How AI Is Transforming Skincare: Lessons from Building India's First AI-Powered Beauty App

The dermatologist was visibly frustrated. We were sitting in a conference room at Pers Active Lab, reviewing the latest iteration of Skin Beauty Pal’s recommendation engine, and she was trying to explain — for the third time — why the algorithm’s skin classification was clinically imprecise.

“You are categorising skin as ‘oily’ or ‘dry,’” she said. “That is what a beauty magazine does. Skin is a complex organ. A person can have an oily T-zone, dehydrated cheeks, and sensitised skin around the eyes — simultaneously. If your AI cannot see that, it is not giving personalised recommendations. It is giving personality-quiz-level answers.”

That conversation changed how I thought about AI consumer products forever. The technology was impressive. The image processing could analyse a selfie and identify visible skin characteristics with reasonable accuracy. But the gap between what the AI could technically detect and what constituted genuinely useful advice was enormous — and bridging that gap was not a technology problem. It was a product problem.

The Intersection of AI, Dermatology, and Consumer Trust

Pers Active Lab was building something genuinely novel in 2019: an AI-powered skincare application that used image processing to analyse a user’s skin from a photograph and deliver personalised product and routine recommendations. Skin Beauty Pal would eventually launch as India’s first app of its kind.

My role was leading product positioning and go-to-market strategy. What I walked into was a classic technology-first product: brilliant engineering, credible AI capabilities, and an almost total absence of consumer-centric thinking about how to translate that technology into something people would actually trust and use.

The challenge was threefold:

Trust. Indian consumers — especially women making skincare decisions — did not trust algorithms to understand their skin. They trusted dermatologists, they trusted recommendations from friends and family, and they trusted brands they had used for years. An app claiming it could analyse your skin from a photo faced immediate scepticism.

Accuracy perception. The AI did not need to be perfect to be useful. But it needed to be perceived as accurate. If a user took a photo and received a recommendation that did not match their own understanding of their skin, they would never use the app again. The gap between algorithmic accuracy and perceived accuracy was a critical product challenge.

Actionability. Even if the analysis was accurate and trusted, the recommendations needed to lead somewhere useful. “You have combination skin” is a diagnosis, not a solution. The recommendations needed to be specific, actionable, and connected to products the user could actually purchase.

Working with Dermatologists and Engineers on UX

The most educating aspect of building Skin Beauty Pal was the three-way collaboration between dermatologists, engineers, and the product team. Each group had a fundamentally different mental model of the product.

The engineers saw a classification problem. Input: image. Output: skin type and condition labels. The goal was algorithmic accuracy — matching the AI’s classifications to ground truth established by dermatologists.

The dermatologists saw a clinical advisory tool. They wanted nuance, caveats, and disclaimers. They were uncomfortable with definitive statements about skin health made by an algorithm without a physical examination. Their professional training — and their professional liability awareness — made them cautious about every claim the app could make.

The product team — my team — saw a consumer experience. The user was not looking for a clinical diagnosis or an algorithmic classification. They were looking for guidance. Help me understand my skin. Tell me what to do about it. Make it simple.

The UX we eventually built was a carefully negotiated compromise:

Step 1: Photo capture with guidance. We designed the photo capture flow to maximise image quality — consistent lighting, proper framing, removal of makeup. This was a UX challenge because every additional instruction added friction, but the AI’s accuracy was directly dependent on image quality. We ran multiple iterations to find the minimum viable guidance that produced usable images without causing users to abandon the flow.

Step 2: Analysis with transparency. Instead of presenting a single skin type label, we broke the analysis into visible dimensions: oiliness, hydration, texture, pigmentation, sensitivity indicators. This gave the dermatologists their nuance — the app was not saying “you have oily skin” in a reductive way. And it gave users a richer, more educational experience that felt more trustworthy than a single-word classification.

