Startup looks to help HSBC, financial firms spot money laundering

Published on Jul 23, 2018

Major financial institutions are drowning in data.

With hundreds of thousands of customers, and literally billions of records scattered across multiple industries, sectors and geographic regions, accurately detecting money laundering is not so much like finding a needle in a hay-stack as finding a needle in the entire hayfield.

In the search for more efficient solutions, major financial institutions are turning to artificial intelligence (AI). In April this year, HSBC announced it had partnered with startup Quantexa to integrate its anti-money laundering (AML) technology into HSBC’s systems.

The development is likely to be perceived as a welcome step from HSBC, which has been heavily penalised for money laundering failings and has had to assure regulators it will clean up its act and improve on compliance.

The bank, which has a vast customer base, is looking to use Quantexa’s technology to identify potential money laundering activity by analysing information that is available internally and externally, as well as transactional data within a customer’s wider network.

“HSBC is continuously looking for ways to build on our existing capabilities to detect and prevent financial crime. We are looking forward to working closely with the company to utilise its technologies as we become more intelligence led in our approach to financial crime risk management.”
HSBC Global Risk COO Ray O’Brien

In addition to its work with HSBC, Quantexa is also exploring partnerships with a number of other financial institutions, according to Quantexa co-founder and director Alexon Bell.

It recently partnered with Deloitte, whose financial crime team will work with Quantexa to flag potential money laundering activity within financial institutions.

Tom Scampion, EMEA financial crime leader at Deloitte, said his organisation looks forward to working closely with Quantexa “in an intelligence-led approach to combating financial crime.”

How it works

Quantexa’s technology has been applied to transaction monitoring, trade based money laundering and market AML as well as a range of non-financial crime related uses. The system is based on open source software, which makes it relatively easy to adapt for a variety of use cases. Bell sees the open source base as a key part of Quantexa’s rapid success since the company was founded in 2016.

“[The co-founders] looked at the market and we noticed a few things happening. There was a very large surge in open source adoption by banks, so they’re moving towards non-proprietary platforms. The reason for that is the cost comes down.

“We also saw some frustrations from customers around having to go back to the vendor to make changes. We formed Quantexa to enable our customers to use our products in the ways that they wanted, without being tied to us.”

Quantexa’s technology is not a silver bullet, and is intended to augment rather than replace professional expertise when it comes to financial crime. The goal is to help organisations to sort the signal from the noise, for example by using the sheer number-crunching capabilities of AI to reduce the number of false positives or to link records across multiple data sources, which can then be analysed by a human.

“So for a compliance person trying to understand the flows of money and the types of risk, because we’re on Hadoop they can simply plug in the enterprise standard reporting package and then start to pull in information about that which we can service to them, but we service them with connected information rather than the one-dimensional information that they may have from other platforms.”

“For investigators interacting with the Quantexa platform, we generate alerts but we also consume alerts from other systems, link them all together and present the holistic view to investigators.

“This is quite an intuitive interface, and we are seeing very large productivity gains. So literally things that take days are being done in minutes, because we’re able to automate some of the linking and data gathering which the investigator would previously have to do manually,” he explained.

Overcoming challenges

“We formed the company to have a software platform, enterprise scale, and unlimited in the amount of data that it could connect and use for different purposes,” said Bell.

“The problems in the past were about volume and scale. The largest live platform that we’ve got is about nine billion records. It sits across five different regions and it’s rolled out to lots of users. What we know from past experience is that connecting things in networks and resolving entities at that volume is really difficult.”

“Quantexa has been able to crack that with a forensic investigation capability for very large organisations to connect their customers across these billions of records, across multiple countries and systems, and understand how they are connected,” Bell said.

“We do that both from a reference perspective – so bank accounts in different countries – and also who owns them – for example, are they connected to companies, beneficial owners, offshore structures – and who they transact with – are they individuals or companies, and what risks do those companies present.”

“That holistic picture is the unique core of Quantexa, and then it’s about applying different analytics on top of it to solve different problems.”

Accuracy, algorithms and AML alerts

According to Bell, the platform also finds connections which investigators might otherwise miss, for example linking records where details such as names are slightly different. Artificial intelligence models need to be trained on pre-existing data sets, and the accuracy of their results rely heavily on the quality of the training data they are given.

As the saying goes, “garbage in, garbage out” – that is, no matter how good the algorithm may be, if it is trained on inaccurate data then it will produce an inaccurate result. Quantexa uses training data from the financial institutions it works with, but it also draws on data from other sources such as credit reporting agencies and company information databases.

“We are basically connecting all of that data together and then applying the risk scores, so we can take all the AML alerts, all the sanctions screening alerts, fraud alerts, mule account alerts, and we build that picture,” said Bell.

“So you can imagine that normal people have very boring networks and transactions – you know, you work somewhere, you go on holiday, you shop in the same places, you don’t really connect to sanctions lists or networks of [politically exposed persons], your networks tend to be fairly contained and not as interesting as the bad people.

“The PEPs connect to other PEPs and they connect to high risk jurisdictions and shell companies, so their networks are a lot more complicated and they have a lot more red flags littered across them.”

Professional expertise is crucial to provide the context around the raw data. Quantexa’s team works closely with staff at their client organisations to develop an understanding of what the risks look like in their industry and to build that knowledge into the model to ensure data is captured and categorised accurately.

“I will trust the algorithm, but what I don’t trust is the data the algorithm uses. If the data is not categorised correctly, it does cause problems,” said Bell. For example, it is crucial to understand the reasons why a particular report is categorised as either accurate or a false positive, and to ensure false positives are not accidentally categorised as good results or vice versa.

AI and the future for financial crime detection

“AI is incredibly powerful, but we also blend artificial intelligence with human intelligence. When we look at trying to detect money laundering, we sit with investigators, financial crime threat mitigation people, and ask them ‘what does AML look like in markets, for you? Or in trade, for you?” Bell said.

“And when we sit with investigators we ask them ‘why was this alert significant, and why was this one a false positive?’ And then what we do is harvest that information and programmatically include that in our analytics because we know that these are high risk typologies or common risk factors for that organisation.”

Bell is confident about the future for AI in fighting financial crime, including expanding Quantexa’s technology to cover other areas of AML. “AI has applications across the space, without a doubt,” he explained. An exciting part of Quantexa’s development has been watching partners take the open source software and adapt it to their own needs.

“There may be 50 or so people across our installations at the banks who are Quantexa trained, so they know how to build our entities and networks and configure scores within the platform. We have a very large partner ecosystem also coming on board and those partners are helping with existing installations but also they are finding new areas and new applications of our technology that we hadn’t thought about to help solve some really difficult problems,” Bell added.

About the writer: Melbourne-based Elise Thomas has a background in international affairs and a strong interest in financial crime, data and technology issues.

Advance your CPD minutes for this content,
by signing up and using the CPD Wallet

Get started