Leveraging AI to improve the customer experience

Technology is continually improving the customer experience in the insurance industry. One area that has been a problem, not just for the insurance industry but for the broader financial sector, is unstructured data, and in particular, text documents and emails with attachments.

Automation has struggled with unstructured data, but now artificial intelligence (AI) is being used in pilot projects in the insurance industry to tackle the problem and process unstructured data, with the aim of improving both the customer and the employee experience.

The email challenge
The number of emails being sent in the sector, between customers and brokers and insurers, to manage for instance international insurance programmes is steadily increasing – driven by growing complexity and regulatory pressures. In order to manage this efficiently, the sector needs to make use of technology as an intelligent augmentation. The challenge with emails is that the sender has no idea whether it has been received, whether it is being worked on, or when an answer might be expected. It is like sending an email into a black hole. But with AI, the time taken to answer an email effectively will be reduced drastically because the email will reach the appropriate person or team much faster with all the associated data (eg attachments) structured and prepared.

In the past, it was almost impossible for technology to understand unstructured data. However, this has been made possible in the last few years as a result of the maturity and availability of AI experts, algorithms and greatly increased computing power, which has become available at a cheaper price. As a result, AI solutions can read, scan and understand emails and attachments, extract information and, based on the classification of the email, flag it and create a workflow or forward it to the correct recipient.

Based on the content and attachments, the AI engine can for example understand who the customer is, the broker, the line of business and the subject of the email. With this information it can then check the insurer’s systems to find all the information needed to define a label, possibly enrich the text and forward the email to the correct person.

AI can also summarise the content of the email and any documents attached, and prioritise the email by deciding if it is time-critical, for example an urgently needed insurance certificate. It can then create a workflow or perform/trigger any other task and send a request, together with the customer information, to the right team with the right priority.

Learning from feedback
Perhaps the most important element of AI in this context is the ability to learn over time, using feedback from humans/employees. If the AI engine is not sure that it is right, it will always ask back to a human to verify the decision that it has taken. The human has the right to overrule the AI engine and every time this happens, it leads to an improvement of the system. In the end, it is always the human making the decision where the AI is not certain enough, and in this way the AI learns and enhances the correct decision-making. Basically, AI solutions have become a ‘smart companion’ – it is about intelligent augmentation, ie human and machine working together seamlessly. However, initially, a large amount of labelled data is needed to teach the AI engine how the system works.

The advantage of all of this is that by using AI-based augmentation, insurance service teams can significantly reduce efforts for non-value-added work and spend more time with customers working with them and helping them, rather than spending time looking up data in systems and keying it from one to another, which can be done faster and more effectively by AI.

For example, the AI can compare two versions of a document and summarise any differences in seconds – saving all involved stakeholders from having to read through large documents.

The end goal is to automate the processing of emails and its back-end integration and at the end of the day to reduce the time taken for the customer and broker to receive an answer. Another goal is to improve the data quality and to enhance not just the customer experience but also the employee experience, by avoiding re-keying and repetitive tasks and allowing them to serve the customer in a quicker and more efficient way.

Current projects
AI-based solutions around processing emails have a high potential for scaling i.e. they are transferrable to projects that can be rolled out internationally. At Zurich, we focus on use cases linked to email and document classification, triage and routing, as well as integration into the back end for key underwriting processes. And currently we are working on the main languages used for international programmes.

But AI technology could also be used to translate policies – so if a risk manager in the US opens a policy from Hong Kong, it could be translated from Mandarin into English by AI technology. This is something that we are exploring at the moment.

We have started to use AI to process emails within several European countries, and further countries and regions will follow next year. It is important that the AI is retrained for each country, using specialists, working hand in hand, and learning by experience and feedback. This is crucial and it is vital that humans support the project to ensure that the AI is able to learn and improve to provide the best possible service, and this must be integrated in an easy and efficient manner in day-to-day tasks, in order not to increase workload.

The industry is still in the early stages when it comes to exploring AI technologies. It’s important to test different use cases and work very closely with the respective business units, to make sure to implement technology that improves processes and customer satisfaction and supports our employees to work more efficiently.

Contributed by Joel Agard, lead digital BA, digital and new technology, international programmes, Zurich Insurance Group

Back to top button