A decent-sized bank recently held what it called a “Digital Summit,” bringing together senior execs from various business units and functions to discuss the bank’s opportunities regarding digital technologies.
Predictably, a good portion of the discussion involved customer data and analytics. Well, to be more specific, predictive analytics. Actually, to be even more precise, next best action models.
A rather large fintech vendor–who I bet most of you have heard of–presented on the topics of analytics and next best action (not “offers,” mind you) models. The presentation contained more BS than a field of cows.
This particular vendor could not refrain from spewing more numbers regarding the improvement in response and conversion rates its clients have seen from deploying their technologies–60% lift here, 54% improvement there, 2x this and 4x that. Absolutely no substantiation of the claims.
Fortunately, I don’t think many of the execs bought into it. At least, that’s the sense I got, when the president of the retail banking group asked me at the end of the day to sum up my thoughts. I told them that I was afraid that they were placing way too much faith in just one type of analytical model–next best action–at the expense of other models, at the expense of more fundamental data management capabilities, and more importantly, at the expense of higher business priorities.
The look on his face–as well as on a lot of others’–told me my fear was actually unfounded. They were looking for confirmation of what they silently believed.
Why was I down on next best action? The vendor’s pitch was basically: Get a 360 degree view of the customer, and then you’ll have the data you need to determine where to go with the relationship, and that, at any interaction, you’ll be able to recommend what the customer’s next best action–not just offer–is.
My take: Nonsense. Total freaking nonsense. First of all, if you integrate all the data you have about your customers across lines of businesses and channels, what you end up with is all the data you have about the customer. What about the rest of their financial relationships? What about their past history? What about their future goals? What about…a thousand other things that you really need to know in order to make the right recommendation to a customer?
The vendor here demonstrated absolutely no understanding that, when a next best action model is developed, the recommended actions must be pre-defined. What that means–and I had to explain this to the head of the retail banking division–is that there’s a pretty good likelihood that many ancillary services that the bank provides could be left out of the model because of an insufficient volume of data about them.
Reality is that there can be an infinite number of “best next actions” a customer could take. A bank lacking a well-honed, well-developed, long-history of analytics deployment isn’t a good bet to get it right on the first try. Especially not working with a vendor that provides technology, and not marketing analytics services.
The underlying problem here is…well, there are probably a bunch of problems at play here. Let’s focus on one of them: The definition of analytics.
At one of the breaks during the day, I was chatting with one of the attendees who hit the nail on the head when he said: “I don’t think we’re on the same page regarding what analytics is.” Bingo. He went on to say: “We’re limiting the discussion to just advanced statistical techniques.” Jackpot.
In his book Competing on Analytics, Tom Davenport included what he called an Analytical Maturity Model:
I wrote about this three years ago, pointing out that the model is wrong. There is no shortage of organizations that do predictive modeling and have optimization models that do not “compete on analytics” or have anything that comes close to a “competitive advantage.”
But there is one aspect of this model that should be quite useful to banks–and specifically to the one I’ve alluded to above. It’s the spectrum of elements that comprise analytics.
What Davenport got wrong (sorry, Tom)–and that bankers need to understand–is that you can’t place a value judgement on the various elements. That is, standard reports are not “less valuable” or “less important” than ad hoc reports, which, in turn, are not less valuable or important than queries, alerts, or advance statistical techniques.
If running an ad hoc report gives you the information you need to make a great decision, taking six months (and who knows how much cost) to build a predictive model that helps you come to the same decision is not better.
What Davenport got right (gotta give the man a little credit, no?) was that, in building an analytics capability, you don’t start in the upper right quadrant of his model. The lack of understanding on this point on the part of the FinTech vendor was just another thing that ticked me off about them.
If they’re so good at analytics, I would expect them to understand two things: 1) That there’s a path to go on (from the left side of Davenport’s model to the right), and 2) That the key to analytics success is not in integrating the data or deploying technology that enables sophisticated analytical techniques, but in using the data in the context of the organization’s sales and support processes.
This is why I told the bank I was fearful they were putting too much faith in next best action models: I didn’t think they knew how to use them. Will their branch tellers really be willing to make product suggestions? Will their branch sales people feel comfortable doing what the model suggests if their gut tells them otherwise? Is there any consideration of what the contact cadence should be? That is, if an offer is shown when somebody logs into their account on Monday and isn’t accepted, should we repeat the offer when she comes into the branch on Tuesday? Should we move to a different offer–oops, I mean action?
Bottom line: Successful deployment of analytics means business process change. Some bankers–and a lot of vendors–need to stop thinking of analytics as some kind of panacea that’s going to magically improve marketing performance.
Redefining analytics in banking means thinking about it as “the effective use of data” and not narrowly as just advanced statistical techniques.