In an economic environment greatly affected by the COVID-19 pandemic, chatbot technology provides a significant opportunity to minimize hands-on human interaction while providing necessary services to customers. Quesnay recently interviewed Chris Rivard and Anthony Peccerillo from CodeObjects, which powers the chatbot provider Insurbot.ai. Below is the Q&A from our interview:
CHATBOT’S USE OF TECHNOLOGY
The main type of technology that chatbots utilize is artificial intelligence (AI). Using AI, chatbots are designed to behave like humans but they are accessible at any point of the day by voice and/or text. Chatbots with AI are easily scalable, but limitations do exist and human hand-off is still necessary at certain times.
Q: How many types of chatbots do you believe there are?
A: We don’t differentiate the chatbots by technology, but more by the implementation. In general, there are 3 different “types” of chatbots:
The first type of chatbot is targeted at supporting a specific task. It asks a pre-configured set of questions and is looking for specific responses. Resetting a password would be a good example of this, the bot is looking for specific info like a current password, followed by additional specific input; the new password. This process executes and then moves on. This chatbot cannot perform interpretation or processing language. It cannot perform outside of its intended abilities.
The second type of chatbot has a deeper knowledge of certain topics and is much more task-oriented. This bot has a deep understanding of the intents it is capable of completing, but it is contained within these tasks (like making a payment or viewing an invoice). It understands the complete scope of the question asked based on a specific transaction.
The last type of chatbot is assistant oriented. This is a mix between the first and second types of chatbots. Basically, the chatbots have the ability to guide conversation flows (things like digressing from one thing to another, asking to repeat a question, allowing the caller to interrupt the chatbot or tell the bot to wait). However, from a business knowledge perspective, it has a much deeper knowledge base for a specific domain that it is supporting, in our case the insurance industry. It can interpret and identify key business entities and requests and with confidence respond in context. The end result is an assistant that can converse and respond with domain details.
Q: With today’s technology, how can a chatbot make the consumer’s life easier?
A: Technology, especially mobile devices, allow people to do a lot more on the move, and on their own time, it’s no longer a 9 to 5 window. Chatbots allow a company to remain available to its customers 24 hours a day. It is also a great way for companies to interact with their clients on a proactive basis. Companies can be more of a holistic service provider by sending alerts, such as payment reminders, and storm notifications and mitigation aid, in the event of severe weather. Because of the multi-platform connectivity, customers can respond with the method of their choosing.
Chatbots can recognize unique user visits, capture conversations, and when that user returns rather than starting at the beginning, can pick up a conversation where it was left off. All adding to the personalized conversational abilities designed at making consumers' lives easier.
Q: Do you believe chatbots can be fully trusted to handle sensitive information and change customers’ services?
A: We believe that if the right measures are taken, then chatbots can be trusted to handle sensitive information. Ultimately, the chatbot isn’t the issue. The challenge is dealing with the channels of message delivery, like text messages, phone applications, and voice communication. Channels like these cannot adequately protect the data. However, as soon as the chatbot determines there will be an exchange containing personally identifiable information, it can direct the consumer to a secure link. Chatbots have the ability to change communication channels from voice to text, redirecting the user to a link where something secure can be entered and stored in the system of record by the chatbot.
BUILDING AND MAINTAINING CHATBOTS
Chatbot providers help companies build and deploy chatbots. The devised steps are: structuring data, developing and training the chatbot, testing, and deployment. The companies that hire the providers can choose to be as involved as they want in each of these steps.
Q: Which step of creating a chatbot takes the most time and why?
A: Training the chatbot takes the most time. We recommend a supervised learning approach, which means constantly using analytics to detect inconsistencies and better ways of interacting with the end-user. Being able to interact with different ethnicities, ages, and other demographics, allows the chatbot to learn the many ways a person can ask the same question or different ways to answer a given question based on the audience. The chatbot needs to understand the languages and vernacular of the domain. As new information is available, we can provide annotations that provide instruction to the bot on how to interpret this new data. Training is continuous. If you continually add to the bot’s knowledge base, its abilities will be greatly enhanced.
Q: After a provider like yourself comes in and creates the chatbot for the client, what is required on the client's end to maintain it?
A: After implementation, there is a significant amount of time spent on training, as we referred to in the previous question. However, this is more in relation to the teams dedicated to supporting the technology. It’s important to involve both business and IT while supporting the chatbot on the client’s side. Regular performance metrics should be analyzed by business experts because these folks understand the subject matter. For example, if you are implementing a service-oriented chat, a corporate social responsibility lead would be a great person to have to oversee the bot because they are proficient in the tasks the chatbot will be handling. They will be able to evaluate successes and point out where the additional training needs to take place. Ongoing training involves working through the day to day conversations, where customers introduce the chatbot to new terms that the chatbot needs to be trained on.
CHATBOTS IN INSURANCE AND BEYOND
The insurance industry is changing and customers are looking for more customized services, which have made chatbots a valuable tool. Insurance companies already store an incredible amount of data. With AI, they can utilize it to improve internal efficiencies, mitigate risks, and improve their customer’s overall experience.
Q: In your opinion what is the ideal use case for an insurance-based chatbot?
A: Insurance based chatbots are well suited to handle repetitive, low-touch tasks. We are not saying these tasks are low urgency or priority, but they don’t require a high human touch. Examples of this include: inquiries to check a balance or to pay on an existing bill, requests for documents, such as a declarations page, ID card or certificate of insurance, or when someone needs to file an initial first notice of loss. These are all examples of transactional tasks that are easily handled by a chatbot which allows a company to better use its human staff in more effective ways.
One of the challenges with advisory bots, that are attempting to satisfy more complex scenarios or scenarios that require human reasoning, is that programming biases may be present. Currently, people are dedicated to addressing these challenges and identifying chatbot bias, but until these technologies become mature, we feel that insurance companies will not rely entirely on a decision made by a bot. We feel that bots are best used to augment decision making rather than being entirely responsible for making the decision.
One example of a low-touch task in insurance is filing claims. Traditionally, companies are challenged with supporting the claim volume that arises during a natural catastrophe. To handle the unusual spike in call volume, they are forced to outsource the call center function to a supporting organization where increased wait times and the margin of error are much higher than that of a chatbot, due to inexperienced resources. When Hurricane Michael hit the Gulf Coast of Florida, one of our clients was able to file 30% of all claim volume related to the storm with our Virtual Assistant (Insurbot.ai) saving them in the area of $1 per minute during the post-storm call spike. Not only was it a better and faster experience for their customers, but it also showed great business value. Read the whole story, here.
Q: What tasks do you believe chatbots should never perform? And which industries should it not penetrate?
A: There are really no industries that cannot see some benefit from AI and chatbot systems. The augmentation of services can always be improved on as long as technology doesn’t try to replace people. It is very important to be aware of bias as interactions that could involve a high level of bias should look to be avoided or highly scrutinized. Human bias can creep into the AI world through the data and algorithms that are programmed and the chatbot ultimately cannot be faulted for its bias, it is the people writing the encryption that pass along their own biases which are the issue. The other issue is the data used to train the chatbot, if there is not a broad enough sample size used, bias is sure to exist because of limited scenarios and examples for it to pull from. While these situations may never be avoidable, ways to regulate and protect against it need to be in place.
Contact us if you’re interested in learning more about CodeObjects, Insurbot.ai, or how to partner with Quesnay.