Using AI in banking customer service and sales may not deliver the expected benefits

Even before the arrival of artificial intelligence, banking had gone through decades of major development and automation, bringing clear benefits to customers. The same development will continue with AI, and much of the work currently done by people will change in the coming years. The real question, however, is which work should be automated.

For banks, customer service is one obvious opportunity. The more questions a robot can answer, the fewer people are needed and the cheaper it becomes to serve each customer. For simpler tasks, this may make sense. If a person wants to know when their bank card will arrive or where to find an account statement, it may not matter much whether the answer comes from a person or a machine.

But there is one challenge with financial matters: AI may be smart and fast, but in this field, technology alone is not enough. Trust is also needed, and trust is built over time.

Taking out a home loan, financing a company or falling into payment difficulties affects a person’s life in a different way from most everyday services. In these situations, there is not always a predefined question and one correct answer. The customer may want to explain their situation, compare different options or simply understand why the bank reached one decision or another.

That is why I would be rather cautious about the idea that the best bank of the future is one where customers have as little contact as possible with bank employees. In fact, the opposite may happen. If most technology becomes equally available to banks, human expertise and a good customer relationship may become an even more important differentiator than before. It is easier for a competitor to copy an AI model or a good self-service solution than a trustworthy adviser.

At the same time, banks have many areas of work that customers never see and that people do not necessarily need to do themselves. One fairly typical example is searching for information. The financial sector has a large number of regulations, internal rules, guidelines and procedures. They are necessary, but every day someone still needs to find answers from them. Often this means that one employee asks a question and another, highly qualified employee searches for the relevant section.

In 2025, Coop Pank’s internal support specialists were contacted on average 300 times a month. To reduce the time specialists spent searching for information, Coop Pank introduced an internal support chatbot. Since the launch of the internal chatbot, employees have submitted nearly four times more questions to the robot each month.

This shows that by automating back-office work, the workload of specialists decreases on the one hand, while on the other hand, making it easier to ask questions means that people look for answers more often than before. With the help of AI, employees can also search for answers to questions that previously took up their own working time. According to our assessment, the introduction of the internal support robot reduced by around 80% the manual work previously spent on resolving standard internal support questions.

This is a different way of thinking about automation. The goal does not have to be for people to ask fewer questions or receive less help. They can receive more help, while specialists have more time for cases where simply finding the right paragraph in a document is not enough.

The most expensive time is lost on small things

The same phenomenon is familiar in IT development. In larger companies, small development tasks are constantly being carried out, and a very large share of the time may be spent before the developer even starts developing.

The person making the request describes what they need. The developer reads it and asks for additional information. The requester clarifies. Questions move back and forth via email or another communication channel, and in the end, expensive development resources are used for work that has little to do with programming.

This is a part that AI can take over quite well. Coop Pank has introduced an agent that asks the requester the necessary questions and, based on the answers, prepares structured input for the developer. The development decision is still made by a person, but they can start working on it earlier. This may seem like a small thing until you start adding up how many people deal with these kinds of minor interruptions every day.

This is probably where the next productivity gains from AI will be found in many companies: in fairly unexciting places. In other words, AI will not take over the work of entire departments. Instead, its role will be to automate hundreds of small tasks that currently take up people’s time simply because no better solution has existed so far, or because organisations are stuck in old habits.

AI should not take the lead in customer communication

Today we too often see situations where the human role in customer communication has been fully replaced by AI that offers only standard answers to common problems, but is unable to think along with special cases or offer a suitable solution for the customer. Yet when it comes to bigger decisions, we need a person on the other side to talk to. To use an analogy from healthcare: AI may help a doctor analyse an MRI scan or detect anomalies in a large data set, but when it comes to making a diagnosis, prescribing treatment and making complex decisions, we need a person we can look in the eye and whose judgement we can trust.

In banking, the situation is quite similar. When making major financial decisions, customers need a person in front of them who understands their situation and, along with giving a recommendation, also takes responsibility for it.

However, AI can be used as a tool to support customer communication. In sales, for example, AI can already help a bank employee compare products, find the necessary information, prepare for a conversation or draft a follow-up contact. This allows the customer service employee to quickly access relevant and important information, while still communicating with another person and checking the facts.

A customer service employee who spends much of the conversation searching for information is unlikely to be able to provide good service. A better result may come when the service employee or salesperson can access the information within seconds and use the remaining time to answer the customer’s questions.

Banking differs from many other fields in that it cannot be based on the assumption that it is enough for AI to be right most of the time. Data security, verifiable sources and human responsibility will remain important even when language models become much better than they are today.

That is why the success of AI adoption should not be measured only by how much working time or staffing resource it saves. A much more meaningful question is what the savings are made at the expense of.

If AI reduces the time a lawyer spends searching for a document, a developer spends asking for missing input, or a customer service employee spends gathering information, then AI has been implemented in the right way. But if, in the name of efficiency, it becomes increasingly difficult for customers to reach a real person, the savings may be short-lived and quality may suffer over time.

AI gives banks the opportunity to make work faster and more efficient, but strong customer relationships are not created by algorithms. Even in the age of the smartest technology, trust, empathy and the ability to truly understand people will remain values that only human beings can provide.

The article was originally published in Postimees.