12. April 2019

“AI only works with a strong business case”


Artificial Intelligence // AI is gradually making inroads into banking.
Mario Crameri, Head of IT & Operations Swiss Universal Bank at
Credit Suisse, is the brains behind it. In our interview, he provides
an insight into the requirements that have to be met.

The hype surrounding artificial intelligence (AI) is nothing new. Similar expectations of technology appeared back in the 1960s and 1990s, but were abruptly dashed. This time, however, Mario Crameri, Head of IT & Operations Swiss Universal Bank at Credit Suisse, does not anticipate another AI winter, as he explains in discussion with ti&m. Improved computing performance, AI algorithms, and above all the enormous quantities of data available mean that the tipping point for an AI breakthrough has been reached. The time has come, he says, to engage with the technology.

Credit Suisse leads the way with AI
AI is already present in many everyday devices and services, whether it’s smartphones, TV boxes in the home or services such as Netflix. AI is in widespread use in many sectors, for instance in expert systems in medical technology, like IBM Watson Health. Even agriculture is finding applications for AI and increasing revenues by determining optimal pesticide usage. Insurers and above all banks continue to be somewhat cautious. This is primarily a result of the value they place on data security. But for Crameri, banks are actually predestined to use AI. After all, they handle huge volumes of digital data. An extremely high proportion of bank processes could in his opinion be supported or even performed by AI in the future. Credit Suisse is an AI pioneer amongst Swiss banks. An initial pilot project for the customer advice support team was launched in mid-2017 and successfully rolled out at the end of 2018. Credit Suisse chose this area because it had vast amounts of data that could be used to train the algorithm. Crameri also saw the potential to introduce AI relatively easily there. First, the bank prepared a traditional proof of concept. Then it was a question of deciding whether to use open source or a vendor product for development. The bank eventually decided on the latter.
The advantages of this were that there was no need to set up in-house support. In principle, all that was left to do was to build the gateways and interface, and to feed the algorithms data. For this purpose, however, Credit Suisse decided on an open design, which, as Crameri says, offered the best of both worlds.

Introducing AI requires change management
Credit Suisse also decided that initial deployment would be primarily internal, so as to involve users actively and facilitate change management. According to Crameri, AI is bringing about enormous reductions in workload for payment support staff. Previously, they had to spend a large part of their time answering the same standard questions. This has now been taken over by AI, enabling staff to give their time to the really knotty problems. Feedback from staff has been “predominantly very positive”, according to Crameri. But without investing in change management to convince staff of the benefits, this would not have been the case. Moreover, it was important to ensure that the solution provided the promised added value from the start. A half

“The biggest mistake is to roll out a system without making the necessary support available from day one.”

finished solution would have led to a great deal of frustration and jeopardised the acceptance of the entire project. “The biggest mistake you can make is to roll out a system without making the necessary support available from day one. Because the system cannot do everything at the beginning – it is still learning,” emphasises Crameri. Data quality is another key for the success of AI projects in his view. Credit Suisse has very well-maintained data, partly as a result of regulation. Data must be 100 per cent complete and accurate, says Crameri, or else algorithms will not learn correctly. Crameri also takes the view that when businesses are introducing AI, they should choose an area where the potential of AI will quickly become apparent. They should also appoint sponsors to take the project forward. By this Crameri means cheerleaders on the executive board as well as users who are favourably disposed towards the technology. Ultimately the introduction of AI is not a question of technology but of changing the operating model of an area.

AI is often only part of a label, not an integral feature
Crameri observes that many solutions currently carry an AI label without actually including any AI. Many chatbots are actually only more sophisticated FAQs or call waiting loops, he says, because they process a fixed set of questions. To qualify as AI, a solution must be based on machine and self-learning. However, this applies only to a handful of chatbots today. To describe a system as AI, it is not enough to feed a machine with questions and answers, emphasises Crameri. The same is true of roboadvisors. They generally optimise portfolios on a procedural, linear basis rather than using AI. A further development would be an application that learned independently on the basis of all potential transaction data and improved portfolios. However, robo-advisors are still a long way from being able to do this.

AI is still inferior to humans in many cases
The biggest overhead with AI is still training. Google needs 300 million images, for example, to enable it to differentiate a dog from a cat. A young child only needs to see a few dozen images to tell the animals apart. In this respect, humans are still clearly superior to artificial neural networks, says Crameri. And it will be many years before computers outstrip humans. “We are right at the beginning of this development,” says Crameri. The benefits of AI will only become apparent when artificial algorithms are able to train each other. For example, when a robot has learned to recognise certain rules, such as rules for assigning activities to specific areas, and is then able to use this information for other purposes, either to undertake some of the activities itself, or to pass on the information to other robots. One day, it will be possible for learning rates to increase exponentially in this way. But that day will not be today or tomorrow; these processes are still far too complex for machines.

The business case decides, not the technology
AI has enormous potential for banking, according to Crameri. Inspired by the initial AI project, Crameri’s team are constantly coming up with new use cases for AI. The banks, including Credit Suisse, therefore consider themselves as being on a long journey towards AI, one which is far from over. “The banking sector can only transform successfully if it uses technology wisely. AI is a key technology in this context,” argues Crameri. However, businesses should not make the technology their primary focus. The underlying use cases are much more important. Because in the final analysis users and the business have to see added value, whether it carries an AI label or not. Structures should therefore be optimised so as to prepare them for the use of new technologies. As with any transformation, the real challenge is change management. Added value must be clear both to those concerned and to the company as a whole. “We want to deploy AI chiefly to achieve efficiency gains and improve service levels, both for staff and customers,” says Crameri.

Dr. Mario Crameri
Dr. Mario Crameri

Mario Crameri has over 20 years’ experience in a range of managerial positions in the finance sector, including with Credit Suisse in Zurich and New York, and with the private bank Julius Baer. His primary focus has been strategy and business development, digitalisation, and application development. Mario Crameri has a Master’s degree in Business Information Systems and a Doctorate in Information Technology from the University of Zurich.