Wellframe Ride-Alongs: Care Without a Foundation

Three surveys ran alongside the ride-alongs. Two went to staff, covering onboarding and day-to-day efficiency. The third went to members directly and pulled in the largest pool by far: 2,800 responses. Separately, I ran 25 in-depth interviews and validation sessions with our own internal subject matter experts, to pressure-test what the field research was telling us before it went anywhere near a roadmap.

I owned everything downstream of the research, too: synthesizing all of it into the internal insights deck, running the client readouts, and presenting the findings live at one of our quarterly customer conferences.

The research collapsed into five themes. Staff had no stable foundation: some were toggling between as many as 9 different systems and spending 20 to 45 minutes just building a member profile before they could make a single call. Retention lived or died on the first day. And staff were still held to telephonic documentation requirements that never adjusted for how differently digital actually worked, so they were quietly doing double the paperwork just to stay compliant.

Onboarding was its own fight, and a different kind of problem than the others. It was sales, even though nobody on staff had been hired or trained to sell anything. Most of these were cold calls: a member picks up the phone, and within two minutes has to hear that they're not healthy, that a stranger wants to help fix that, that it's free, and that step one is downloading an app. That's four reasons to hang up before anyone's said anything reassuring, aimed at clinical staff who'd never needed a sales instinct before. Only half of invited members ever converted.

In spite of all that needed to be overcome to be successful, the fifth theme is where the real story is.

The ah-ha was what showed up once it was dead. Per touch, the work on either side of the interaction shrank hard too: telephonic prep and documentation together ran close to an hour, digital cut that to ten or twenty minutes. But staff weren't doing eight of these a day anymore. They were doing forty. Ten to twenty minutes, forty times over, adds up to as much daily overhead as the old hour-long version ever did, and at the high end, more. The bottleneck didn't shrink. It moved.

We'd spent years optimizing the middle of the interaction, and we'd won. Now the problem lived at the ends.

Every theme got broken down into multiple specific design questions, simplifying the problem instead of leaving it as a vague direction. The staff-foundation and onboarding themes became: how might we enrich a member's history so staff aren't rebuilding it from scratch every time, and how might we cut telephonic onboarding down using digital and marketing tools that don't need a live person on the phone. The Day 1 theme became: how might we match a member with genuinely relevant content the moment they onboard, instead of the generic welcome program that was underperforming everywhere. The overhead-and-documentation theme became: how might we take the pre- and post-interaction work off staff entirely, since that's exactly where the cost had moved.

This wasn't a slide deck that sat in a drive. Getting there meant manually synthesizing 60 hours of ride-along observations against three separate surveys and 25 SME interviews, by hand. There was no AI to help find the pattern connecting something a survey respondent wrote in a free-text box to something I'd watched happen on a ride-along weeks earlier; every one of those connections got made manually, until five patterns emerged instead of a hundred scattered anecdotes. Once they were named, this went to our Senior and Executive Leadership Teams directly, and became an ongoing roadmap roadshow across the company.

We took the same roadmap on the road externally too, running sessions with customers to walk them through the shift in direction before any of it shipped. For everyone who'd participated in the original research, we gave back individualized findings of their own: their own workflow gaps, their own change-management opportunities, training tips specific to their team, pain points pulled directly from their own workflows. It resonated. The support that came back was significant.

The features themselves went through their own prototype and test cycle. That's its own story, not this one.

Twenty distinct opportunity themes came out of the original Ride-Alongs. Sixty percent shipped as real product within a year, roughly sixteen named features. Personalized onboarding links replaced generic invitations. A Day 1 activation flow replaced the old first-touch experience. Members could browse and request enrollment into care programs that actually matched their condition instead of defaulting into the generic welcome track, directly answering the finding that condition-specific programs kept members where generic ones lost them. And the AI staff assistant idea, named explicitly in the original research as something staff wanted, shipped as a GenAI summarizer pilot in about eight weeks, championed personally by our CPO. I stayed on as a stakeholder on that one, weighing in on what design produced and shaping the final look, not running it day to day.

That eight-week pilot didn't stop at eight weeks. It grew into something that now runs across 189,000-plus eligible members, summarizing more than 99% of them without hitting the model's context limit, tested at 3,000 daily active users. It didn't get there overnight. It got there because we stayed on it.

The data backed the shift toward personalization hard, and it showed up everywhere we looked. For example: what members saw when they onboarded played a huge role. Nearly a third of members stopped engaging entirely after the day they onboarded, and the biggest predictor of whether that happened wasn't the health plan or the staff caseload. It was what program they'd been placed in. Members onboarded into the generic Wellness program had a 45% abandonment rate within two weeks and an NPS of 41. Members onboarded into condition-specific programs like maternal health had a 17% abandonment rate and an NPS of 63 over the same window. Same platform, same company, same staff. The only variable was whether the content actually matched the person.

The clearest signal the work mattered wasn't in that data, though. It was that individual customers started asking for their own dedicated ride-along sessions after seeing what the original research turned up for them. I ran several more rounds specifically for two of them, mostly conducting those sessions myself, which pushed total observation time well past 100 hours across the full program.