Step 3: Recommendations with rationale. Every product recommendation was accompanied by a brief explanation of why it was recommended, connected back to the analysis results. “Based on the dehydration detected in your cheek area, we recommend a hyaluronic acid serum” gave the user a logical chain from diagnosis to action. This dramatically improved trust in user testing because the recommendation did not feel arbitrary.

User Research: 15 Conversations That Shaped Everything

I conducted approximately 15 in-depth user interviews during the persona development phase. These were not usability tests — they were exploratory conversations about how Indian women currently made skincare decisions.

The findings were revealing and directly shaped our positioning:

Finding 1: The information gap was real but not uniform. Women in metros — Mumbai, Delhi, Bangalore — had access to dermatologists, beauty influencers, and a growing ecosystem of skincare education content. Women in tier 2 and tier 3 cities had dramatically less access. The information asymmetry was stark. A woman in Lucknow or Jaipur might have one dermatologist available, with a three-week waiting list, and her primary skincare information source was advertising.

This finding shaped our market positioning. Instead of competing with dermatologists in metros (a losing proposition — no app replaces a doctor), we positioned Skin Beauty Pal as a first-step guide for women who lacked easy access to professional advice. Not a replacement for dermatology, but a bridge.

Finding 2: Skincare routines were heavily influenced by social proof. The women we spoke to did not make skincare decisions based on ingredients or clinical evidence. They bought what their friends recommended, what they saw beauty influencers using, and what their mothers had used. Science-based positioning — “our AI uses convolutional neural networks to analyse your skin” — would alienate more users than it attracted.

This finding shaped our messaging. We led with outcomes (“discover what your skin really needs”) rather than technology (“AI-powered skin analysis”). The AI was the enabler, not the selling point. This was a lesson I had first learned in FMCG brand management: lead with the consumer benefit, not the product feature.

Finding 3: Privacy was a significant concern. Uploading a bare-faced selfie to an app felt vulnerable. Multiple users expressed discomfort with the idea that their unfiltered, makeup-free photos would be stored somewhere. This was not abstract data privacy concern — it was personal, emotional vulnerability.

We responded by building in explicit privacy controls: local-only processing where possible, clear data deletion options, and transparent communication about what happened to uploaded images. We also made the photography guidance explicitly permission-based — explaining why we needed a makeup-free photo and what we would (and would not) do with it.

The GTM Strategy: Ad to Website to App Install

Building the acquisition funnel for Skin Beauty Pal taught me several lessons about go-to-market for AI-powered consumer products.

The funnel was straightforward in structure: paid digital ads drove traffic to a landing page, the landing page drove app installs from Google Play, and the in-app experience drove engagement and retention. Simple in concept, complex in execution.

The ad layer. We tested multiple creative approaches and discovered that before-and-after imagery (showing skin analysis results) dramatically outperformed product screenshots or feature-focused ads. Users responded to the promise of personalised understanding — “see what your skin really looks like” — rather than generic skincare advice.

The landing page. The landing page served a dual purpose: convince users to install the app, and set expectations for what the app would and would not do. This second purpose was critical. If users installed the app expecting it to cure their acne, they would churn immediately. We needed to promise value without overpromising outcomes.

We settled on a positioning that emphasised three things: understand your skin better, get personalised recommendations, build a routine that works for you. Notice the verbs: understand, get, build. They imply a journey, not a miracle. This subtle framing reduced the expectation gap that is the biggest killer of consumer AI products.

The app install-to-activation flow. The most critical metric was the conversion from app install to completed first analysis. A user who installed the app but never took a photo was a lost user. We optimised this flow ruthlessly — reducing the number of screens between install and photo capture, simplifying the onboarding, and creating urgency (“see your personalised skin analysis in 60 seconds”).

Our activation rate — defined as the percentage of installs that completed a first analysis — reached levels that the team was proud of, contributing to my receiving the Best Performance Award for Q2 2019-20.

What AI Consumer Products Get Wrong

Working on Skin Beauty Pal crystallised a perspective I have been developing since: most AI consumer products fail not because the AI does not work, but because the product around the AI does not work.

