YC-backed startup Proven wants to make it suck less for women to find skincare that works for them. The co-founders are taking what they describe as a “rational, logic-based” approach to figuring out which ingredients might be most appropriate for each individual.
As a TC Disrupt battlefield founder once memorably put it during her on-stage pitch, the beauty industry makes a whole lot of money from a whole lot of BS. And skincare falls squarely into the ‘full of it’ category, with its expensively marketed pseudoscientific claims touting ‘miracle’ fixes that most definitely aren’t.
Proven’s co-founders, Ming Zhao and Amy Yuan, say frustration when battling with this BS via their own skincare issues ultimately led them to found the business together. Zhao had had a stressful job in private equity which she credits with “really wrecking my skin”, while Yuan suffered adolescent skin problems and also has allergies that can affect it.
“After trying numerous products and investing — I saw it as an investment, in expensive ‘miracle’-promising products — nothing really worked for me,” says Zhao. “So I became very frustrated and I felt betrayed by our beauty industry. And eventually what actually worked for me were customized products that were made for my by a few different facialists. So that’s how the optimize idea of tailoring products to exactly someone’s situation, someone’s skin, first came to my mind numerous years ago.”
Yuan’s computational physics background informed the data-focused approach they’re taking with Proven. “I’ve done a lot of big scale supercomputing simulations,” she says. “And, I thought, given my background why don’t I just write an AI engine that gathers reviews for me to find skincare products that are automatically fitting to my skin. And when I talked to Ming about it we immediately had this spark — and started crawling data.”
Their core idea is to see whether deep learning and machine learning algorithms can distill useful information from millions of online testimonials for skincare products, plus a much smaller subset of publicly available peer reviewed scientific research papers — turning a mountain of what is obviously very variable data into, what they hope, is a formula for programming customized skincare products that work.
They’re focusing on skincare purely for women because it’s women who’ve written the millions of online product reviews underpinning this data + AI play.
“The average person spends 45 minutes to 1.5 hours researching products before they buy any beauty products and even after they buy based on the research that they’re able to do, 55% of people are still unsatisfied post-purchase. And that’s because of the proliferation of information that’s out there. No single person is capable of reading the amount of information there is in order to make a sound decision,” argues Zhao when asked why they think their approach can work. “Which is why we’ve built the largest database of beauty.”
Their database combines data on hundreds of thousands of skincare products culled from millions of users reviews. At this point Zhao says they’ve used their AI engine to analyze more than 8 million reviews and testimonials — “of basically anybody who’s bought a skincare product, a beauty product and has written a comment about it online”.
“In this database it also has more than a hundred thousand beauty products that have been talked about. So basically everything that’s on the market. As well as more than 20,000 ingredients — as well as 4,000 peer reviewed scientific articles on skin and on ingredients and on what works for skin,” she continues. “So it is not just reviews but it’s combined with scientific research.
“On our team we also have an award winning cosmetic chemist who is the person who helped to formulate all of our products. We also have dermatologist advisors on our team who put the human touch on top of the big database knowledge base.”
Potential buyers must first fill out a survey on Proven’s website, answering questions about things like their age, ethnicity, skin type and their skincare priorities. After which they’ll be emailed custom products they can buy — which will in turn be blended by drawing on Proven’s database of AI-distilled testimonials to match crowdsourced learnings to what an individual customer knows (or at least claims to know) about their own skin.
The service isn’t live yet — but will be soft launched in the US next week.
“The database is really powerful. It has all of the information and has more than 10 years of consumer testimonials on various skincare products,” adds Yuan. “One surprise that we had going into this space is how little research was out there on people’s skin and then what kind of ingredients would be effective on what kind of skin and in which environment.
“And then we feel like… why they’ve researched so little is because there’s not enough data to back it up — unlike pharmaceutical research where funding can go in and there’s clinical trials and a lot of different funding sources. Skincare is sort of in an awkward position.”
Globally, the skincare, beauty and cosmetics industry is estimated to be worth some $445BN at this point — a figure that’s only set to keep growing in an age of selfie obsession and perpetual digital self promotion.
So any skincare company that can come up with a slicker formula to help women find effective products could be a real game changer.
But, at the same time, there’s undoubtedly a lack of high quality data to drive genuine change. And without regulation of BS claims, well, misinformation is free to masquerade as eye-catching marketing. And that’s why pseudoscientific nonsense is so lucrative. And why there’s little incentive for the industry to change.
