Vial & Vale

In a design innovation course at UC Berkeley, my team of five had a semester to design a real product, from first interview to working prototype. Most teams get a problem to solve. We had to pick our own, so the first decision was choosing what was worth solving.

We chose skincare because beauty matters to people across every generation. How you present yourself is tied to your identity, and the stigma around it runs through all of society, which makes finding products you actually trust a problem worth solving. But picking the space was the easy part. We still had to figure out why finding a product that works is so broken in the first place.

Beauty shoppers face real financial and emotional exhaustion from trial-and-error buying. Women spend around $170 a year trying to find products right for their skin, and one in nine have cycled through ten or more that never delivered (OnePoll for AmLactin, 2,000 US women). The misses pile up as wasted money and wasted packaging. The market keeps answering with more options and louder claims, but more choice was never the problem.

So we narrowed our problem space into a question we could design against.

How might we help beauty consumers discover sustainable products that fit their needs on the first try?

We ran five in-depth interviews with people who buy their own skincare, focused on how they decide what to buy and where that decision falls apart. We mapped every quote onto an affinity map to surface the patterns.

1. Fear of wasting money keeps people from trying anything new.

Once someone finds a product that works, they stop experimenting. The risk of paying for another miss outweighs the upside of something better.

2. People leaned on other people rather than products.

Our participants trusted other people’s experiences because nothing about the product itself told them whether it would work for them. Reviews, recommendations from family, and even influencers with the same skin type are more reliable than the ingredients label on a bottle.

3. What they couldn’t verify, they couldn’t believe.

Inconsistent ingredient labels, product photos that mislead, or reviews that felt planted: the information meant to build confidence was doing the opposite.

A smaller thread ran underneath. A few people wanted easier access to sustainable brands, which we later folded into the product’s sourcing detail rather than treating it as a headline need.

All three patterns point the same way. Rather than there being a lack of options, the problem was that nothing gave people a reason to believe a product would work for them.

We generated a wall of concepts and sorted them into three kinds of bets. Likely to work: the safe plays, like AI-driven recommendations. Likely to please: the crowd-pleasers, like a subscription sample box. And the dark horse, the boldest and least certain, where we put a virtual try-on.

Owning prototyping, I sketched all three as lo-fi wireframes first, to pressure-test the flows before committing to visuals.

I built out all three, leading with the recommendation and sample box ideas.

But the try-on was the one everyone leaned in for. It made the app feel like the future, the thing you’d screenshot and send a friend. It looked strongest on paper, so I built it into a full prototype. Point your camera, pick a foundation or blush shade, and watch it render on your face in real time.

It was the most exciting thing we made. Then we put it in front of real users.

We ran a first round of usability testing on the prototype with our original interviewees, an early formative check to see which concepts held up before we moved into polish.

The recommendation flow drew quiet approval. People liked being able to see why a product was suggested, which was a good insight for us. But two other features split the room.

The sample box landed. Being able to try before buying visibly eased people’s anxiety about committing to a product. It answered a fear we’d been hearing since our very first interview.

[Mark’s quote]

The try-on did the opposite. People enjoyed it for a moment, then asked what it was actually for.

[Maira’s quote]

Maira, one of my interviewees, said the quiet part out loud. The try-on shows you a shade on your face and calls it a decision, but it never tells you whether the product will work on your skin. The feature we had bet on, sketched, and built was answering a question nobody was asking.

I cut it.

That was my call, and cutting something you personally built is hard. But with that, the real signal surfaced. What people actually wanted was proof that a product would work for them before spending a cent. The real problem was never personalization or polish. It was trust.

Cutting the try-on left a gap, and the testing had already shown us how to fill it. Every feature that survived showed its reasoning to build trust, the one thing the try-on didn’t. So, taking charge of designing the new iteration, I built the replacement around one idea: not just tell people what to buy, but show them why it fits.

It starts by actually getting to know your skin.

