Complaints management with generative AI: adding a new layer to the mix
I have recently been involved in a process to add a new layer to a customer complaints handling process for a global bank. The addition of generative Artificial Intelligence (AI) has added new capabilities that have speeded up the process, increased the volume of complaints handled, and improved the consistency of customer experience.
The problem with complaints
Customer complaints happen in any business. No matter how good you are, sometimes things go wrong. The key is to handle complaints quickly and effectively, and provide a satisfactory resolution for the customer. Complaints should also receive the same resolution regardless of when or how they arrive, or who handles them.
Traditionally, complaints have been dealt with by employees, usually complaints handlers. This ensures that the process is human. However, with the best will in the world, it can be less than ideal. The provision of scripts means that the process is standardised, which is often better for compliance purposes. However, it can feel a bit dehumanising to customers. Complaints handlers also have limited capacity, which means that at busy times, there may be long waits on the phone or for a response to an email or web query. What’s more, they may experience fatigue or even burnout in especially busy periods.
Complaints and analytics
We have been providing analytical support for complaints management for some time now. Analytics such as intelligent decisioning systems can help by providing suggestions for ‘next best offer’. They can also take into account the value of the customer in generating this offer. For example, it may be worth providing a better offer to retain a higher value customer. Visual text analytics can provide a dashboard to help complaints handlers review the situation more rapidly.
This was the situation for this customer. They were already using existing decisioning solutions, but wanted additional capability to:
- Be able to summarise long email threads or conversations that had taken place over several contact episodes. This would mean that a new operator could quickly catch up with and understand the situation without the customer having to tell their whole story again.
- Automatically generate a reply to the customer, to improve the consistency of responses.
- Enable the complaints handler to review the reply before sending it, to ensure that it was appropriate.
The ideal solution for this seemed to be the addition of a Large Language Model (LLM). We chose to use ChatGPT for this, but other LLMs would work in a very similar way. The process is shown in the diagram below:
In summary, the dashboard sends the customer ID to the intelligent decisioning solution from SAS. This draws down the necessary information from the customer database, such as churn probability and customer value. The full text of the complaint and any previous interactions is then sent to the LLM, which provides a summary. That summary is sent to a visual text analytics application, which provides categorisation. The decisioning system recommends a ‘next best offer’, and this is passed back to the LLM to generate a reply to the customer. The operator reviews that reply and can then send it using the complaints dashboard.
It is important to stress that the LLM alone does not solve business tasks. The key is to integrate it into a large and complex process, layered into the orchestration and governance. Of five important functions in the complaints system, broadly two (summarising the complaint to date, and generating a reply) are handled by the LLM. The other three (automatic categorisation of the issue, generating the next best offer, and orchestrating the full decision-making process) are handled by SAS Intelligent Decisioning and SAS Visual Text Analytics. This is not to overlook the role of the complaints handlers, who must check the response before sending it to ensure that it is correct, consistent and appropriate.
The benefits of adding generative AI
What benefits were seen from adding generative AI? We were able to decrease average complaints handling time by 20%–40% and increase the volume of complaints handled by up to 20%. The average response time decreased by 30%–40%, and the complaint resolution time by 20%–25%. The cost of handling complaints also decreased by 8%–15%. Complaints handlers found that they could offer a more consistent customer experience, and with increased personalisation of the process. This is likely to lead to happier customers and employees, as well as lower costs: a huge win for everyone concerned.
