AI personal stylist (my learnings from Better Wardrobe)

 
In the last 1 year, I might have seen 20 people interested in the idea of working on an AI personal stylist startup. I hate when YC uses the word ‘tarpit’ for ideas that everyone thinks should happen, but there’s no need - I think an AI personal stylist is a tarpit given how many people think of it and even then there’s no good AI stylist startup. Here’s my experience and thoughts about it.
I spent 3 months working part-time on this Mar-May 2023 with my friend Yash.
It’s not a big problem: to validate the problem we ran an experiment where we asked people to send some of their photos to give them feedback on how to dress better and what to buy next. Users who initially said, “Oh I really want this” didn’t feel enough value to pay $10 for this. This is a bad sign. We removed the paywall and made it free. Then the users felt friction in finding photos. Oh oh. So we went to their Instagram and collected photos ourselves. Then we sent them the report of the feedback, which either they didn’t even check at all or didn’t follow at all. This is a really bad sign. People who feel they have a bad dressing sense and need help also have found comfort in their way of dressing. There’s no urgency for them to look better.
It’s not even the best solution: The target audience is the “bad” end of the fashion spectrum who don’t understand/enjoy fashion. On the other end even if need input it’s rare and they’d prefer discussing with a friend. I used to think there were so many input variables that the output space would be also very large. It’s really not. Technically, you could just write a one-pager for the target audience on what the trends are and what they should buy to get to the top 25%. Here’s how I would write one for a guy in his 20s. Get the following
  1. 3x Solid t-shirts - (Black, white, one more)
  1. 2x Oversized Graphic t-shirts
  1. 3x half sleeves printed shirts
  1. 2x SemiFormal full selves shirts
  1. 2 jackets
  1. 2x jeans - (NO slim fit, either relaxed or baggy)
  1. 1x baggy cargo trousers
  1. 2x shorts
  1. 2x Sneakers (1 solid, 1 fancy)
I’ll maybe add brands or links directly too. This should solve 80% of users’ problems. Rest they’ll add with time based on their interest. In fact, I think someone should make a “Revamp your wardrobe box” service like this as a fun side project (hmu if you want to work on this with me). Once someone gets to the top 25%, going further needs interest in fashion, which if the users have they won’t need the AI stylist or it wouldn’t add enough value.
It’s a bad target audience: People who need guidance are the people who aren’t interested in fashion. These people don’t enjoy the act of spending time and money on fashion because they don’t see the value. This is a bad group to solve for. For example, if you recommend someone to wear a black t-shirt, they’d prefer a $7 Jockey T-shirt and not a $30 Zara t-shirt. The total money they spend and the time they spend on fashion is quite low.
It’s not easy to accept recommendations - The reason I got interested in this is because my sister who goes to a fashion school told me I dressed badly and gave me recommendations on what to buy. I slowly started to dress better. One thing my sister told me on Day 1 was to wear baggy fit and oversized clothes. As much as I respected her opinions, I did not accept that suggestion. It took me a year to actually start wearing baggy, which I was completely aware was the trendy and the right thing to do. Even now, I am slowly getting comfortable wearing it. Do you get what I am saying? Am I boring you?
 
Summary: People who need constant feedback and improvement in their outfits tend to be people who are rich enough to afford actual stylists. Everyone else - just ask your sister, female friends, girlfriend, or even ChatGPT is not that bad.
 
What I would do differently: The initial hypothesis was “Only rich people can afford a stylist, technology will make it accessible for everyone”. This implies people broadly want this, but can’t spend a lot on this. I would just try to validate this. I think we did a good job of not building the product and did everything manually which was enough to get most of these insights. We were slow in sharing the link to this experiment mostly because we were expecting it would magically reach the right people. Only slowly we realized that wouldn’t happen. I would just spam everywhere I can for people to come on my landing page and pay a small amount / small effort to indicate they actually want this. I think talking to a lot more people about how they shop would have helped for sure. A lot more here means at least 100 people. Broadly, for a consumer idea going forward, I’d set my goal to find the 100 people who LOVE your product. If I am unable to, then there’s nothing there.
 
Potential wedge: The thought of working for the “bad” end of the spectrum bad me think of reversing this and working for the other end i.e. power users who either spend a of time or money, ideally both, on buying clothes. My hunch is the way these people buy clothes has changed a lot in the last 5 years - people buy from Instagram, D2C brands, and thrift shops. These power users have also become more social with their clothing. The top FashionTech startups are largely the same as they were in 2015. Could it be the case that a new platform built for these top 0.1% users might make sense in 2025?
 
I am young, dumb, and broke, and might have written this poorly and derived incorrect insights. If you feel anything above could be better or you want to discuss more, please reach out.