Figma
GitHub
User Interviews
Codex
Claude Code
Supabase
VercelThe opportunity
Skincare shoppers are surrounded by claims: “clinically proven,” “brightening,” “reduces fine lines,” and “pore minimizing.” But in the moment of deciding what to buy, most people have no clear way to tell whether a product can actually deliver on those claims.
Existing skincare tools mostly focus on ingredient safety. They can help answer, “Could this ingredient be harmful?” But that does not answer the more common buying question:
Will this product do what it says it will?
The problem is also shaped by an important constraint. Claims around acne, eczema, and rosacea can cross into regulated medical territory. SkinSavior needed to make evidence clearer without implying diagnosis or treatment.
The challenge was to make scientific evidence useful for people who do not want to become skincare experts.
The product bet
We found a crowded category. Products like Yuka, EWG Skin Deep, INCIDecoder, SkinSort and Paula’s Choice already offer variations of safety scores, ingredient explanations, routines, recommendations and community features.
Rather than replicate those features first, we made a narrower product bet:
SkinSavior would grade the certainty of evidence behind each specific product claim.
This created a clear point of difference and gave us a disciplined build order. We chose to build the evidence-grading experience first, then defer familiar but less differentiated features such as routine tracking, compatibility scoring and dupe finding.
Those features may matter later, but they would have made the first version feel like another skincare utility rather than a product with a distinct reason to exist.
Existing tools answer
“Could this ingredient be harmful?”
- Yuka
- EWG Skin Deep
- INCIDecoder
- SkinSort
- Paula’s Choice
SkinSavior answers
“Is this specific claim supported by evidence?”
- SkinSavior
Discovery
Competitive landscape
Safety is covered. Evidence is not.
Our competitive analysis showed that ingredient safety and feature parity are well-covered. The gap was not another score. It was a way to evaluate whether a specific claim is well-supported.
| Tool | What it scores | Per-claim evidence |
|---|---|---|
| Yuka | Ingredient safety, single score | No |
| EWG Skin Deep | Hazard rating per ingredient | No |
| INCIDecoder | Ingredient function and flags | No |
| SkinSort | Match score, routines, dupes, community | No |
| Paula’s Choice | Ingredient ratings by editorial review | No |
| SkinSavior | Evidence certainty per claim | Yes |
A second gap: analysis without context
While reviewing these tools, we also noticed features breaking in ways that had nothing to do with missing data. In 1 case, a TikTok creator scanned a topical flea treatment for dogs and told her followers it was unsafe. The app’s own description noted the ingredient was common in veterinary practice, with a caveat that most likely meant it shouldn’t be ingested by people. She misread the result, but the app set her up to, because it judged the ingredient on its own instead of the product for what it was. In another case, a skincare analysis app evaluated hair care the same way it evaluated face products, telling users whether a shampoo would break them out and where it fit in their skincare routine.
In both cases the analysis wasn’t wrong, it just didn’t know what kind of product it was looking at. SkinSavior only covers skincare today, so this hasn’t affected us yet. But it shaped a rule we plan to build in before expanding into new categories or letting users add products from outside the catalog: evaluate the product’s category and intended use first, then grade the ingredients against that context.
Early user interview
Directional signal before expensive decisions
We conducted an early semi-structured interview with a skincare enthusiast to pressure-test our assumptions. This was directional research, not a representative sample, but it produced several useful insights to validate in future interviews.
We learned that:
- Even people who research skincare products do not naturally think in ingredients.
- People make decisions based on desired outcomes, product purpose, texture, affordability, and whether they can realistically repurchase a product.
- Trust is already fragile; shoppers recognize the influence of marketing, sponsored reviews, and vague claims.
- A future dupe finder should prioritize purpose, product type, and texture—not just ingredient similarity.
- A skincare diary was less compelling than we initially assumed.
These findings reinforced the evidence-first direction and changed parts of the roadmap before they became expensive to build.
Product and experience design
The central design challenge was turning evidence quality into something that feels understandable and useful at the moment someone is evaluating a product.
Decision 01: Evidence model
Per-claim evidence instead of one ingredient score
The first concept used a single evidence score for each ingredient. We rejected it because the same ingredient can have strong evidence for one outcome and weak evidence for another.
The final approach evaluates claims separately. A product’s brightening claim and pore-minimizing claim can receive different evidence grades, helping users understand what is supported and what is uncertain.
Multiple randomized human trials have measured niacinamide's impact on skin pigmentation and the appearance of dark patches. These studies observed changes in how pigment is transferred to skin cells. While the evidence is strong, most trials involved small groups of participants, which limits how broadly these findings can be applied to all skin types.
Minimizes pores
Clinical studies suggest that niacinamide can improve the appearance of pore size by strengthening the skin barrier and reducing inflammation. While the effect is visible, the evidence is strongest when niacinamide is used consistently over several weeks.
Supports collagen
Some studies suggest that niacinamide may help improve skin elasticity and support the appearance of firmer skin. The evidence is promising, but more research is needed to confirm how consistently these benefits translate across different skin types and ages.
Decision 02: Explanation
Making trust visible
A grade alone is not enough. The experience needs to explain what it means, show the reasoning in plain language, and give users a path to understand the supporting research without overwhelming them.
The product experience moves from:
- 1Product claim
- 2Evidence grade
- 3Plain-language explanation
- 4Underlying research
Decision 03: Personalization
Designing a profile around needs, not a single skin-type label
Instead of forcing users into a single category such as oily, dry, or combination, the quiz profiles independent needs across oiliness, pigmentation, aging, and sensitivity. It also includes phototype and safety-related questions.
This gives SkinSavior enough context to personalize the experience without requiring users to know what ingredients they need.
Decision 04: Platforms
Designing consistently across web and mobile
The product uses a shared design language across web and mobile while adapting hierarchy, navigation, and interaction patterns for each context.
Building and shipping the product
To make the product’s trust promise real, the evidence-grade system could not rely on an AI model to decide what was true.
Architecture: Grading pipeline
The product uses a three-stage pipeline
-
Stage 1Extract
Research and study information is extracted into structured fields.
-
Stage 2Grade: deterministic rules
Deterministic rules based on evidence frameworks calculate the certainty level.
-
Stage 3Translate: AI, read-only
AI translates the completed result into plain language without changing the grade.
This approach keeps the most important output—the evidence grade—consistent, explainable, and separate from generative copy.
Shipped: Prototype scope
What is live today
- A skin quiz and saved profile
- Per-claim evidence explanations with ingredient breakdowns and safety flags
- Product, ingredient, routine, and community browsing
- Global search
- Add-a-product-by-URL flow
- A catalog pipeline with product and ingredient data
- Web and mobile experiences built from a shared foundation
Learning and next steps
Research earlier
Even one early conversation changed the roadmap. It would have been better to test assumptions sooner, before scoping features such as a skincare diary.
Build trust into the system and the interface
Trust was not only an evidence-framework problem. It affected the grading logic, content language, information hierarchy, and visibility of sources.
Product restraint is a design decision
The most valuable decision was not adding more features. It was choosing not to build the expected parity features before the product’s evidence-first value was clear.
Next steps
The product is currently being validated as a web experience. The next phase focuses on proving that the evidence-first experience is useful before expanding its scope.
Validate the core experience
Test whether people can:
- Understand what each evidence grade means
- Feel more confident comparing products
- Distinguish evidence quality from ingredient safety
- Find the explanation useful without feeling overwhelmed
- See enough value in the core product before parity features are introduced
Expand the product experience
Once the core experience is validated, the next product opportunities are: