[devit]

Assuming this can recommend clothes that someone is going to immediately love based on previous purchases, this seems incredibly valuable for all the fashion e-commerce sites starting from Amazon, and could in fact turn any large site with this technology into a monopoly at least until someone replicates it.

Of course it needs to actually work though, which is not clear from the article (categories like "comfy" aren't going to cut it, you probably need a sophisticated deep learning approach on product images plus brand identity data and maybe Instagram posts with a lot of training data).

Obviously you can also run a store or an affiliate-based site yourself with it, but the problem is that you are going to be missing data on customer's taste; maybe you could exclusively cater to people who love posting their photos on Instagram, connect to their Instagram account and understand their fashion taste from their posted photos - or you could even support imitating someone else fashion's taste by looking at their Instagram profile.

[ezzzzz]

Yeah, this is sort of confusing to me. I know that B2B/SAAS is a preferred model for VC these days, but I'm sort of baffled that you wouldn't just take the tech that's been developed (assuming it works) and just integrate it with your own e-comm platform. Like, couldn't something be bootstrapped here? Maybe add a subscription/service model to create revenue?

[josephjrobison]

I agree with you, the product looks pretty cool and promising, and seems it could be a good fit within Amazon or something with a vast array of clothes.

[saalweachter]

There's kind of three problems you run into with that sort of thinking.

1. Is there a difference between great recommendations and random recommendations?

Maybe purchases are bottlenecked on money rather than desire, so your customers can already find as much stuff as they can afford to buy. Maybe your customers want to browse and look at literally every product on your website and will find it on page 1 or 100. Maybe your customers need to see a lot of options to realize how much they like the perfect recommendations, so showing them what they'll end up purchasing first doesn't really make a difference. Maybe your customers are lazy and will buy the first shirt you show them regardless of how good it is.

2. Can you get by being dumb if you have enough data?

A company like Amazon has soooo much data on shopping preferences based on past purchases. They can construct extremely naive models from their 20 years of logs and expect them to perform very well given your ten years of personal shopping data with them.

3. Does your superior algorithm create a moat?

For any interesting problem, you can get like 80-90% of the quality while spending 1% of the programmer-hours as the best of class. So if you're best of class, does your extra quality buy you anything, or is 80% as good as you good enough for most customers? Eg, can your potential acquirers just halfass something and get all the value you provide for a fraction of the cost, or do they need your extra years of experience to compete?