(Image credit:Monika Kazak.)
Progressive’s (Mayfield, Ohio) announcement that it has adopted of artificial intelligence (AI) chatbot capabilities to its customer acquisition channel comes as no surprise to those involved in the machine learning and natural language processing (NLP) space. Although they are not first in the insurance market, their decision makes obvious sense—and as a top 10 insurer, it serves as a signpost for the industry.
Today’s consumers expect their inquiries to be answered on their terms. If I am driving and I want to call your company from my car, you had better be there to answer. If I send a text message, I expect an answer right away. However, don’t expect me to continue the conversation right away—I’ll respond when I’m ready: The relevant term is “asynchronous communication.” And the communication must be omnichannel as well—though I submit a claim via the phone, I may want to check its status later via voice using Amazon Alexa.
With its mind-boggling adoption (1.4B active customers), Facebook messenger is the ideal platform to deploy chatbots to carry on the insurance conversation from the desktop to the mobile channel, and vice versa. The caveat is that the bot must intelligent enough—that is, it must have sufficient NLP capabilities. It must be able to keep customers out of an eternal loop of “Sorry, I didn’t understand that question.” And it must not present customer with a limited set of options inadequate to the kinds of problems they are likely to need solving.
In other words, when designing (or choosing an NLP engine) insurers must to go mile-deep and inch-wide into their ability to handle specific workflows—such as getting the AI trained on producing an auto quote, rather than a variety of customer support question. If the AI can’t understand the most common answers to seemingly straightforward questions, not only it will lead to high chatbot abandonment rates but it will damage the brand perception and lead to even further customer dissatisfaction.
When a customer enters “I need a quote for my Mustang. My ZIP code is 90210 and I drive it for fun only,” the chatbot must readily recognize multiple intents and properties to avoid asking the questions for the already-provided answers. Furthermore, it must allow for previously entered data to be corrected on demand, without dragging the customer through to the beginning of the whole process again. For example, if after entering the ZIP code, age, and vehicle details, a customer must be allowed to modify her age by simply understanding the phrase “change my birthday to…” And the conversation them must continue from the point where it previously left off.
Reengaging the Customer
Chatbots must have sophisticated reengagement capabilities as well, which is it their best advantage over webform driven workflows. When there are three questions left to produce a rate quote, the chatbot must bring the customer back into the conversation without being “spammy” or annoying. It should be able to do this regardless of where the conversation started or paused. Let’s say a customer comes to an insurer’s website to get a quote, then leaves without completely entering all required information. A chatbot has to be smart enough to reengage via Facebook Messenger to bring him back to close the deal.
To make such chatbots even more useful, they have to include all current and future communications sources like SIRI, Alexa, twitter, Facebook using text and voice interfaces. To remain competitive and cost effective, insurers are most likely better off by partnering up with the vendors specializing in building virtual insurance agents as opposing to industry agnostic platforms.