Proven’s founders say they’ve done a lot of cleaning and structuring of the data in their database before processing it for patterns. Even using fraud detection algorithms to try to weed out sources of fake reviews. On top of the cleaned and structured data they’re then applying various machine learning and deep learning algorithms to try to link particular ingredients with beneficial outcomes for different types of users.
But the big question is whether poor and/or low quality data — even if you’ve managed to scrape together an awful lot of it — can really lead to useful AI-powered decisions.
Where skincare is concerned, that remains unproven — unlike this startup’s name.
And with so many other unseen factors at play that can also affect people’s skin, such as diet, exercise, lifestyle, even genetic conditions, which won’t necessarily be being expressed within the limited confines of an online review, well, it’s just not clear whether anything of real worth can be distilled from such partial and fuzzy data.
Although there would certainly be poetic justice if the beauty industry ends up being successfully disrupted thanks to millions of user reviews debunking its not-so-miracle cures.
We’ve not been able to test Proven’s service at this nascent stage. And clearly it’ll take time for its own user testimonials to roll in. But if you look online, you’ll find skincare reviews are rife with dissatisfaction.
So even if all Proven offers is doing some of the legwork to help people decide what to buy (or avoid buying) that’s at least a partial incentive — given that many women will already be spending lots of their own time and money locked in a frustrating trial and error process of looking for skincare that works.
The co-founders also say they’re looking at ways to visualize the learnings they’ve extracted from their database — to try to make that information more accessible.
Away from the mainstream beauty market, the even more expensive skincare option is to pay for custom products from a dermatologist or other skin specialist. A route that can be effective, as it was for Zhao, but can also be prohibitively expensive. Certainly it’s not accessible to the mass market.
Zhao and Yuan say they want Proven’s skincare products to be accessible but they also argue that beauty products priced too cheaply can be perceived by women as ineffective or undesirable. So they also won’t be setting the price bar too low.
“One of our goals is to make beauty inclusive, and that is inclusive from many different angles — in that we’re not just making products for a certain subtype within a certain ethnicity. We want to be able to help many people with their skin issues, across ethnicities, across geographic locations,” says Zhao. “So in terms of pricing too we want to be approachable. But what is funny though, we know from our data that women don’t consider a product to be of high quality unless it is above a certain price bar. That’s just how we’re built. So we try to signal that our products are of the highest quality — because they are.”
For people with sensitive skin the challenge of finding effective skincare can be a full on nightmare. (Nothing says ‘unhappy customer’ quite like having paid for a emollient that actually makes your skin even worse.) So if Proven can narrow the risk of encountering irritants that’s also going to be compelling for at least a subset of consumers.
Though product customization can also be risky from a business point of view. Because if a personalized product ends up disappointing, that customer will have few avenues to explore to come back for more.
The mainstream skincare industry does claim to cater to different skin types. But its categories tend to be fairly broad-brush, and arguably just make matters worse by creating yet more skincare products which buyers need to factor in to ‘buy and try’.
The lack of regulation on the beauty industry also makes it impossible for consumers to be confident in any of the claims being made by any of these products. A ‘miracle’ snake oil can (and frequently does) sit on a shelf next to a more basically packaged and less expensive moisturizer that contains essentially the same ingredients.
The same could be true of ‘AI-distilled skincare’ too of course. So for now it remains to be seen whether Proven’s personalized skincare products end up delivering more effective skincare than the average pot of cream plucked off the shelf.
After all “personalized” is just another word that sounds good but doesn’t in itself mean very much. So “personalized skincare” may end up being just another nice sounding but hollow claim.
On the other hand, if their “AI engine” actually manages to distill some valuable intelligence from millions of product reviews it could be a very winning formula. Beauty industry product promises that don’t disappoint would be a disruptive innovation indeed.
Proven says it will initially manufacture its skincare products itself, in the US, using an ingredients philosophy the co-founders sum up as “just what you need and nothing that you don’t”. They are also excluding some common but controversial beauty product ingredients, such as SLS, parabens, alcohol, triclosans and animal byproducts. (Although their products are not equivalent to fresh cosmetics, such as the custom preparations you might get from a dermatologist, as they do include some preservatives.)
While they’re starting with skincare — offering a small range of day and night serums, toners and creams to begin with — Zhao also says they see potential to expand into other wellness products if the personalized touch flies.
“We’re starting with skincare but we’d love to do the same thing… within all of the wellness category, because there’s nothing more intimate for me personally and for a lot of women I know than their skincare, than their bodycare, than their haircare,” she says. “The things that they put on and therefore are absorbed into their body. So we want to help everyone to have a more personalized experience with these essential, important categories.”