Automation in Banking

By Lars Hamberg, Co-founder and CEO, Gavagai 

Modern companies want to know everything about their connected clients—who they are; what they feel; who they work with, work out with, sleep with; what they are looking to buy, when, why, from whom, at what price…and so on and so forth. Most businesses want to know their clients as intimately as possible. Online companies within the travel, hotel, music and dating industries have to use advanced and automated client analytics to survive in a competitive low-margin business. They keep tweaking their systems—i.e., the online client analytics and the client offering at the center of the online business model—in very short project cycles.

Truly digital companies, such as Spotify and Tinder, regularly analyze hundreds of millions of events in order to understand their clients and to remain one step ahead of them—on an individual level. Ironically, the finance industry is the only industry I know for which the legislator has to step in and order the operators to know their clients:  “Know your client, or you will be hit with legal sanctions!” In the light of reputation risks, fines and penalties, as well as missed business opportunities in the digital age, that’s quite bizarre, wouldn’t you agree?

Banks have a lot to learn from the digital world, and they have to become much more agile. Interestingly, for many years, banks have repeatedly been told to look outside their own industry in order to adapt to a digital world, in which most activities are done online and over the smartphone.

However, most banks haven’t listened. Most banks—but not all—seem helplessly behind the curve. They generally make a lot of noise about digitalization, opportunities and about “banks being IT-companies”…but it is still mainly noise. Why is this? Maybe they were too profitable and content to do anything about this real threat to their business?

Nobody knows, but there are possible explanations and possible industry ramifications: Maybe many banks are behind because the most recent wave of digitalization and advanced online data analysis has coincided with an extremely benign trend for the operational environment of banks, in which banks are not allowed to fail and where credit has been very cheap for a very long time. Maybe it is because many banks operate in inefficient markets, with local oligopolies, high barriers and insufficient competition.

Until now, banks mainly have focused on using new technology as a shield, or as a means to cut cost, with a focus on the automation of back-office processes, advanced anomaly detection in transaction data for fraud-detection purposes, and similar cost-saving measures. With ROE (return on equity) for global banks under pressure from the increasing costs for regulatory capital, some banks are finally starting to take automation seriously.

Some banks have started using new technology as a sword, or as a means to make more money in the front-office, by producing better products or by optimizing distribution. For some time, automation by means of machine learning from structured data has been widely used in trading and in asset management. Now we see an increase in strategies leveraging unstructured data as a possible alpha source. This has coincided with the emergence of machines that are learning to read and understand unstructured data on scale. Some banks have also started using machine learning for targeting clients in product campaigns, with dramatic hikes in conversion and profitability. Within asset management, some banks have started using AI (artificial intelligence)-driven tools for alpha generation and distribution; and within retail banking, some banks have started leveraging AI-driven customer insights to increase profit margins on targeted clients by as much as 40 percent on average.

Currently a lot of industry attention centers around chatbot solutions in retail banking. Chatbots are interesting from many perspectives but have to become better at understanding natural language. Most cognitive and conversational solutions still only operate in a few languages and don’t perform well enough when they have to use a translation layer. A poor chatbot solution is neither progressive nor good for business. We can expect a major pickup in performance when all solutions operate natively across languages.

Robo-advisory is also very much in fashion and deserves a special mention. Three conditions need to be satisfied to make robo-advisory models deliver to expectations:

In the first place, models must provide advice with regard to trading in portfolio components with different investment horizons—not to be confused with the trivial task of re-balancing on a single-horizon assumption basis. Models need to take into account that individual portfolios may have components with different investment horizons. Matching individual risk profiles and needs with customized-model portfolios is neither new nor very exciting, in my view. Successful robo-advisory models will have to handle real-life situations, such as a scenario in which, for instance, a client wants to invest different portions of his or her portfolio with different investment horizons. The robo-advisory model will have to provide real advice with regard to the size limits of the new speculative-portfolio addition, beyond which the risk profile for the entire portfolio diverges “too much” from the client’s risk profile. Robo-advisory models of today rely on the concept of single horizon, which is a basic assumption with no explanatory value. In theory, it is assumed that “assets with the same risk should have the same expected rate of return”, and vice versa. This is a basic theoretical postulate, which has no bearing on real life and is easily falsified. In real life, the rate of return is the result of the capital-weighted average of investors with different and individual investment horizons and, consequently,  different risk preferences. Hence, the risk in the investment strategy depends on the time horizon and not only on the risk in the asset per se. This is very important, and for as long as robo-advisory models don’t operate with a risk concept that integrates the intended investment horizon for various components, robo-advisory will never be really useful in real-life advisory situations, in my view. Such a multi-dimensional risk concept is, however, possible with new technology, including artificial intelligence. 

Secondly, advisory models must also, ideally, generate some viable evidence to suggest that the investment advice was suitable to the best of everyone’s knowledge at the time when the advice was given. Settlements for alleged unsuitable advice—or breaches of fiduciary duties—have become a consequence of large market corrections. What is viewed as suitable at one point is not necessarily viewed as suitable in a future dispute, and history shows that no court is immune to hindsight bias. Hindsight bias is highly prevalent and works in favor of claimants and increases the risk in advisory services. Clients in the financial system are ultimately paying for this increased risk, one way or other. There is a multi-year research project looking specifically at the described challenges, and the problem described above has been solved with state-of-art AI.

Thirdly, there must be human “touch” in the advisory situation. While the machine does the actual work, the profiling and the modeling “beyond human capacity and cognition” and so on and forth, a human being still needs to handle the social aspect of the relationship. Customer-insight data shows the importance of the social aspects of the client meeting. Soft factors, such as feeling as if you’ve been seen and looked after, are very important from a satisfaction and profitability point of view. A good client experience even trumps the actual terms and conditions in many cases. Most seem to agree that hybrid models—with a machine-supported advisory service—is the most profitable approach.

The banking sector has been remarkably slow to adopt new technology, but some are catching up fast. In my view, the incumbents will not become Regulatory Pipelines (“Reg Pipes”) that are simply providing the plumbing for the more agile and innovative fintech operators, as some seem to believe. Consolidation has begun and will continue. Viable solutions find their way into the incumbents, most of which now have their own innovation labs.

Besides the regulatory plumbing and their legacy systems, the incumbents have an asset, which is very hard to replicate. Banks have unique and extremely valuable sets of historical transaction data on their clients. Using new technology, banks have the opportunity to leverage this unique asset, enabling them to know their clients better than Google and Facebook combined. When banks start using the latest technology to leverage their client data, we may see a different type of banking operation with services going beyond traditional banking business.  Banks cannot sell their data so they have to use it themselves, maybe by staying one step ahead of their clients, by automatically predicting their clients’ individual behaviors and needs, and by providing offers that you wouldn’t typically associate with a bank.