘trained’ on a set of data through which it establishes a logic structure, and the logic structure is then applied in a ‘decision algorithm’ to infer outputs without supervision.

Axyon founder, Daniele Grassi, had seen the rise of AI in the computer engineering department of the University of Modena in northern Italy from about 2006. But, not wanting to build another solution searching for a problem, Grassi remained patient until 2016 before he saw a possible application to finance.

The company participated in the accelerator programme at ING and was introduced to the capital markets side of the bank.

“AI has been around for many years and is widely used in retail banking,” Abbenhuis says. “In capital markets, there is an awful lot of data, but as it has been slower to adapt to emerging technologies there is still room to create transparency in otherwise opaque markets.”

The company’s first product is SynFinance, which comes as an off-the-shelf platform that analyses the entire history of the syndicated loan market as captured by Refinitiv.

Abbenhuis says: “Because we can see every bank that has participated in nearly every syndicated loan and also the high-level structure of those loans – which is far more data than any human mind can individually process – we can spot patterns that help syndication desks assess the liquidity of a new deal and, in turn, offer better advice to their clients.”

Own data

Banks can also add in their own data: which financial institutions its syndication desk invited to participate in previous deals, who accepted and who declined.

Abbenhuis suggests that in a primary loan market dominated by the top 10 arrangers who together corner 40% of the fees, a tool like SynFinance can level the playing field, and perhaps allow a borrowers’ smaller relationship banks a means to better assess likely support for a new loan structure. It’s a potential differentiator.

benoit legrand 160x186

Benoît Legrand,

Benoît Legrand, chief innovation officer of ING and chief executive of ING Ventures, says: “ING Ventures is constantly looking for fintechs that help us to deliver a differentiating experience for our clients. Axyon AI specifically is of interest because of their use cases in AI that can bring value to financial institutions, thanks to their predictive solutions.”

He confirms that, as well as being an investor in the company, ING’s wholesale banking advanced analytics team and the syndicated lending team are using SynFinance to enhance their activities.

UBS banking analysts for now find that AI is still most commonly being deployed in the replacement of retail banking middle- and back-office staff and in improving risk and fraud management, as well as compliance and regulatory functions.


There is an obvious and long-running battle being waged here between banks to wring efficiencies from these essential and mundane processes and to provide a low error rate service to customers and reduced costs for shareholders.

Shifting market share in wholesale funding is a very different battle. And if deep learning can suggest potential supporters for a loan deal beyond those a bank’s syndication team can identify themselves – even experienced syndicators don’t know what they don’t know – where else might SynFinance work in capital markets?

The obvious next step would seem to be into other forms of debt.

“With UniCredit, we will explore the domain of bonds,” says Abbenhuis.

The Italian bank doesn’t go into details but is clearly doing more than just investing in a promising startup company for a financial return.

ranieri-de-marchis 160x186

Ranieri De Marchis,

Ranieri De Marchis, co-chief operating officer at UniCredit, says: “In our view, fintechs can effectively generate ideas for banks, adding to the digital transformation already under way. 

"We are strongly committed to developing innovative services and we are now proud to be investing in and partnering with Axyon AI in order to accelerate our digital transformation and further enhance the quality of our client advisory services.”

Axyon AI has also developed a tool for fund managers called Iris, which provides periodic predictions of variables used by asset managers to develop the investment products they sell to customers. These include asset performance metrics such as volatility, Sharpe and Sortino ratios.

“We are absolutely not asset managers ourselves, but we believe that the parameters on which they base their strategies are often backwards-looking,” Abbenhuis says. 

“So, if we can predict how those parameters, such as volatility, might behave in the future and offer managers a view on what’s likely to happen, they can sensitize their strategies based on those likely scenarios. We’re currently experimenting with generative adversarial networks, where we pit two neural networks against each other, to simulate future market scenarios.”