IBM uncovers rapid rise of Watson in everyday banking


IBM has spent the last 6 to 8 months trying to help the global banking community to move to the next stage in banking, which it calls the Cognitive Bank, using AI to improve efficiency, reduce churn and find new business opportunities. It revealed this in a briefing this week, along with much of its findings to how to go about the process of addition cognitive computing to the finance sector.

IBM’s webinar was mostly about normal every day checking account style banking, including the offering of mortgages, counter services and new systems like mobile payments and wealth management.

The key is straightforward according to the VP of Banking Analytics, Boxley Llewellyn, who says that banking wishes to becomes more customer focused, and so wants to know its customers better.

IBM talks about behavior based customer insights (BBCI), Know Your Customer (KYC) for which it has developed systems built around its Watson AI systems, and its own special Financial Transaction Manager (FTM).

It all starts with behavior says Llewellyn, taking data that the bank has on what people spend, and what they spend it on, how they manage their money, for instance what systems they rely on, and where they travel to, what their social profile says about them. “We talk to banks and they have a lot of data, and when we raise the issue of adding internet held data on their customers, they say ‘Let’s just start with the data we already have.’”

This data can map life events, and IBM says it has listed over 40 separate life events that can change behavior and create a financial event, everything from the first ever paycheck to enrolment in college, entering a relationship, graduating, getting married, starting a business, buying a car or a home, putting their own children into college, retiring, relocating, getting an inheritance or paying for a parent’s medical care, or making a career change.

These can lead directly and indirectly to overdrafts, pay rises, health crises, product churn, attrition. Effectively banks are looking to move to a stage where they can predict and anticipate this type of event and the financial transactions that go with it, instead of describing it as it happens and having little reaction time.

There is also the possibility of radical cost takeouts. Instead of offering inappropriate products to everyone of a certain age, the banks can offer a much more directed approach to offering transaction types that are welcome, approaching the client just as his or her need arises, rather than asking them if they want help managing their retirement funds, just as they need a loan instead. This leads to a reduction in costs, but other outcomes include selling more stuff and creating loyalty.

One of the biggest issues is working out when a family is ready to churn their mortgage. If a bank can be ready with the right mortgage option that suits a change of circumstances, then it can prevent the customer casting about looking at rivals.

But all of this relies on the ability of Watson to predict behavior, the rest can be pretty much automated. IBM claims that it can now predict cashflows in bank accounts with 94% accuracy looking a week ahead and 87% looking a month ahead. It is important that this information doesn’t force the bank to do anything that looks “creepy” or aggressive. Having your bank manage call to say “I am worried you will go overdrawn this month, can I sell you a loan,” is perhaps the wrong side of this equation, and instead they must be sensitive about what they do with the prediction. Pushing out generic loan offers to customers with a high risk of going overdrawn makes it look like this is just serendipity, not your bank keeping too close a track of your finances.

IBM also says that Watson can predict 9 different life events with 95%+ accuracy and offers a 91% improvement in eliminating irrelevant offers, and has done this at scale for 3.6 million customers. The real nugget is predicting 50% better than all previous systems, when a customer might churn.

Now IBM did not give us details of how this works, and what data they are taking onboard. It’s not tough if you look at someone’s Facebook page, to work out if they are getting married, or moving house, but IBM insists that most of this is based purely on data that the bank sees, so what are you buying and is it on a credit card or a checking account or an online payment.

So if that’s what AI can achieve without external internet held data like Facebook, what more could it do with it. IBM’s sales pitch is straightforward to each bank, “What would it be worth to you to have a 6 month prediction to 90% accuracy that a customer was about to churn,” and anyone in banking is likely to say yes if you can prove that you can do it.

And given that most of the top 100 banks in the world use IBM for data processing services of some kind of other, this is IBM taking its existing brand and turning it into cash. If a brand new AI start up knocked at a bank’s door and said, give us all your data and we can help you become more efficient, many would not get past the front door.

This is especially true since IBM says that in order to get banks onboard it has to deal with lines of business manager, chief marketing officers and not just the IT division of any given bank. And the trick is to use the entire bank to create a single channel view and yet each of the outcomes will mean that Watson is indicating that bank contact needs to go through a particular part of the bank, so it’s a curious mix between omni-dimensional data collection and single channel marketing.

Someone asked Llewellyn if there were any data quality issues, and he said that only one bank has said “come back when we have fixed out data integration problems” and what he normally replies is “let’s see how successful Watson is with the data that you have,” and most have been more than satisfied with the results.

by Peter White, CEO, Rethink Research, AI Trends contributing editor