Automation, as a buzz word, is closely tied to Robotic Process Automation (RPA) tools. Most applications of Robotic Process Automation (RPA) tools relate to repeatable back office processes which work on structured inputs. However, it is common understanding that more than half of the work in an insurance company’s back office gets done through an incoming communication, either in form of a call, e-mail, text chat or physical mail. How do we handle this? – Technologies that pop up are attended BOTs, conversational BOTs enabled by Natural Language Processing (NLP) and intelligent text extraction from scanned documents using OCR (Deep Learning-based). This blog focuses on conversational BOTs and Natural Language Processing, and how they can complement Robotic Process Automation RPA tools.
Interactions of customers with insurers serve as moments of truth. Therefore, it is very important for insurers to effectively handle the interactions that are initiated through calls, e-mails, social media, text chats and documents that come through mailrooms. The conversations cut across all processes - new business to policy servicing to claims. Based on our experience, nearly 50 – 70% of conversations are related to enquiries seeking some information and a good proportion of them are straightforward responses. This fact opens up the avenue for both efficiency as well as better customer servicing. If some of the simple conversations can be automated, human workforce can be used for more engaging conversations, while the wait time for more complex conversations can be reduced. Apart from this, e-mails and incoming documents act as a starting point for a lot of processes e.g. payment to supplier. There is a limitation to automate these using RPA due to unstructured inputs in the form of attachments and the need to interpret the intent of communication
Is there a solution that can handle both these needs of the insurers – improved customer experience and operational efficiency? Yes. Natural Language Processing-based conversational BOTs (Voice, Chat and E-mail Bot), supported by OCR in certain cases, could help.
Is NLP a hype to bring in cost savings? I’m afraid not. While it definitely works on the premise of automatically handling conversations, cost savings through FTE reduction is a very simplistic outlook of evaluating this. Before getting an answer to the above question, let us look at some of the typical challenges faced by the customer service operations of an insurer.
Challenges faced by customer service operations
- Are Millennials and Gen Z expecting more? – Most insurers across the globe are targeting the millennial and Gen Z customer segment who are used to the digital way of life. To name a few, they expect multi-channel engagement, round-the-clock support and instant response at their convenience. Responding on weekends and after office hours is increasingly becoming a difficult expectation to manage. Organisations need to evaluate whether their customer and intermediary service expectations fall in this category.
- When was the last time we had to prepare for a catastrophic event? – As discussed earlier, handling customer enquiries on normal days in itself is time consuming. Consider a situation where an occurrence of a catastrophic event would trigger a sudden peak in the inflow of claims and enquiries! Well, this is a nightmare for any insurance company’s operations. There is a scramble to set up stop gap claims support teams. As the customers are already in panic, it is very crucial for insurers to handle enquiries and acknowledge claims on a timely basis.
- Are customers frustrated about the amount of time it takes to reach a specialist? – Yes. As customers, how many times have we called support teams with anxiety and faced the frustration of options-based IVR or front line support staff not being able to put us through to right team? We wish we were transferred at the first instance. To provide best-in-class customer service, it is very important to transfer anxious and irate customers to a specialist as soon as possible.
- Documents attached with correspondences – It is common for the operations team to receive communications with attachments like survey reports, supplier invoices, proof of loss documentation etc. So, it is important to not only interpret the communication, but also process the information in an attached document e.g. automatically interpreting a supplier invoice and initiating the payment process, updating claim estimates based on survey reports etc.
While these are some of the common problems in contact centre operations, is technology equipped to handle this?
- Voice is still most common form of incoming conversation: Based on our experience, nearly 70 - 80% of conversations are still initiated through a voice call. To handle voice requests through technology, there are two steps – converting voice to text and then interpreting it through NLP. This means that the accuracy of the outcome completely depends on the quality of speech to text conversion.
- Long sentences are less likely to be interpreted correctly: Customers tend to provide lengthy details on their problem, which in turn is a challenge to NLP. Conversations with too many words are more likely to face errors in interpretation.
- Conversations covering multiple topics: The tendency of humans is to set the context of the call / chat. For example, when they initiate conversation, they say, “I am calling to know the cash value of my policy and withdraw if the balance is sufficient.”
- Problem of false positives: More than not understanding, interpreting wrongly is a bigger problem, which has a larger impact in terms of customer experience. Imagine a conversation where the customer is enquiring about the premium due date, while we respond with the premium amount!
- More than categorization: People believe that the prime responsibility of NLP is to identify the intent of the customer query. However, it is even more important to extract relevant information for further processing. For example, when customer asks “What is the premium due date in my policy P987656,” it is important to extract the policy number to process the request.
- Focus is on problem solving, not on imposing technologies: To some of the challenges, it is very important to emphasize on problem solving rather than applying technology. For example, there are problems like invoice processing wherein 70% - 80% of invoices could be from empanelled providers, and thus received in a standard template. So, a choice needs to be made based on the proportion and number of variable templates.
So, does this mean technology is not mature to handle different types of conversation? Absolutely not. It is about applying technology with clear objectives and focused outcomes.
- Appreciate the nuances of each channel (voice, text chat and e-mails) when designing solutions
- Choose a technology to solve problems and not the other way around. The ideal flow starts with identifying the problem, coming up with a list of solutions and leveraging the best technology out of that. On the other hand, forcing a technology to be a part of the ecosystem for the purpose of exploration in spite of having a simple effective solution will result in resource wastage
- Eliminate false positives at first interaction and transfer them to human servicing or schedule for human intervention if the request is during non-office hours
- Eliminate potential errors by leaving out conversations which can fall under this category
- Understand customer sentiments and try not to solve complaints and anxious requests automatically. Allocate them to priority teams at the earliest
- Have an iterative approach to problem solving so that the chances of success are much higher. For example, categorize enquiries and service requests first before moving on to call types. Have clear business rules based on insights on confidence levels to determine between automatic and human handling of the query
- Paraphrase the query to the customer to confirm the understanding when there is ambiguity instead of providing an incorrect response. For ex : multiple topic queries, multiple intents matched
- Educate customers that they are interacting with a virtual agent in voice and chat channels which will change the way in which they interact
- Complement service teams with prediction even if conversation handling has not been automated. Assisting the customer service representative with cues on insurance process based on the topic of discussion, customer emotion derived from his/ her tone, speech pattern, pitch & pace, etc., will help in enhanced customer service
To summarize, an ideal process of handling conversations starts with defining the problem domains, choosing the right technology, and focusing on solving majority of transactions and not exceptions. Based on this, the next appropriate action could be determined - responding with an answer or a query for clarity or transfer to human agent. This approach would help yield better results.
Thus, defining the applications of conversational solutions and NLP to solve business problems is about choosing the right battleground so that there is no negative impact on customer experience.
Refer Mindtree’s offering and experience on Intelligent Automation in Insurance