Trust had to be built before a single product appeared. So the experience opens with a short skin test, a guided set of questions about your skin type, concerns, and values.

• It uses image-based choices, so people aren’t forced to self-diagnose.
• Every question has an “I’m not sure” option, so nobody has to fake an answer they don’t have.

Every recommendation, sample, and match that follows traces back to something the user actually told us, which is what makes the rest of the experience feel earned instead of imposed.

Instead of simulating a product, we let people actually sample it.

The try-on promised to show a product on your face, but users told us that still didn’t prove it would work. So we stopped simulating and just let them sample.

The skin test generates a personalized box of three samples drawn from your matches, so you test on your own skin before committing to a full size. The limited number of samples was intentional, to make every pick feel deliberate.

We made every recommendation explain itself.

This was the quiet win from testing: people liked seeing why a product was suggested, so I built it in everywhere. Because the feed is driven by the skin test, each match can tell you why it’s a match, tied directly back to what you shared. Product pages open into plain answers:

• what’s in it
• where it’s sourced from
• who it’s for

I defaulted to the transparency people said they were missing.

Minimalism here wasn’t a style choice. It was an accessibility decision.

Our first version of these screens used busy colors and decorative fonts. It looked nice, but it had poor contrast and visual hierarchy. So I scrapped it and rebuilt with high contrast, simpler type, and a simple layout. Not only because minimalism looked cleaner, but because we wanted to make the product legible to as many people as possible. It was the same instinct as cutting the try-on. Substance over surface.

The first round told us which ideas held up, but because we’d tested with our original interviewees, people were already sold on the problem. So we ran a second round with five participants from our extended network, none of whom had seen the project or heard the pitch.

We gave them three core tasks. Find a product for your skin. Judge whether they’d trust the recommendation. Get to a sample. And underneath all three, the thing we most wanted to know: whether their confidence actually moved.

Confidence rose from 2.6 to 4.2

We asked each person the same question on a 0 to 5 scale, before and after using the app: how confident they felt they could find skincare that actually works for their skin. The average moved from 2.6 to 4.2. With only five participants this is a directional signal rather than statistical proof, but it points the same way as the rest of the round.

5 of 5

completed the skin test unaidedin about a minute, against a 45 to 50 second target

4 of 5

trusted the recommendation from its description and sourcing

3 of 5

got to a sample unaidedone needed a clarification, one struggled to find the calls to action

Task success and time

The skin test worked cleanly: all five completed it unaided in about a minute, slightly over our 45-to-50-second target but fast enough to feel light. Judging trust was the strongest result: four of five said the product descriptions and sourcing information were enough to make them trust the recommendation, which is the exact thing the whole app was built to earn. Getting to a sample was the roughest: three of five managed it unaided, one needed a single clarification, and one struggled to locate the calls to action.

What broke

That last participant gave us good insight into where the app was faulty. Their trouble finding the calls to action signaled that the sample flow had buttons that didn’t read clearly enough as the next step. It’s a fixable UI problem, not a broken concept, but it’s the kind of friction that tends to lose people. I’d prioritize it in the next iteration.

I designed a skin scan I couldn’t actually build

The scan returns numbers, Moisture 33 percent, Sensitivity 66 percent, like the app read your skin. But it didn’t, so there’s no model behind those scores. In a product built on trust, that’s the one screen that breaks the promise. I’d either tie the scores to real questionnaire inputs or cut the scan and let the skin test do the work.

I never fixed the moment the flow lost people

Testing showed it clearly. The people who stalled all stalled in the same place, because the sample flow’s buttons didn’t read as the next step. I saw the problem and ran out of time to fix it. It’s a small UI fix, but it’s the kind of friction that loses the users you worked hardest to earn.

I’d watch the decision unfold over time

A skincare decision plays out over weeks, from hopeful purchase to slow doubt to giving up, and a single interview flattens that into a story people rebuild from memory. Next time I’d run a diary study, logging decisions over a few weeks so I could watch confidence move in real time instead of asking people to remember it.