Electronic solutions to your anti-money laundering responsibilities

RBS International Banking chief executive John Owen observed earlier this year that anti-money laundering (AML) “provides no particular competitive advantage to any one party”. Despite the lack of commercial imperative, 66% of financial services firms saw AML and KYC (know your customer) budgets rise during the past year, and 61% of firms reported an increase in headcount, according to Veris Consulting's Global Cost of Anti-Money Laundering Compliance 2013 survey.

The research captures the views of 284 respondents in 46 countries, primarily compliance professionals from retail, business, private and wholesale banks, plus broker dealers and other institutions. It found that the majority of costs stemmed from automated transaction monitoring systems: 75% of respondents cited them as the greatest expense.

Given the threat of fines from regulators, compliance with AML and KYC can be considered cost mitigation. By August 2013, the Reserve Bank of India had identified 28 banks that were failing to comply with AML and KYC legislation. They included public and private sector lenders, big and small, and multinational banks, and were subject to punishments ranging from cautionary letters to fines of Rs3 crore ($480,000).

There have been higher-profile cases; HSBC, ING and Standard Chartered paid $1.9bn, $619m and $667m, respectively, to US authorities in 2012 for past breaches of AML and KYC rules. For banks facing a wave of change programmes stemming from regulation, any way to reduce this burden is welcome.

Artificial intelligence

Aniruddha Paul, chief information officer at ING Vysya, which offers retail, business and private banking in India, employs artificial intelligence (AI) to selectively monitor transactions. "We prefer to use a sniper rifle rather than a machine gun to identify what is happening on a portfolio,” he says. “For that we use quite a bit of AI. We use relational database technologies, such as Oracle, to support our AI.”

The use of AI allows the bank to track unusual behaviour more flexibly. For example, a bank may want to monitor the frequency of transactions, which can indicate money laundering. Using a rules-based engine it can determine that if the number of transactions in a month goes above 50 then that account should be put on close watch. If the engine uses AI or fuzzy logic it will not specify the value of 50 transactions, but rather trigger an alert should the frequency of transactions substantially increase from the average, reducing the number of false positives that are created by the unique features of an account owner.

Mr Paul notes that previously a firm would have needed an online analytical processing cube, which involves replumbing the data tables of a relational database into new structures to provide more efficient categorisation and retrieval of data. By creating a fixed structure, these cubes were fit for purpose but also inflexible as requirements changed, limiting their value. 

Better hardware

“You don't need to do that now,” says Mr Paul. “It is possible to use simple relational structures and hardware firepower to deal with issues that arise. Data from transactional systems is de-duplicated and cross-checked on the basis of business rules and against fuzzy logic; algorithms that are written to come up with back-end detection and specification of values.”

Improvements in the use of hardware have allowed banks to use a simpler technology stack to support KYC/AML than was previously needed. “In the past few years that firepower of the hardware has gone up significantly,” says Mr Paul.

An example of improved hardware is the use of in-memory analytics. This takes data from the random access memory chip from which it is instantly available, rather than from memory storage – often a disk – from which it has to be retrieved.

“With in-memory techniques, processing speeds can be 1000 times greater,” says Alexon Bell, of analytics and business intelligence firm SAS. “We had a customer for whom performing multivariant analysis – analysing the input into models – took seven days. We have driven it down to 84 seconds.”

The bigger picture

Larger banking operations have a higher level of complexity than smaller firms, making the task of applying standardised checks to account opening and transaction monitoring processes more challenging.

“Some global banks are operating in more than 50 countries, with hundreds of millions of customers, and have to analyse [data across] all of those customers to find out whether there is money laundering, terrorist financing or breaches of controls,” says Mr Bell. Using relational databases can be restrictive for these banks as they have to cross-reference many types of data and documents, with different regulations in each market.

“Relational databases are fabulous with transactions and, of course, that is what banks do, but they are not good at sequential analytics i.e. one thing happens, then another thing and then a pattern forms, as one needs when looking for money-laundering,” says Professor Mark Whitehorn, chair of analytics at the University of Dundee's School of Computing.

Economies of scale

Alternative models of data storage and retrieval, such as Hadoop, the open-source platform based on Google technology, are possibly a lower-cost alternative. The amount of code needed to write queries in Hadoop is significantly less than for databases, which reduces the testing time, likelihood of errors and complexity of redesign.

“DB2, Oracle and Teradata [database software] are incredibly expensive when you start scaling (volume),” says Ravi Kalakota, partner at systems integrator Liquidhub. “Hadoop is much less costly to scale for the same level of data volume. Traditional solutions are also expensive when you handle big data characteristics: volume, velocity and variety. When all three are evolving it becomes a major issue.”

One option for dealing with rising IT costs is for banks to share the burden with one another where possible, something that RBS’s Mr Owen called for in his speech in April.

Markit, a provider of data and analytics to capital markets firms, has teamed up with operations management firm Genpact to develop a centralised customer administration platform for banks that will provide KYC checks. “Compliance with KYC and AML adds a necessary but large cost burden to our industry,” says Michele Trogni, managing director and global co-head of Markit’s solutions business. "We see an opportunity here. If we can deliver a centralised and standardised on-boarding service for the banks, it can be a strategic advantage in terms of quality and time to market, in addition to reducing costs.”

Drawn to the centre

By keeping a single central source of data for counterparties, the model should eliminate the challenges caused by old data or duplication, while a single standardised process facilitates more efficient on-boarding for the client. “With a centralised model, a hedge fund can understand exactly what it needs to provide to that one hub. It can be verified once and distributed to everyone,” says Ms Trogni.

HSBC and Morgan Stanley have both signed up for the service, and with Mr Owen’s tacit support it would appear to have hit the right note with senior management at the big banks. Nevertheless, monitoring customers’ transactional data for AML will remain an internal affair and that means a radical rethink of data storage and retrieval. One bank technologist says that the big database vendors “have been left behind” by small firms building extensions to Hadoop that will make it secure and stable enough to be used in financial services.

“Banks we are talking to are going to the pure-play Hadoop vendors,” says Mr Bell. “They are not going to the traditional database vendors to look at their Hadoop extensions.” Mr Whitehorn adds: “This is revolution.”