Expert Note

How artificial intelligence is revolutionising client contact
Banks and insurance firms can benefit from artificial intelligence (AI) in many ways – for example by introducing virtual assistants for their contact centres. This leads to significant cost reduction, increased effectiveness and efficiency and higher levels of client satisfaction.

ELCA recently started a number of AI projects for Swiss financial institutions. In this interview, Dr Yves Burki, Head of Business Line for Financial Services and BI & Big Data at ELCA, speaks about the concrete impact AI is having on business, its future potential and how the market is evolving.

Yves Burki, Head of Business Line BI & Big Data and Financial Services

What are chatbots and virtual assistants/agents?
Virtual assistants/agents are computer programs you can interact with to request information or perform actions. As clients are interacting with a robot instead of a human, the agents are referred to as artificial assistants or chatbots, “robots that chat”. One of the first well-known consumer chatbots was launched a few years ago by Ikea who used an online assistant called Anna; more recently, Microsoft released Tay, a chatbot which learned from its users’ conversational style. Nowadays, the technologies have evolved considerably, so you can also interact with virtual agents by speaking to them e.g. via mobile phone. 

Virtual assistants use natural language (NLU) techniques to understand and process requests and come up with an answer. Largely based on machine learning, NLU belongs to the broad field of AI and offers many new opportunities in the area of client contact and customer service for companies, particularly for financial institutions. ELCA has invested a great deal in expanding its expertise in this area and developing a team of specialists.

What advantages do artificial or virtual agents offer?
Both the organisation using virtual agents and the clients interacting with them benefit greatly from this technology. Organisations can significantly reduce their operational costs in terms of customer service or call centres, as these costs are mainly based on the people employed and time spent on the requests – and there are often recurring standard requests that could easily be processed by a virtual agent. The good thing about virtual assistants is that they can use machine learning techniques to improve continuously based on experience from previous interactions. All this, together with the fact that organisations can reduce the number of people they need to train for client-facing roles, means additional savings for companies. 

Therefore, the return on investment of virtual assistants is high and very easy to calculate: if you move x% of your current requests to a virtual agent, you save the corresponding amount of time and work needed in your contact centre. 
In addition, virtual agents also offer advantages to clients, who benefit from instant service, around the clock, without having to wait in a queue – while enjoying consistent levels of quality in the service they receive. Last but not least, virtual agents do not suffer from fatigue or mood swings and have endless amounts of patience.

The return on investment of virtual assistants is high and very easy to calculate.

Aside from customer service and contact with clients, in what other areas can financial institutions benefit from virtual assistant technologies?
Virtual assistants can be used in other domains as well; one example is IT support teams answering employees’ requests. But besides the “conversational” aspect offered by virtual assistants, NLU techniques in general offer a large number of promising further uses for financial institutions. To just name a few:

  • NLU allows you to automatically “understand” what is published in your information sources and carry out actions in certain circumstances, for example informing somebody about a transaction and even executing it.
  • You can add metadata to archived documents in order to improve the relevance of enterprise search.
  • The technology can be used to automatically verify documents (or other textual information sources) in specific processes, for example those involving internal guidelines or compliance with specific regulations. This is very useful whenever a significant amount of time is spent on manual document verification, for instance in trade finance or legal departments.


The finance sector is in many aspects quite sensitive. How do clients react when they talk to a virtual agent instead of a human?
On the one hand, people are already aware of and used to the fact that companies use computer programs for client-facing roles – at least to a certain extent. So when you call a contact centre, for example, you often talk to a machine first that asks you some questions to categorise your request more precisely. In terms of customer satisfaction, a leading virtual agent solution called Nina from Nuance – which, incidentally, we are currently implementing in a Swiss bank – has reported very high rates. While the first three figures in the graph demonstrate the positive impact of virtual agent solutions on client satisfaction, the last figure of 98% repeat users proves the great potential of virtual agents in terms of standardisation and cost savings – and the benefits they can offer banks.

Implementing such a solution requires specific expertise, mainly in the NLU domain.

What should companies take into account when considering virtual assistants? 

Implementing such a solution requires specific expertise, mainly in the NLU domain. This includes precise intent categorisation - understanding why the client contacted the bank and what they want to do - and entity extraction – i.e. recognising useful parameters such as dates and names to fulfil a certain task. Since most of the solutions currently on the market have not been available for very long, experts are required to evaluate the best solution for each specific context.

There tend to be two types of solutions:

  • Framework solutions are the most established and include the previously mentioned Nina solution. These are often cloud solutions allowing users to define the conversation logic from end to end. In a first step, users specify examples of requests corresponding to the intents they want to support; in a second step, solution designers define how each intent is to be handled by configuring a tree-like workflow. Note that these solutions require companies to predefine all the intents a system is supposed to address, which involves a considerable amount of work. ELCA has developed additional NLU components that complement the existing framework solutions so that they are able to handle requests that have not been predefined.
  • Application programming interface (API) solutions are designed to be integrated into a global solution as specific modules. An example of this technology can be found in Microsoft’s Cognitive Services, which feature an NLU module called LUIS. Like the framework solutions above, API solutions can also be programmed for business-specific intents and with specific vocabulary. 
  • Because both approaches are still in the early stages, they have certain limitations. For example, they sometimes fail to preserve the context of a conversation, forgetting what the user mentioned a couple of minutes earlier. In addition, due to their off-the-shelf nature and limited configurability, they tend to assume standard linguistic expressions are used, which can be problematic when dealing with clients who use certain terms or dialects, for example in Swiss German.

Furthermore, using these cloud-based solutions involves sending potentially sensitive information outside Switzerland, which may rule the solution out as an option altogether for many Swiss banks. That is why we are currently working to develop a fully Swiss-based solution.

Personal Information
with prefix, e.g. +44 or +49
I would like to be contacted by ELCA
Would you like to receive the ELCA newsletter?
Please specify your request
If you would like to add documents to your request, please attach a file here.