Genocide’s Private Banker: Why it’s Perfect Timing for Machine Learning in Anti-Money Laundering
Money laundering is nasty stuff and stopping bad actors is good for the world. On May 1, 2015, BNP Paribas was hit with the largest financial crimes compliance fine in history, $8.9B, for providing services to the Sudanese government from 2002 until 2007 while it was sanctioned over carrying out a genocide. Financial crime directly supports child trafficking and terrorism financing. Applying machine learning in this domain stops the spam that’s jamming up the system, helping find more bad actors, cut $80B of bloated cost structure in compliance, protect identities, and improve customer experience.
Genocide’s Private Banker
Arbab’s mother woke him up one night during the genocide of a UN-estimated three hundred thousand citizens in Darfur. The Sudanese militia formed circles around villages like Arbab’s, attacked with guns and machetes, raped women, and went house to house burning the villagers alive in their own homes. Arbab was thrown in his home to burn alive, saw a way to escape, and ran to a tree where he had agreed to meet his mother if separated. Arbab was separated from his family for three years before reuniting with his mother, who even now makes him call her if she can’t visibly see him. The conflict in the Darfur region of Sudan began in February of 2003. At least 400,000 people have been killed; more than 2.5 million civilians are displaced.
Halima’s father died defending her village against an attack from the Sudanese militia.
She still does not know what happened to the rest of her family. She was later punished by the authorities for speaking out about the government’s lack of support for her population. She was transferred to a clinic where she had to take care of sexually assaulted women. She once had to care for over 40 girls who the militia barricaded inside a school and raped while their parents remained outside. Eventually, Halima was taken away by the militia, beaten, raped, and told: “We’re going to let you live because we know you’d prefer to die. Isn’t that clever of us? Aren’t we clever, doctor?”
On May 1, 2015, BNP Paribas was hit with the largest financial crimes compliance fine in history — $8.9B — for, according to regulators, “central bank for the government of Sudan.”
In 2007, a compliance officer at BNP Paribas warned colleagues, “In a context where the international community puts pressure to bring an end to the dramatic situation in Darfur, no one would understand why BNP Paribas persists [in Sudan] which could be interpreted as supporting the leaders in place.” BNP processed over $6bn in transactions on behalf of Sudan’s government and banks, provided credit and financing, stripped identities to clear transitions through the bank’s sanctions screen systems, and routed transitions through puppet accounts.
In an executive meeting in 2005, a Swiss compliance officer shared strong concerns about the bank’s Sudanese banking business. BNP’s then group COO Georges Chodron de Courcel dismissed the concerns and directed that no minutes be taken in the meeting.
BNP has made massive strides since that fine, taking a leading role in reviewing and improving internal processes for stopping financial crime. But the problem is extremely complex; and technology is unlocking new ways to stop criminals in their tracks.
The International Financial Crime Problem
About $4.38T annually is laundered, according to the UN Office of Drugs and Crime, or about 5% of global GDP. It’s estimated that the global financial services system spends as much as $100B a year on financial crime compliance, and perhaps $15–20B+ on enterprise software and analytics alone, but is still suffering $35B in regulatory fines annually.
More aggressive combating of financial crime started with the G7 establishing a global Financial Action Task Force (FATF) in 1989 to tackle money laundering, and it broadened its mandate to include terrorism financing after the 2001 terrorist attacks. As regulatory enforcement took hold, fines accelerated. Between 2011–2014, AML-related fines grew from less than $1B annually to over $15B.
The intelligence community leans on regulators, who lean on banks with rules-based systems which then backfire and spam the bank, regulator, and intelligence analysts with false alerts. This spam is so immense that individual banks like HSBC or JPMC can spend billions and have 15K+ analysts reviewing alerts coming from these anti-money laundering risk screening engines.
Why it’s Perfect Timing for Tech Innovation in AML
Historically, banks have resisted innovating in compliance because it is a lose-lose proposition. If they try a new ML approach and it fails, regulators can fine them. And if they try a new approach that succeeds (and therefore discovers a bunch of previous cases they missed), regulators can respond by fining them. So they were deterred. But that has changed. Regulators are encouraging and sometimes even mandating innovation, including AI/ML. Between the immense pressure of the fines, costs and headcount escalation combined with the reduced friction to do to new regulatory acceptance of machine learning and the move to the cloud, the timing is perfect to set a machine learning layer on top of all these rules engines and cut $80B in cost out of this $100B in annual spend.
The Joint Statement released last year is a great example of how the US banking regulators see their role in promoting innovation in the financial crime compliance area. They clearly stated that AI and digital identity are the future of the compliance industry and committed themselves to encourage the innovation by de-risking experiments with technological vendors and accepting failures during the pilot programs. This trend is seen across the globe with agencies like FCA, FINMA or JFSA establishing regulatory sandboxes and Fintech licenses to accelerate deployment of innovative compliance solutions in Top tier banks.
Machine learning can be applied to many different problems in AML and KYC. In particular, its biggest impact will come from more efficient and effective risk screening. Machine learning and natural language processing models, sitting on top of trusted historical data from vendors like Dow Jones, LexisNexis, Thomson Reuters, and Accuity — as well as internal client and transaction information — will be used to streamline and increase risk visibility and all forms of list screening, sanctions screening, media screening, payment filtering, and transaction monitoring. Other valuable use cases include tasks like automating workflow, aggregating and deduplicating data, or enriching entities can help make screening and investigations more efficient, but in the end it will be 80% more efficient screening that reduces the size of the haystack, allowing us to find more needles.
