Creating a Social Phenomenon in HealthTech

BabyPeek tells you who your baby might be. Eye color, hair texture, whether cilantro will taste like soap to them. All of it read from the baby's DNA, in a blood sample the mother already gave for a clinical screening test.

The point was never the data. It was the moment a parent pictures a specific child.Someone with your partner's green eyes who will probably hate cilantro the way you do. That feeling is what people paid for, and it is what separated BabyPeek from a category built entirely around risk and reassurance.

The hard part is that none of it is certain. Every answer is a probability. A 70% chance of green or hazel eyes is a real answer, and it will still be wrong sometimes.

So the design had to do two things that pull against each other: create a real feeling of closeness to a baby nobody has met, without ever pretending to know more than it did. Warmth and honesty at the same time, for people who were never going to read a confidence interval.

I led design and product management end-to-end, working with product, marketing, engineering, clinical, and legal partners. Leadership, meanwhile, was measuring something else entirely.

Every trait result opens with a sentence a parent can repeat out loud, on a card colored as the trait itself. The probability sits underneath it, and it shows the full spread rather than only the winning answer. Green or hazel at 70%, blue at 25%, brown at 5%, with each bar colored as the eye color it stands for.

That was the honesty mechanism. Naming the alternatives and their weight is what stops a confident headline from becoming a promise.

Every screen carries an Other factorssection, which is the part I argued hardest for. Eye color admits that babies' eyes often change from blue to brown in the first two years. Hair texture lists products, hormones, humidity, and ethnicity, each with its own drawing. It is where the product volunteers what it cannot know, inside the experience rather than in a disclaimer nobody reads.

The science sits behind progressive disclosure. Open “How it works” and you get eumelanin and pheomelanin, or a drawing of how a follicle's shape decides whether a strand curls. Closed, the screen is a warm sentence and a number.

That was the instruction from leadership. It's a mood, not a target, so the first thing I did was turn it into a journey the team could actually act on: patients discover BabyPeek, purchase it, feel delight at the results, and share them. Four stages, each with its own metric, each a place a design decision could bite.

The business context mattered. BillionToOne was doing roughly $72 million in revenue that year, almost all of it from the clinical test. BabyPeek was a $99 optional add-on, paid out of pocket, that only a fraction of patients ever opened, and it was expected to grow into a meaningful revenue line of its own.

I combined behavioral data with interviews. The data showed what was happening. Only the interviews explained why.

I analyzed the demographic and behavioral factors associated with purchase (age, ethnicity, insurance type, marital status, location) in spreadsheets and Tableau, then interviewed 20 participants who had interacted with BabyPeek in the past week. I treated the demographic patterns as associations, not causes. They told us where to look. The interviews told us why.

From that I built six personas and shared them with the VP of Product, VP of Marketing, and Engineering Lead before any feature work began. They changed the question in every brainstorm that followed from “does this sound good?” to “which persona does this serve?”

Every patient visits the portal to collect their UNITY results, so the top of the funnel was effectively everyone. 45% opened the BabyPeek information page. 4.5% bought it. Nearly everyone who showed enough interest to look did not convert.

Research surfaced more barriers than we could attack at once, so the funnel decided the order. I framed three hypotheses and chose the ones we could learn from fastest.

Hypothesis 3 produced the most interesting design problem. We wanted to remove the price barrier, and we couldn't simply make BabyPeek free: giving away a paid add-on attached to a clinical test raises inducement issues.

So users unlocked BabyPeek by sharing it on social media. Free access had to be earned, and every unlock became both a distribution channel and organic social proof. The constraint produced a better mechanism than “just make it free” would have.

I tested it before launch. The value exchange landed and the redemption flow did not: users didn't know what to write, got lost in the promo code process, missed a verification step below the fold, and felt five steps was too many. One finding outranked the rest. Some users were not ready to announce their pregnancy publicly, and no amount of flow polish fixes that.

I collapsed five steps into one screen, moved verification into view, applied the promo code automatically, and wrote ready-made post copy so nobody had to compose an announcement from scratch.

Our PM proposed an online trait guessing game: parents enter their own traits, guess the baby's, and the product reveals the answer with a line like “the baby has mom's blue eyes.” It was a good instinct about what would spread, and I had a serious problem with it.

Asking parents to input their traits, then revealing the baby's, implies the prediction came from the parents. It didn't. It came from the baby's DNA. The feature would have quietly taught users something false about the science, in a product whose credibility rests on being honest about what it can't know. It was also expensive to build correctly, against a deadline, with no appetite for the validation it needed.

Rather than argue, I wireframed it. Mom has brown hair, dad has brown hair, the baby has a 55% chance of blonde. Now write that sentence so it's clear, accurate, and still fun. Once the copy problem was on screen instead of in the abstract, the complexity spoke for itself and we agreed to shelve it.

What we kept was the part the PM was right about. People want to play this with their friends. So I shipped it as three printable games where friends and family do the guessing and the report reveals the answers. Same social moment, no false causal story.

Peek-a-Boo is the one I'm proudest of. Players rank how likely each eye color is, then guess the percentage chance of curly hair or cilantro aversion. Scoring gives five points for an exact match, three if you land within 5%, one if you land within 10%.

That is probability calibration as a party game. To play it well you have to think in distributions and accept that a confident answer can still be wrong, which is exactly what this product needed people to understand. Nobody sits through that explanation. They will play it at a baby shower.

The page copy does the same job in one line, reaching for something people already reason about correctly:

The three hypotheses were where we started, not where we stopped. Over the following year I shipped most of the option space: lifecycle marketing, influencer collaborations, checkout experiments, science clarification, customer reviews, shareable trait results, the reveal moments above, and a price test across $149, $99, $29, and $9.99.

Two things we chose not to build: a physical keepsake book, too expensive to produce for the return, and partner participation. The shipped landing page is the clearest record of everything else.

The campaign worked and the business case still didn't close. BabyPeek missed its standalone revenue target, and a 75% conversion lift wasn't enough to change that. My read is that the gap was structural rather than a conversion problem: at $99, on a product only a fraction of patients ever opened, the volume needed to make it a meaningful revenue line was never realistically in reach. We had by then tried most of the levers available, which is what made that conclusion credible rather than defeatist.

So we stopped trying to make BabyPeek a business on its own terms. We repositioned it as a way to bring patients to UNITY. Parents who wanted the trait report asked their OB-GYN for the screening test it rides on, which moved BabyPeek's value from direct consumer sales to driving adoption of a clinical test that insurance covers, and to the clinic relationships that came with it. The arithmetic makes it obvious in hindsight: a BabyPeek unlock was $99, while the screening test it rides on sells for several hundred dollars and is billed to insurance. One patient who asks her doctor for UNITY because she wanted the trait report is worth many BabyPeek sales.

BabyPeek sells for $9.99 today and can no longer be bought on its own. It exists to make the medical test more attractive. Letting go of the original vision was hard, and it was the right call.

Legal constraints produce better design problems

The inducement restriction that stopped us handing BabyPeek out for free forced a mechanism that was simultaneously compliant, viral, and respectful of the user. It was a better answer than the one we were prevented from giving.

The honest version of a feature is usually still a good feature

The guessing game didn't need to be killed. It needed the part that misrepresented the science removed. Getting to that took wireframes rather than argument, and the result kept what my PM was right about.

What I would do differently

I scaled a solution before testing the motivation underneath it. We knew adoption was low. We didn't know whether the barrier was price, trust, or emotional readiness to share a pregnancy publicly, and those call for different products. I would run smaller experiments to separate those first and let the answer choose the lever, rather than building a campaign and learning it afterward.