Unleash the Power of AI in Insurance: How to Drive Efficiency and Growth
There’s lots written about AI (even in insurance!) these days, but how many brokers are actually using it? Very few, if any. This blog is about one-way brokers can actually (actually) make use of AI in a real-world way.
Why would brokers want to use AI? Commercially speaking, brokers want to do two things: drive revenues up and costs down. Ideally, AI should do both of these things.
The Aim
The aim of Ignite’s new data-retrieval ‘AI Layer’ tool is to create an amazing data-engaged call centre assistant and chatbot that saves brokers’ staff and their customers time.
This ‘AI Layer’ will answer customer queries without the pesky need to actually go find the data and type out a response. If this works it will:
- Improve out-of-hours customer interactions so customers are happier, more likely to recommend and renew
- Reduce admin time per policy, either through more efficient call centre work or customers self-serving, leading to lower staff requirements and reduced overheads
- Create a feedback loop into systems and products so that they continuously improve
The Use Case
Take a given broker with 100k policyholders (although it could be 10k or 1m). They currently have 6 staff (this is a real-life use case).
They want to quadruple their policy count in the next 2-3 years.
Do they want to quadruple their staff count? No, they want to halve it. Doing so would equate to a saving of over £0.5m a year in staff costs alone.
How do we achieve that? A good start is by allowing customers to do more themselves on a self-service portal, but… we already do that. They can already do all MTAs, cancellations, renewals, direct debits, etc. Those that cannot help themselves resort to live chat agents.
If we can handle more customers at once then we need fewer agents rather than more. To do that we need to provide responses for agents so they don’t have to spend time digging around for data and typing answers.
A bit of experience
In 2021 Ignite launched an AI (NLP and ML) Chatbot to answer common customer queries (such as how to change direct debit dates, or how to make an MTA). In 2022 we processed nearly 100,000 chats with 1,000s of customers.
But this chatbot was not connected to the customer database, so even though it answers 50% of (FAQ-style) queries pretty well, the second 50% was going to be a fair bit harder.
Ignite’s AI Layer
When customers have a query we send their (typed) query to our AI Layer.
Our aim is to have their query answered by AI. To do this we need to:
- Know the customer
- Not breach GDPR
- Provide useful responses
Ignite’s AI layer is doing three things:
- Fetching as much relevant customer and policy data as possible
- Obfuscating the data so we’re not breaching GDPR guidelines
- Re-formating that data into structured prompting templates so the AI responds usefully
What does it actually do?
It answers questions, quickly and accurately.
For example, if a customer asks “What are my excesses?” it will immediately (2-5 seconds typically) answer “Your voluntary excess is £x and your compulsory excess is £y making a total excess of £x+y” (where x and y are the right numbers based on the policy record). Or if asked, “Am I covered for driving other cars?” it will answer “yes” or “no”. In each case the live chat or call centre operative doesn’t need to either open the customer record or type the reply.
And if a customer asks: “how can I save money on my insurance?” it will answer “I’m afraid I don’t have data available to answer that question” (rather than giving an ultra-polite greatest hits of legal-or-otherwise hints from the wider internet which is what ChatGPT would do).
How does it do this?
Ignite’s AI Layer works in a similar way to ChatGPT, except that instead of just asking the question, we give the engine a whole lot of pretext/context so it can better answer the question. Things like policy dates, costs, cover details, etc.
How do brokers use this?
This tool is already available to brokers using the Ignite policy administration platform. It will be available as a plugin to other/legacy PAS systems in the coming months.
If you’d like to see a demo, hear more about this project, or find out when it becomes widely available, get in touch.
