We Built an AI Sidekick for Insurance. Our Users Didn’t Even See It.

"If you don't know anything about a product, you also don't know what to ask."

This was the problem. Before we started this project, buying disability insurance (DI) on our platform could be a relatively lonely experience. You’d land on our site, facing a complex and deeply personal decision process, and were basically on your own. People requiring expert-level knowledge had to hunt for answers on Google, digging through FAQs, or reaching out to support.

That friction meant we were losing those who needed the product, but simply might not be guided enough. So, we asked ourselves a question: What if we could give every user a personal advisor to help them through it?

The Goal: A Guide, Not a Gadget

We wanted to build something that felt like a real, human expert was sitting next to you, offering a helpful tip at just the right moment. It had to feel native to the product, not like a bolted-on chatbot.

We chose to start with DI because it’s our most personal product, and it’s where people need the most guidance. We aimed to preserve the most effective parts of our existing funnel while seamlessly integrating this new AI advisor into the experience. The idea was that when you used it, you would "still feel at home," just with a friendly expert along for the ride.

How a Great Backend Saved Us Months

Here’s where things got interesting. We budgeted a lot of time for backend integration, expecting a huge headache. We were wrong.

It turned out to be a "huge blessing." Our existing backend was so well-built and the APIs so clean that plugging in this entirely new AI-powered frontend was shockingly fast. In about a week, we had a working prototype.

Instead of spending months building communication systems, we got to focus on the fun part: making the user experience great. It was an amazing moment where we realized how much the teams before us had set us up for success. They had built for the future, and we were building on that momentum to unlock new opportunities.

Our Mantra: "Some Parts Need to Be Broken"

We learned a hard lesson on a previous AI project: we spent too much time trying to perfect the user experience before the core system was ready. We iterated "in the wrong order."

This time, we flipped the script. We built the core system first and pushed to get something functional in front of users as fast as possible. I believe that "some parts need to be broken in order to have that time to fix it." You can't waste months perfecting a theoretical idea in a bubble. You have to ship, get feedback from real people, and iterate. It’s a core part of our Grit and Entrepreneurship values — trying new things without being afraid to fail.

The Humbling Feedback: "Wait, There's a Chat?"

Our biggest surprise from user testing was a humbling one. We gave the AI advisor a huge chunk of the screen and thought it was super obvious.

But our users didn't see it.

They were so focused on the numbers and figures on one side of the screen that they were completely blind to the conversational assistant on the other. It wasn’t until we triggered a pop-up message that they even realized it was there.

That feedback was a gift. It forced us to be more intentional. We added:

• Personalized welcome messages that pop up with a friendly "Hi!"

• Contextual triggers that offer help based on what the user is doing

• A feedback button on every single message, so we could learn directly from our users without being in the room

What We Used to Build It

The stack behind this funnel isn’t magic, but it is intentional.

Here’s what helped us move fast and stay flexible:

• Frontend: React with Layered UI states separating chat from core funnel logic, Framer Motion for smooth animations, responsive-first layout system

• AI layer: Prompt-managed LLM interactions, intent classification for user queries

• Feedback loop: Message-level feedback triggers + prompt refinement pipeline

• Backend: Existing APIs with added chat endpoints, session persistence for conversation context, and structured data serialisation for AI context

• Collaboration: Insights from our Safety Net AI project and internal knowledge base

The result? A conversational layer that felt native to our product and didn’t break our backend.

Why This Matters for Insify (And for You)

This project was about more than just a new feature. It proved what’s possible when you have a culture where every new idea is "welcomed with open arms" and then rigorously challenged to make it better. It’s a place where you can find the solution to a hard problem just by talking it through with a colleague–a process I learned is called "rubber ducking"(Thanks for being my rubber duck, Ralph!).

I've learned that being a perfectionist doesn't help. Trying to reach perfection is what helps. This project, like any other, isn't perfect, but every iteration brings us closer to our goal: simplifying complexity and keeping it human.

We’re not just building insurance products; we’re building a new kind of insurance experience. If you’re an engineer who’s motivated by that kind of challenge and wants to build on a foundation that enables speed and impact, we should talk.

We're hiring.