Santosh Vakil, Directeur Développement des Affaires
François Dufour, Vice-Président, Développement des Affaires
What is a conversation? The Oxford English Language Dictionary defines conversation as “a talk, especially an informal one, between two or more people, in which news and ideas are exchanged.” That’s easy. We humans talk amongst ourselves all the time, formally and informally, between two people or as a group sometimes exchanging ideas or discussing events. Let’s remove the “Human” component at one end. Let’s replace that end with a machine. Now, let’s revisit conversation. Is it possible? Can there be a dialogue between a human and a machine? If yes, can there be a meaningful one? Conversational Artificial Intelligence makes it possible.
Most of us think chatbots as Intelligent systems. They are intelligent, up to a certain degree. Almost all Chatbots have pre-fed metadata, capable of performing simple tasks like booking an appointment or providing an address. We have been disappointed far too many times by such chatbots by not providing us with the answers we look for, or not able to perform tasks that are slightly more complicated than basic functions. In reality, there is no Artificial Intelligence involved, forget Conversational AI. So, what makes a chatbot successful? In simple terms, a successful chatbot should make you feel like having a conversation with an actual human being. The conversation needs to be two sided and consistent.
To achieve a consistent chatbot, a line has to be drawn in the sand. On one side we have the functional and consistent chatbots and the other the wide majority. Inconsistencies, missteps such as contradictions are commonly noticed within the latter category. The two key differentiators that usually tip the balance in favor of a robust platform are long-term memory and semantic understanding. At Koïos Intelligence we have developed an algorithm that fulfills both conditions.
Our in-house probabilistic algorithm will give a value between 0 and 1 to any intention of a dialogue. For example, once we have defined a set context, we attribute a probability to a specific sentence. For example, if one says “Is the colour of my car impacting the price of my insurance?”. The bot would actually inquire into its database set and retrieve the correct answer such as below with a confidence level of 0.975.
“Contrary to popular belief, the colour of a car has no incidence on an insurance premium.”
This simple yet efficient method allows the AI to map a clear path for an appropriate answer which in return offers consistency and reliability. It also simplifies the learning capability of the chatbot and streamlines its processes.
Specificity is an important aspect to take into consideration. When one engages into a conversation they usually prefer having a specific answer rather than a generic one. Abandon rate is a major concern for most clients. We aim for the lambda user to show a high level interest in the interaction with the chatbot. It is primordial that we keep the user on the same and only platform. The chatbot should be able to answer any parallel questions to a specific subject and eventually recenter the client to the transaction.
The more engaging and personable the chatbot is, the better the overall conversation flow.
As a prospective customer, you would expect empathy from your insurance agents. A highly successful chatbot like Olivo will empathize with your current scenario based on the statement or questions asked. A classic example would be an individual undergoing divorce asking the chatbot to remove the spouse name from the insurance policy. An emphatic chatbot would make sure to be sensitive while providing the information and completing the task.
An important aspect of an intelligent chatbot is the ability to recall data or information about a particular prospect that was conversed weeks or months ago. Like a baby, AI enabled chatbots train themselves with information. They segregate important information and update it for further use. Constant updating of the metadata in a particular domain of a conversation marks a well evolved chatbot.
A successful chatbot can save time and money on acquiring new clients and effortlessly service existing ones while providing a human touch. In the Insurance domain, it can also provide analytics that can prove useful to Brokers and MGAs to increase business methodically. But how many of such conversational AI enabled, ML and NLP powered chatbots truly exist? In theory, maybe a few. In reality, Koïos Intelligence’s Olivo is one of them.