In 2008 Andy Kirby, an employee of the highways department in Swansea, Wales, had to design a road sign that read “No heavy vehicles, residential area”.
Although less than 30% of the country’s population speaks Welsh, the law indicates that all signs must be bilingual, so Kirby sent the phrase in English to an official translator and, upon receiving the answer, sent it to print.
The phrase ended up being an early meme as “Nid wyf yn y swyddfa ar hyn o bryd. Afonwch unrhyw waith i’w gyfieithu” means “I’m not in the office right now. Please send any translation work”
This is a hyperbolic analogy to how analytics works in the traditional financial industry. Those who decide do not speak the language. Those who speak the language do not decide. The one who sees it from the outside cannot believe it.
I was fortunate to work in Banking for more than 20 years, doing so in five different countries. However, the only factor that distinguished my career was accidental. I think I am the only element at the intersection of three apparently exclusive sets. 1) People who worked in a bank branch, 2) people who were part of the management committee of a bank, and 3) people who know what heteroscedasticity is.
Banks are the companies with the most availability of information about their customers. And they have been for decades.
To be more precise, 10 years before Google or Amazon existed, the average bank could already know where a person lived, for whom and in what physical place they worked, where they moved, how much they earned, who they lived with, how much they spent on each type of trade, whether his savings increased or decreased, the goods he acquired or insured.
But I say “could know”, because he did not know.
I have the personal and somewhat domestic definition that the analytical area of a bank should have the sole objective of “increasing the value of the client portfolio while pretending to know them”.
And even in that the banks have failed. We do not accurately measure the value of the portfolio. And we fail to credibly pretend that we know our customers.
There are numerous reasons for this situation, and many can be explained by the genesis of data processing in the financial industry.
An example of them is the so-called “The Plaza’s Internet Problem”. In essence, the best hotels in New York adopted the internet before anyone else, but they were tied to long-term contracts and large investments in physical infrastructure that caused, after 10 years, to be the places with the worst internet in Manhattan.
In the same way, many banks were pioneers in the use of early and inefficient technologies, which they accompanied with a disproportionate and irrecoverable investment that prevented them from quickly migrating to the most appropriate ones.
Another set of problems are those that derive from the “Flying Airplane” of the financial industry.
The technological agenda of a bank has always had an almost exclusive focus on operational continuity. In which “the plane continues to fly”.
And this is a major challenge for two main reasons: First, banking is the sector with the most intense relationship with customers. As we know, intensity is a measure of periodicity and importance of interactions. Although there are industries, such as telephony, that have more frequent relationships with their customers, the criticality of each interaction is lower. Dropping a call is an inconvenience. The disappearance of money in all my accounts is a potential heart attack.
Therefore a bank can never stop, and must keep its stack of legacy systems running permanently. Each new element in that set increases the complexity of the system and its operation.
Second, that stack will always be intricate because banking is a flood industry. Almost all the leading banks are the sum of dozens of pre-existing institutions, bank mergers, financial incorporations, insurance company acquisitions.
Operational continuity implies, then, ensuring that “the squadron of aircraft continues to fly, ideally without colliding with each other.”
In short, there are at least a dozen valid, multifactorial reasons why banking has always been deficient in customer analytics. But the most interesting motif, at least from the point of view of its consequences, is what we at N5 call “The Welsh Road Sign”.
“The Welsh Road Sign”
The Welsh cartel problem is an imperfect analogy to that of banking analytics because it assumes that only two languages exist in this domain as well.
But the tower of financial babel is higher.
Technicians don’t understand managers, managers don’t understand technicians, channel teams underestimate both, customers criticize everyone.
And I allow myself to mention some anecdotes that illustrate it.
I am going to start with a pre-Cambrian one, from the time when intelligence teams still had a greater focus on human channels than on digital ones, because I think the conclusions are still valid.
Many years ago I interviewed the head of CRM for a multinational bank. He was a Doctor of Statistics, a university professor, and highly respected by his team for his extensive technical knowledge.
He was also very friendly and answered all my questions thoroughly.
One of them was “what projects or advances are you most proud of in your career at the bank”. He explained to me that, using Machine Learning (a revolutionary technique at the time), he had developed a very robust model that made it possible to predict with high reliability which customers would take out a consumer loan over the next 30 days.
I was very interested and asked him how long he had been using it, to which he replied more than a year.
I also asked him how the model was applied.
He told me that the branch executives received a list of clients among whom they could predict that 90% of those who would contract a loan were found.
It seemed very auspicious to me, so I asked him the most obvious question: “how big is the list that you send?”. He was silent and I noticed that he had never asked himself that question. He told me “I don’t know… it’s variable, it depends on the portfolio… I can look it up. It is important?”. I didn’t tell him, but from my point of view that was the only thing that mattered. Pointing to a list and saying “90% of those who will hire a loan are there” has no merit if the list is not very short.
For example, I can know precisely that 100% of the people who take out a loan this month are already on a list. The list is called “inhabitants of planet earth”.
Finally he looked up the data and showed it to me.
They were roughly as follows: An executive had a typical portfolio of 800 clients. In an ordinary month, 18 of them took out a consumer loan. The base that the executive received to manage had 280 clients, and collected 90% of the “successes”, therefore, 16 sales.
From my long experience as a branch executive I can assure you that a list of 280 clients to reach the goal of a single product (the typical incentive system has 8) is exactly as valuable as the aforementioned list of inhabitants of the earth.
I imagined, therefore, that the level of effective use of the model would be very low.
I asked him, and that fact was very clear to him. “This is the big problem we have. Channel people don’t understand anything. Only 0.8% used this campaign last month.”
I made a quick account. The bank had 500 executives, 4 had used the list. 9 people worked in the Models area.
The second anecdote, and I promise that in the end it will be clear how they are all related, is much more recent.
I was talking to the Chief Innovation Officer of a mid-sized bank. He had recently been hired from another industry and was finishing defining a strategic plan to present internally.
He was a creative person, he understood user experience and he had correct intuitions regarding the usual “pains” of the financial client. But he had no technical grasp of the analytics world, and he had only a few weeks of banking experience.
I asked him if he could summarize his plan for me in three concepts.
He said to me, briskly holding up three fingers on his right hand and pausing dramatically: Artificial Intelligence, Deep Learning and Blockchain.
The probability that this is a factually correct answer is virtually nil, unless the question was: can you list three buzz words?
A strategic plan that starts from specific solutions is comparable to a doctor who chooses three medications from a medicine cabinet and says “I already have what I will prescribe for the next three patients who arrive”.
The third anecdote is perhaps the most interesting.
The protagonist is the CEO of one of the largest banks in the world, and one of the most brilliant people I have had the opportunity to interact with.
We will see it in the next entry of America Digital News
