4 Steps in Building an AI Powered Chat Assistance Insurtech Application
Growthbotics.com was tasked with building a B2B/B2C AI powered chat application for an American insurance financing company, IPFS. IPFS provides financing for insurance providers and holders. The chat application’s main feature was its AI NLU (Natural Learning Understanding) capabilities and basic chat banking services like account balance checking and loan application. Here are the 4 main steps in building an AI powered Insurtech app:
- Ensure Adequate Data Privacy and Security with VPN and Password-Enabled API.
As with all of our clients in this industry, data privacy and security are the utmost importance in building any AI applications. We have specifically developed a VPN enabled HTTPS REST JSON API with authorization so IPFS’ consumers would be able to securely access sensitive data like their account balance.
2. Build out AI NLU by feeding historical data, lots of it.
Next, we spent many months building up the AI NLU capability of the chat assistant. Training AI requires hundreds and thousands of data feeds and supervised learning. We collected various historical chat transcripts and compiled a database with responses. With these initial datasets, the AI chat assistant app would be able to expand its “knowledge base” and extrapolate intents by backpropagation.
However, we would still have to hard code the AI responses for possible user inputs. The snippet code below shows how we coded in the responses according to various intents:
Once done with the responses, we are ready to deploy the application to various channels like Telegram, Skype, LINE, Facebook and Website. Below is an example of our AI NLU in Telegram. When users type “insured”, the AI chat assistant would be able to respond accordingly.
3. Create a professional persona for your AI chat assistant
As a chat assistant, visual design is fairly simple and limit but includes rich cards, quick replies buttons, gifs and image sliders. Based on our experiences in the finance and banking industry, a professional, clean user interface trumps a flashy one. For IPFS, we built a telephone customer service persona to the AI chat assistance, coupled with functional buttons.
Here is a screenshot of the AI that was built into their website:
4. Set up a customer feedback system to continuously improve your AI
In order to collect feedback and improve the AI app, we added a feedback menu so users can rate their experiences:
With this feedback menu, we built a simple visualization dashboard to see how well the AI application was reviewed:
After series of improvements based on initial feedback, the AI assistance is now under a trial phase but has already gained traction with over 20 unique users every week. Growthbotics is excited to see a wide-scale launch of the AI application in the insurance industry.
Like our blog post? Send a message over on growthbotics!
