Don't [always] blame the messenger - when an AI-bot should just be a bot
By Madison Mehlferber
Executive Summary
There could be over $30 billion in shifting auto insurance premiums up for grabs every year in an industry that is only now adopting customer facing AI chatbots into their quote delivery. McKinsey predicts AI could offer 1.8-2.8% additional insurance industry revenue generated largely in marketing and customer operations.
Research published in the Journal of Marketing suggests that those who perceive the price as high are 49% more likely to still pay for the worse-than-expected price offer if delivered by a non-gendered AI bot, and 89% more likely to pay for better-than-expected price offer if delivered a gendered AI persona. This is based on an experiment on concert ticket buyers, but may well hold for highly price sensitive auto insurance buyers.
Optimizing their use of AI chatbot personas requires both experimentation and also better predicting how an insurance quote will meet (or not) price expectations of purchasers.
Implications: where AI has, and has not, touched the insurance industry
The US auto insurance industry accounted for $260 billion in yearly premiums in 2022, roughly 20% of a $1.4 trillion US insurance premium market. According to an Accenture report, about $170 billion in total US insurance premiums (so estimating $34 billion in auto insurance premiums) are available for the taking every year as customers consider switching their coverage options to gain the greatest satisfaction with their insurance company - encouraging insurance executives to design the most opportune and cost efficient customer experience possible for overall financial gain (“Research Report: Transforming Claims and Underwriting with AI,” Accenture, 2023).
According to a 2019 survey by LexisNexis, 80% of the top twenty US insurers had adopted AI (and 70% of auto insurers), but used predominantly in marketing, underwriting, and claims (“Hype or Reality: The State of Artificial Intelligence and Machine Learning in the Insurance Industry,” LexisNexis, 2020). For AI adoption in the insurance sector in the future, McKinsey predicts there could been 1.8-2.8% additional industry revenue generated largely in marketing and customer operations (“The economic potential of generative AI: The next productivity frontier”, McKinsey, 2023). Customer facing AI is the next frontier.
Unlike dealing with auto insurance claims (where individualized emotion or empathy towards the situation would impact customer satisfaction in a trying time and present greater risks for integrating AI chatbots), auto insurance quotes are predominantly done instantly and online. Entire industries are built around presenting quote comparisons, where cheapness is often the overriding metric of concern. With today’s technology, introducing an AI chatbot to that quotation delivery system would be both seamless and inexpensive – and potentially at GREAT benefit to the insurers themselves in winning more quotes.
The typical win rate, in this case conversion rate from website visits for quotes, within the insurance industry varies from 3% (“The keys to a profitable, productive insurance ecosystem”, Property Casualty 360, August 2022) to 15% (“Most Essential Conversion Rate Optimization Statistics for 2022”, Enterprise Apps Today, September 2022) depending on the source – 9% say, simply using quote forms and no chat bots. Increasing that conversion rate could for a larger auto insurer could unlock tens of millions of dollars in additional premiums.
Findings: consumers reward, and punish, different AI-bots
In today’s ever-increasing technology-driven world, company executives are faced with financial pressures to replace common customer facing roles, held by humans, with cheaper Artificial Intelligence (AI) agents. But these shifts may not be universally “better” and affect the overall product delivery and consumer decisions in a negative way. A recent paper published in the Journal of Marketing focused on how information is received when delivered by gendered humans (ie online bot personas but with a gender) or non-gendered AI bots (ie generic with no personas) during price offers.
The first study measured the responses of 174 undergraduate students receiving a “worse-than-expected” offer on the same priced concert tickets (the offer landing in a range of price expectations from 1 meeting expectations to 0 not meeting expectations) administered by an AI agent or by a human agent. The data was analyzed and fitted to a binomial logistic regression and further analyzed by a chi-squared analysis revealing customers were 49% more likely to pay if a bad offer was delivered by a non-gendered AI. The second study analyzed a cohort of 290 Amazon Mechanical Turks further utilizing concert ticket price offers. The associated responses were again recorded and grouped into a binomial logistic regression model with chi-squared analysis revealing consumers were 89% more likely to pay on a good quote if delivered by a gendered human agent.
The explanation posited was that non-gendered AI bots are perceived to lack selfish intentions when administering bad offers (so blameless) whereas gendered human agents deserve due credit from good offers, thereby dampening the extremity of negative consumer responses and harnessing the positive.
Imagine a scenario assuming an equal number of customers who perceive the price as high as they do low, and similar conversion rates for both. Using an optimal strategy, where you can predict the price perception of the customer and use the best AI persona for both groups, the conversion rate could improve by upwards of 67%. While true percentages may vary across companies with variability introduced by unique package levels and offerings that tend to be more above or below “favorable” rates, overall a larger auto insurer could unlock tens of millions of dollars in additional premiums.
Moving forward: finding the right place for AI-bots in the user journey
The trends outlined above suggest the need for a hybrid structure with gendered human and non-gendered AI agents to appeal to the unique and dynamic customer base to optimize sales in auto insurance – by taking advantage of the emotion that is drawing customers towards each delivery method:
Day One: Spend time and understand the paper and reproduce the study in the confines of an individual insurance company to design in-house treatment and control groups to determine what the power of matching quote “favorableness” to gendered human or non-gendered AI delivery mechanism may bring.
After: The ultimate success of the program is also based on being able to predict if the delivered quote will be viewed as too expensive or not based on the searchers’ expectations. That will be dependent upon the amount and the types of data that an underlying algorithm receives either by the willingness of the searchers to disclose that post-hoc or from data the company already has on their competitor’s likely quotes.
Finally: Application of AI in the auto insurance industry quotation process presents a strategic advantage offering the potential to become a leader through the implementation of AI driven sales, offering pioneers who adopt such innovative technology to gain market shares before competitors follow. The crux of this strategic decisions will include consideration for where competitors stand in their implementation and when the bot is deployed in the user journey (ie simply to deliver the quote itself, or also as part of the questionnaire process).
Once these criteria are met, an auto insurance company should be able to perform a cost benefit analysis of the results and assess the win rate benefits of using AI chatbots at all in their quote delivery and with which personas to optimize win rate.
____________________________
Madison Mehlferber is a PhD candidate in Biochemistry and Molecular Genetics at University of Virginia and a member of the Graduate Consulting Club at UVA. The research applications proposed in this article are solely the views of the author and do not necessarily reflect the views of the original academic journal article authors nor any individual member of our Editorial Board.