eyeDES: From detecting cancer to spotting market abuse


Peter Lee
Published on:

Features Analytics applies artificial intelligence developed for medical use to spotting capital markets manipulation.

How do you switch careers from fighting cancer through advanced medical diagnostics to combating fraud in financial services?

Well, the obvious route is via a postgraduate degree in computer science, a doctorate in applied physics and years working in artificial intelligence.

In 2014, Cristina Soviany founded Features Analytics, based in Brussels, where she transposed principles coming from her earlier work in cancer tissue characterization with ultrasound imaging to create eyeDES – a new machine learning and artificial intelligence platform to detect financial market abuse.

Having first worked on payments, the company is now concentrating on the capital markets, gaining increasing momentum with global investment banks and exchanges.


Cristina Soviany, 
Features Analytics

“I had been used to analyzing huge volumes of data and identifying with heightened specificity the presence of any abnormal tissue that could be malign while seeking to avoid false positives and indeed false negatives,” Soviany recalls. “I had the idea to apply the same principles to financial fraud, where data is everything but is also constantly evolving. And we began to develop software tailored to the payments industry.”

Then in 2016 one of the banks it had been dealing with challenged the company to help it detect manipulation in the foreign exchange market.

“They gave us a huge quantity of data and asked if we could identify anything interesting in it,” says Soviany. “They were very surprised that, with almost no knowledge of the underlying market, we identified a number of suspicious cases that the bank’s existing system had failed to spot, even while it generated large volumes of false alerts.”

The secret sauce is not to fixate on price action but to look for anomalous behaviour

“Our technology analyses the messaging data, not just the book of orders, and enhances this in real time with additional signals in a market context. The system is based on anomaly detection and identifying how far any market participant’s behaviour deviates from that of other market participants,” Soviany tells Euromoney. “That makes it different from many of the solutions from established vendors, which often operate on fixed rules, are very conscious of price movements and previous examples of market manipulation. They are less quick to detect new anomalies and emerging forms of manipulation, such as spoofing with cancelling of trades.”

eyeDES diagram_780

Those rules-based systems also produce huge amounts of alerts, most of which are not relevant. 

Soviany says: “Our system is not biased around old suspicious scenarios but rather looks at evolving market patterns and detects any new anomalies. That cuts the number of alerts sent to compliance staff by up to 90% and produces only high-quality alerts, with attached explanations of the divergence from normal market activity.”

New application

From an initial start in foreign exchange, Features Analytics is now applying its approach to equities.

Soviany says: “We have been working with a leading global investment bank that gave us equities market data from a specific geography. Its current solution for detecting market abuse was looking only at fixed time intervals, usually periods of most intense activity. 

"Our system is more dynamic. It was able to identify something the existing solution had missed, an account that exhibited a suspicious spoofing behaviour at another moment during the day and this was caught by our solution which uses a parameter-free alerting unit with dynamic thresholding.”

Features Analytics is also talking to providers of market infrastructure –  exchanges and trading platforms ­– as well as to more banks and investment banks focusing just on activity among their own customers and traders.

Dealing with market infrastructure providers may allow a broader view of evolving patterns of market activity and divergences from them. That may help with investigations into layering and wash trading – forms of spoofing in which connected groups appear to buy and sell between themselves in an effort to create a misleading impression of deeper market activity to lure in buyers or sellers.

Growing demand

And why would a computer scientist want to make that pivot from detecting cancer to detecting market misconduct?

Partly because Soviany is an entrepreneur who has spotted an opportunity and set up her own business. Also because the opportunity is big. The increasing frequency, scale and sophistication of attempted fraud, as well as a sharper focus on money laundering by regulators, are driving growing demand across the banking industry for more advanced anti-financial crime tools.

A forthcoming report suggests that the global market for services that detect and prevent fraud is expected to see a compound annual growth rate of 20% over the next six years and will reach $71 billion by 2025. That’s up from $16.5 billion in 2017.

The UK’s FCA produced a first annual survey of reporting firms’ financial crime data in November last year and found that while cyber-crime is the big worry – with identity fraud and theft, phishing and computer malware the most prevalent – the overwhelming majority of 2,000 firms reported almost every kind of fraud is increasing, especially those enabled by new technology.

Proportion of firms that perceived the incidence of fraud increased or decreased, by fraud type, 2017 


Source: Financial Conduct Authority (FCA) analysis of 2000 firms' financial crime data.

In the final quarter of last year CGI announced CGI HotScan360, developed in close collaboration with large banks, which delivers integrated, intelligent and real-time anti-money laundering, customer due diligence and fraud detection capabilities.

“Financial crime is increasingly sophisticated, and new technologies enable criminals to replicate fraud across the globe 24/7,” says CGI vice-president Jan Macek. “In addition, with real-time payments, banks no longer have hours or days to analyze transactions, but must do so instantaneously within milliseconds.”