Fetchr | Thesis on AI x Shopping
Why we moved on from Fetchr and what we learned.
Why We Exited Fetchr
I’ve spent the last two years trying to build a better online shopping experience. We built several ideas in shopping, an AI shopping copilot, a livestream marketplace, creator storefronts, virtual try‑on, and finally an AI personal‑shopping agent which we called Fetchr.
Ultimately, we built something that a small group of people loved using, but couldn’t scale and decided that an exit was in our best interest. We learned a lot on what it takes to build a product people love and on the shopping space as a whole. Here’s a simple summary of our learnings.
1. Why we started Fetchr
When we left school and joined Y-Combinator, my co‑founder and I bounced between ideas: restaurant discovery, AI customer analytics, spend management. We talked to users for two weeks at a time but never found conviction in any idea or problem space.
During the batch, Paul Graham told us we were “weird for working on boring ideas” and should build what actually excited us. For me, that meant the intersection of software, AI, and shopping. People waste hours hunting for the right item and often just give up or don’t buy anything; AI should be able to help them discover what they want.
2. What Fetchr Did
Fetchr asked users for context, we collected their Instagram, Pinterest boards, past buys, measurements, current needs. We used that data to curate picks, let customers approve what they liked, then bought and shipped the items in one checkout.
Because we handled the purchase, conversion was great. Users didn’t want to scroll endlessly; they wanted items they would love and to click “yes” and be done.
3. Why It Still Failed
Logistics creep. A single checkout from multiple brands means multiple shipping and return labels. We had to consolidate shipping and handle every refund. That pushed us from pure software into logistics operations.
Thin margins. We paid retail, drop‑shipped, and lived on affiliate fees. Those fee’s per order don’t cover warehouses, customer service, and return freight unless you hit Amazon scale.
Unit economics catch up. High touch plus low take‑rate is a slow‑motion train wreck. Growth hides it for a while; returns expose it.
Despite this, I still believe that at scale, one could make the economics/business model make sense. However, the company looks more like an online retailer (like Farfetch) than a pure software company.
4. Core Lessons
In shopping, only thing that matters is the purchase. Engagement without transactions is not a business.
If you own the cart and checkout, you inherit the supply‑chain headaches. Software margins vanish.
5. The Market as I See It
AI finally works: ChatGPT can parse “something I can wear to Ibiza this summer under $300,” and Doji’s virtual try‑on looks real. The hard part isn’t the technology or product, it’s building a long term enduring business.
The battle for mindshare. Google and OpenAI embed shopping inside habits people already have. The advantage startups have is focusing on a singular experience tailored for shopping for clothes. They must create new habits from scratch, a user must think “I want new clothes” = “let me go to XYZ”.
Making profit. Discovery apps fundamentally have a lower take rate on the items they sell than a retailer. To justify handling orders, logistics (shipping + returns), and customer service, you need to reach mass scale to even think about profitability. On the other hand, if you choose not to handle this and just do discovery, the user experience will suffer and it will make it very difficult to reach mass adoption (I could be wrong but I would bet on this).
6. Where Opportunity Still Exists
The next winners will accept that:
Owning checkout means owning inventory or logistics—vertical integration, consignment, or warehouses.
New business models matter more than new models of AI. Subscriptions, private‑label capsules, or revenue share on retained customers can realign incentives.
AI creates a unique opportunity for the right founder to not only reinvent the product layer but reinvent the entire stack for an online retailer. This vertical integration across logistics and the platform will also create a durable moat.
7. What’s Next for Us
We’ve exited Fetchr and are taking these lessons forward as we explore the next thing. The problem—helping people buy things they’ll love—remains huge, and I hope someone solves it. If you’re building in this space, I hope our scar tissue saves you time.
Thanks for reading and if you’re interested in what I do next, you can follow me on X. Feel free to reach out if you’re also explaining, I love other smart and ambitious people!
kudos on cutting your losses, and consolidating such crystalline learnings. on to the next one.