In this article, we introduce four scenarios for optimal cooperation between humans and machines when improving the digital customer experience through machine learning and artificial intelligence solutions.
"Improved digital customer interaction", "personalised offers and services", "shorter response times", "avoiding customer churn" – the goals behind many of Zühlke’s current project enquiries from the banking and finance industry can be summarised under the heading "Improving the customer experience". Our subjective assessment that the customer experience is currently the number one topic in the banking world is supported statistically: according to the Digital Banking Report , for 80% of the companies surveyed, improving the digital customer experience is one of the 3 most important strategic goals in the financial industry.
Machine learning (ML) or artificial intelligence (AI) is increasingly being used by financial institutions in projects for improving the customer experience. At Zühlke, we often use the Marketing & Sales process ("funnel") to highlight AI use cases along the customer journey. The current common form of this process consists of two "funnels", which are connected by the purchase or the conclusion of a contract: the first funnel is about winning the customer, the second is about keeping the customer and making them an enthusiastic supporter of a company.
Most of the ML/AI use cases presented are being considered or are already implemented by banks, with the current focus being on recommendation systems. In private banks, however, "recommender engines" are not generally used to acquire new customers, but rather to recommend new products in an existing investment portfolio.
Does artificial intelligence conflict with personal customer relations?
Customer orientation is a decisive competitive factor for banks. Customer relations are based on trust and personal advice from an empathetic counterpart. Against this background, the question arises whether artificial intelligence and an improved customer experience in banking really go together. The answer is yes, provided that the human being is not forgotten. In this article, we present two important success factors for AI implementation projects: the cleverly combined use of humans and machines, and the building of trust in the implementation of AI projects.
Four scenarios for optimal cooperation between humans and machines
Currently, the automation of processes by means of Artificial Intelligence is a much-discussed topic. However, we believe that maximum automation, especially for processes with customer interaction, should hardly ever be the goal. Rather, the aim is to make optimal use of humans, machines and their respective strengths to significantly increase the quality and efficiency of the process. For example, the machine should take over tasks that are repetitive or difficult for humans to grasp; even where very short reaction times are required, machine use should also be considered. Humans, on the other hand, should be able to use their creative, empathic and communicative abilities as well as their intuition and, of course, common sense at all times.
With regard to the process flow, tasks should be clearly assigned to either humans or computers. The following four scenarios are possible (according to ):
Scenario 1: The "transfer of baton": a process is broken down into individual steps, which are taken over either by humans or by a machine, depending on suitability.
Scenario 2: The process is managed by the machine, but at the end a human being decides whether the result of the process is used.
Scenario 3: The human being drives the process, but is enabled or trained by AI (AI shows possibilities, provides insights from statistics, etc.)
Scenario 4: Iterative process loops in which humans and AI alternate. Rather a special case, but one which is already used today in the field of architecture and art with generative algorithms, for example (the algorithms produce contents, such as pictures or architectural sketches).
AI scenarios using the example of personalised investment recommendations
Let us look at concrete examples of how these AI scenarios can be implemented in practice. For this purpose, we choose the topic "personalised investment recommendations", which we have implemented for various private banks in differing forms. All AI use cases have in common that a human relationship manager (RM) should provide tailor-made investment recommendations to their customers.
A first implementation variant of this topic is that an algorithm (not machine-learning, but rule-based) monitors the portfolios of customers according to risk criteria and identifies positions in them that should be sold. The cash amounts thus released are to be reinvested – the choice of investment instruments proposed to the client is made by the relationship manager. Thus, we have here an implementation of scenario 1 above, the handover of the baton.
In a second case we implemented a recommendation system: research employees in the bank are continually writing "investment stories", i.e. reports on possible instruments for investments on a specific current theme, e.g. ecologically sustainable investments. We have now developed an algorithm that automatically provides the relationship manager with lists of clients who are potentially interested in a particular investment story. The RM ultimately decides whether or not to recommend the story to a client. We have thus implemented scenario 2 above: the algorithm carries out the recommendation process, the human being decides at the end of the process whether the result will be used or not.
However, this operation can also be implemented using scenario 3. This means that the RM leads the recommendation process, but is supported by AI, which provides the RM with constantly updated analytical evaluations of the investment preferences of different customer groups, or the success statistics of past recommendations for each individual customer.
The examples show that a specific goal – in this case to provide personalised system recommendations – can be implemented in different use cases and in different scenarios of human-machine interaction. In order to identify the optimal variant for a specific application situation, we recommend that the client carries out an initial project phase – we call this "Vision & Scope" – in which the goal and the constraints of the AI use case are clearly defined. Together with the intended users, we evaluate different versions and application scenarios. The methods of "Design Thinking" are very useful for these tasks. The Vision & Scope phase should be moderated by an experienced AI expert who is able to bridge the gap between technology, business, and human users. It is also his job to build trust in the resulting solution – which brings us to our next focus topic.
A critical success factor: building trust in AI solutions
In our consulting and implementation projects we have noticed that building trust in AI solutions is increasingly becoming a success factor. It is decisive in determining whether a solution can generate value right across the board in a company. Especially in projects that involve changing and developing entire companies into data-driven organisations, this point needs to be given sufficient attention.
Trust in AI solutions should be built on two foundations. The first foundation involves designing the solution, or the system or tool that is to be developed, in such a way that its use generates trust. As a higher-level "design guideline", the Ethics Guidelines for Trustworthy Artificial Intelligence issued by the EU can be used . It makes general statements on ethical requirements for AI with regard to:
- Human control over AI
- Reliability and security
- Data protection
- Social & ecological well-being
The second foundation for building trust in AI is the actual implementation projects. From our "Best Practices" we now outline the most important points, which are in addition to the use of a bridge-building AI expert as discussed earlier.
A vision that is communicated company-wide at the beginning of the implementation projects should explain the "why": Why does the company want to use AI? As well as answering that question, this vision can and should anticipate employees’ potential resistance and worries. Another important element is training: companies that use AI for the first time should gradually train their employees, so that an understanding of the possibilities – and the limitations – of using AI can be developed throughout the company.
An agile, interdisciplinary development team is our standard setup. However, one of the most important measures for us is the joint development of the solution with internal "customers", the users. We involve user representatives as well as the business in the design and development of the solution from the very beginning, which significantly promotes the acceptance of the solution.
In addition to technological and business aspects, people should be brought into the focus of AI implementation projects. A full consideration of the topics "optimal combination of human and machine strengths" and "building trust in AI" are important success factors for AI use-case implementation projects that have the goal of improving the customer experience in banking.
 Digital Banking Report, February 2020. The Financial Brand
 Gartner, 2019: “The Future of AI and the Gartner AI Hype Cycle”
 EU guidelines on ethics in artificial intelligence: Context and implementation. European Parliamentary Research Service (EPRS), Sept. 2019
Dr. Philipp Morf holds a doctorate in engineering from the Swiss Federal Institute of Technology (ETH) and holds the position head of the Artificial Intelligence (AI) and Machine Learning (ML) Solutions division at Zühlke since 2015. As Director of the AI Solutions Centre, he designs effective AI/ML applications and is a sought-after speaker on AI topics in the area of applications and application trends. With his many years of experience as a consultant in innovation management, he bridges the gap between business, technology and the people who use AI.