Specifically, I see three recurring failure patterns:

Pattern 1: Technology-led positioning. “We use deep learning” or “powered by AI” is not a consumer benefit. It is a feature description. Consumers do not care about the technology — they care about what it does for them. The most successful AI consumer products I have studied market themselves on outcomes, not technology. Google does not say “powered by AI.” It says “find what you need.” The AI is invisible.

This connects to a principle I learned early in my career when working on FMCG brand launches: lead with the benefit, not the ingredient. “Contains hyaluronic acid” means nothing to most consumers. “24-hour hydration” does. The same principle applies to AI products.

Pattern 2: Overestimating user trust in algorithms. Users do not automatically trust AI-generated recommendations. Trust is built through transparency (showing how the recommendation was generated), consistency (delivering accurate results repeatedly), and social proof (showing that others have benefited). Launching an AI product and expecting users to trust it immediately is naive.

Pattern 3: Underinvesting in the “last mile” of user experience. The AI generates an output. Now what? If the output is a skin analysis, the user needs to know what to do with it. If the output is a product recommendation, the user needs to be able to purchase those products. If the output is a routine suggestion, the user needs support in following it. The “last mile” — connecting AI output to user action — is where most AI consumer products fall short.

The Dermatologist’s Lesson

I want to return to the frustrated dermatologist from the opening of this post, because her feedback embodied the most important lesson I took from the Skin Beauty Pal experience.

When she criticised our skin classification as “beauty-magazine level,” she was making a point about depth. The AI could detect surface characteristics — oiliness, visible texture, pigmentation. But skin health is multi-layered, influenced by hydration levels, barrier function, microbiome balance, and hormonal factors that are invisible in a photograph.

The honest answer was that our AI could not match a dermatologist’s assessment. No consumer image-processing technology could. But that was not the right benchmark. The right benchmark was: could the app provide more useful guidance than the consumer would have without it?

For a woman in a tier 2 city with no access to a dermatologist, the answer was clearly yes. An AI-powered analysis that correctly identified visible dehydration and recommended a hydrating serum was dramatically better than the status quo of choosing products based on advertising or hearsay.

This framing — not “as good as a doctor” but “better than the alternative” — became central to our positioning. It was honest, it was defensible, and it set appropriate expectations. Most importantly, it kept the dermatologists engaged with the project rather than feeling that we were trying to replace them.

Building in a Domain You Care About

One personal reflection I want to share: working at the intersection of beauty, technology, and health was energising in a way that my previous startup work had not fully been. PitchNDA solved a real problem, but it was a professional infrastructure problem. Skin Beauty Pal was helping real people take better care of themselves. The user interviews were more emotionally rich. The impact felt more direct.

This matters because building products is hard, and the domain you build in affects your stamina for the hard parts. When a frustrating bug delayed our Google Play launch by a week, my reaction was different than it might have been on a product I cared less about. The domain sustained my motivation through the inevitable difficulties.

I share this because I think founders and product leaders underestimate the importance of domain fit. Skills are transferable across domains — I have moved from FMCG to startups to AI — but energy and resilience are influenced by how much you care about the problem space. Choose a domain that sustains you, not just one that compensates you.

Looking Forward

The AI skincare and beauty technology space has exploded since Skin Beauty Pal launched in September 2019. Every major beauty brand now has some form of AI-powered skin analysis or product recommendation tool. The technology has improved dramatically — modern models can detect conditions that were invisible to 2019 image processing.

But the core product challenges remain the same: building trust, setting appropriate expectations, connecting AI outputs to user actions, and remembering that the consumer does not care about the algorithm — they care about their skin.

The lessons I learned at Pers Active Lab — about domain expertise, consumer trust, and the gap between technical capability and useful product — have informed every product role I have taken since. Technology evolves rapidly. Consumer psychology evolves slowly. The product leaders who understand both will build the AI products that actually matter.