We opened the pitch with a person rather than a product. Mona is 65, a retired accountant, retiring this fall and moving into a new setup of life. She needs a Medicare Advantage plan inside the enrolment window — and the one agent she trusts with her health records is on holiday.

Every decision after this had to answer one question: does this help Mona decide, on her own, without feeling she chose wrong?

Before any screen, we ran a sentiment analysis across the research brief and the market. Around 66% of the sentiment was negative and about 30% neutral and adaptable. The themes were familiar but sharp: overwhelming information, cost-consciousness, health concerns, data privacy, a complex and lengthy process, and plans that went quiet after activation.

That number set the brief. We weren't designing a better catalogue of plans — we were designing confidence.

We shaped the enrolment around four moments: onboarding that asks about Mona before it shows plans; prescription savings that switch her to a generic or mail-order when it's cheaper; cost-saving options that put a real number on the screen (save up to $260 on a co-pay); and post-enrolment, where an onboarding checklist keeps going after the purchase.

The last one mattered most. The fear of having chosen wrong doesn't end at checkout — so neither should the journey.

For the second pitch we moved the story forward. Mona is now 68, a loyal member, relocating to another state mid-year — and again in an immediate need for a new plan. This time she stumbles on a Customer Co-Pilot, an AI assistant that stays with her across the shopping journey.

We designed it across three interaction models, so we could show the client where AI helps and where a human should take over.

We framed the Co-Pilot in three layers. What we know (being reactive): her history, her doctors, her claims. What we can know (being proactive): offering contextually relevant information she might overlook. What we should know (being predictive): building personalised health scenarios that help her choose.

Adaptive AI learning let it notice the moments that needed a person, and hand over smoothly to chat or tele-sales — instead of pretending to be one.

The final prototype let Mona ask the way she'd ask an agent — by typing or by speaking — and see her current plan, doctors, network coverage and drug costs update alongside the conversation.

The pitch team took the click-through prototype to the client. It was credited as a significant part of winning the engagement.