When shade matching became a bigger question
Cornell product management course project · Exploratory interviews and synthesis
Five interviews shifted the question from predicting a shade match to helping someone try an unfamiliar beauty product with confidence. Sampling, product information, and the connection to stores deserved attention alongside shade matching.
My role: I conducted the interviews and synthesis as independent coursework, with five participants whose experiences changed the direction of the project.
The full story
The problem
Buying complexion makeup online asks a lot of a photograph. A shade can look promising on a screen and entirely different on your face. I wanted to understand what would help someone feel more confident about trying a new product.
The starting hypothesis
My initial concept was a match score informed by purchase history, reviews, images, and return feedback. Could existing information help predict which shade would work?
It was a plausible direction. It was also a solution I could have become very attached to before learning enough about the problem.
What I did
For an independent Cornell product management course project, I interviewed five people about buying unfamiliar beauty products. I kept the questions broader than shade matching: how they discovered products, what made them hesitate, and what helped them decide.
This was a small exploratory study, useful for finding questions to investigate—not a representative survey of beauty shoppers.
What changed my thinking
The cart wasn’t always the checkout line. Three of the five participants used an online cart as a shopping list, then went to a store to try products before buying. What might look like an abandoned purchase could be a step in a longer journey.
Color was only part of the uncertainty. Texture, coverage, finish, and scent were also difficult to judge online. A better shade prediction alone would leave several reasons to hesitate untouched.
Trying a little could matter more than predicting a lot. Four participants mentioned paid sampling when asked what would make them more confident. Flexible returns also helped. These were directions worth exploring alongside a recommendation model.
A digital try-on still had to earn trust. None of these five participants described a positive previous beauty AR experience; three described negative ones. That didn’t settle the value of AR. It told me that accuracy and credibility would need attention with this group.
A better question
I began with “How might we predict a shade match?” I came away asking “What would help someone try an unfamiliar beauty product with confidence?”
That broader question made room for sampling, clearer product information, useful reviews, and a better connection between online research and an in-store visit. It also left room for shade matching—without making it responsible for solving everything.
What I would test next
I would start with the smallest useful way to reduce uncertainty, such as a clearer route to trying a sample, and learn whether people used it and felt better equipped to choose.
For a future live test, I would define a primary metric tied to the specific intervention, then check secondary signals such as returns and repeat purchase over an appropriate period. Those are proposed next steps, not results from this project.
My role and the limits
I conducted the interviews and synthesis as coursework. No production feature was built or launched, and no conversion or revenue improvement was measured. I’m grateful to the five participants who shared their experiences; their answers made the work better by changing its direction.