CompariSure

Using Messenger for financial inclusion purposes - allowing people to buy insurance policies within Messenger, an end to end process, from first ad click right through to issuing policy documents.

Comment

Inspiration

Too many South Africans are unable to access basic financial services products, such as life insurance, because the distribution costs for traditional insurers are simply too high, thus excluding many people from the market. Furthermore, allowing customers to buy these products via a website online hasn't worked either, especially at the bottom of the pyramid, where most customers are browsing on mobile, and where computer literacy levels are low.

Our founding hypothesis was that by building automated chatbot conversations, and by leveraging the widespread use of Facebook Messenger, we would be able to build a virtual insurance agent that would simplify the process of buying insurance, and also drastically improve product accessibility for the average man on the street.

What it does

Our chatbot sells Funeral insurance end-to-end in Messenger. Users go through the quoting process by selecting how much cover they want, as well as adding various family members to the policy. When the user is happy with their policy, they are shown the final quote and given the option to buy the policy. Thereafter the bot collects relevant information (e.g. beneficiary details, payment details etc), after which the client is provided with a policy document in Messenger.

How I built it

Our chatbot is predominantly a rules-based system, with an overlay of artificial intelligence at certain key stages of the conversation. We built an in-house chatbot framework which is used to power the conversations - this is used to drive users through the conversation towards taking out a policy. We are gradually introducing various levels of intelligence into the system. For e.g., we have recently introduced intent analysis, which is used to classify users' emotions and react accordingly e.g. if a users says something like "Leave me alone", the bot identifies that the user is "angry", and politely ends the conversation.

Challenges I ran into

Despite clear guidelines and prompts, users often don't seem to know that they are speaking with a chatbot, and often reply using natural language. Our process had to be simple and slick (e.g. using quick reply buttons, carousels etc), but at the same time cater for a wide range of user responses. Building in this refinement took months and much analysis into typical user replies / user errors / points of drop off etc.

Accomplishments that I'm proud of

Over 6 months and with our human-assisted chatbot technology, we were able build a "call center" that was effectively the size of 20 people, but with only 3 human agents supporting our chatbot. The increases in efficiency were immense. We're the only player in South Africa that is able to take users end to end in the purchasing of Funeral insurance - we haven't seen another similar implementation anywhere in the country.

The rate at which our chatbot learns and evolves is also quite astonishing - almost not a week goes by without new features being released, or the bot becoming smarter in how it deals with users.

What I learned

Initially we built systems to support our human agents (trying to make them more effective), but we learned that one could achieve more by building new systems from scratch, and then getting human agents to support those systems. Chatbots seem to have not fully taken off in the financial services space because people are trying to apply them to existing old, patch-worked systems. People are trying to build bots that cater for every possible user need - which inevitably leads to poor experiences and unhappy and frustrated users. Our chatbots are built for niche purposes, and solve very specific needs for our clients.

What's next for CompariSure

CompariSure is now in the process of building out chatbot solutions for several new South African insurance providers. We've also had initial discussions with players in East Africa, where there is a strong belief that our technology could help overcome the low levels of insurance penetration seen in the area. While we've started with insurance, we've also began playing in other areas where we feel chatbots could ad significant value (mostly notably the HR space).

We truly believe that by building slick, easy to use chatbots that help people get the right financial product, we can have a significant impact on millions of people's lives across the African continent.

Built With

Share this project:

Updates