Solving Retail's Hardest Problem

You can build a retail tech company today for a fraction of what it cost a decade ago. The infrastructure is cheaper, the UX practically builds itself, and AI models promise to add new discovery capabilities. But the thing that actually makes retail technology work — getting product data into a usable format — still costs exactly what it always has: one engineer, one retailer, one custom integration at a time.

Today, we’re launching a company to fix that. It’s called Octogen, and our first product is Cosimo.

I know this problem because I built one of those companies. In 2013 I founded Bluecore, a retail data platform that grew to a billion-dollar valuation and hundreds of millions in revenue. And the thing I learned is that the hardest problem was never the algorithms, the recommendations, or the UX. It was the data underneath.

At Bluecore, we had an entire engineering unit — Forward Deployed Engineers — whose job was to go client by client, writing custom integrations to ingest, normalize, and make sense of each retailer’s product data. Every retailer’s catalog was structured differently. Every integration was bespoke. It worked, but it was slow, expensive, and never truly solved — each new client meant starting from scratch.

Then AI agents started shopping. Not in a theoretical, conference-talk way, but in a real way. During the 2025 holiday season, AI-driven traffic to retail sites grew 693% year-over-year, according to Adobe. Consumers are discovering and buying products through conversations with AI, and with that comes an opportunity: agents that can parse and reason about products could fundamentally improve how people shop.

But the data problem got worse, not better. The messy catalogs that were already hard enough to wrangle for human-facing retail technology are even less compatible with how agents work. And companies building agentic commerce experiences are running into the same wall we hit at Bluecore: either you throw engineers at the data problem, or you push it onto the retailers. Either way, someone pays — in time, in cost, and in data that’s perpetually lagging behind reality.

A different approach

At Octogen, we started with the hardest problem first — the one that formerly required an entire engineering unit to solve one client at a time. We built automated infrastructure that ingests and standardizes product data at scale, directly from brand sources, without requiring any lift from brands or retailers. At the core is a category ontology — a structured understanding of how products within a domain relate to each other — that lets us normalize any catalog without custom integrations.

The result: 10M+ products, standardized across 10,000+ brands. No FDEs, no retailer data exports.

The other result: an incredibly flexible and usable dataset on which to build next-generation e-commerce technology.

What it unlocks

So what does a truly flexible dataset unlock?

Cosimo is a retail discovery experience built on this infrastructure — and it’s live today:

• Explore and refine using both natural language and structured browsing. You can search the way you’d describe what you want to a friend — “drop-shoulder dresses with taffeta” or “linen pants under $200” — and then refine with clickable filters across attributes, categories, and more. Most shopping tools force you to pick one mode or the other. Cosimo does both from the same foundation.
• It’s OK to care about brands. Try “pants made by Japanese designers” or “scarves like that thick Acne Studios one.” Or find a new brand by typing what you like — or some brands you like — into our brands explorer. In an era of easy distribution, brand identity matters more, not less. Cosimo enriches data at the brand level, not just the item level.
• Be specific about what you want — and what you don’t. Try clicking “More Like This” on a product result and picking your favorite attribute. Rather than a red top you like yielding 50 identical-looking red tops, you might choose similar tops with “bardot” necklines but no polyester. The system knows what makes each product this product — so specificity actually works.

To make this concrete:

• “Scarf in the style of Acne Studios thick scarf, under $200”: Google yields a variety of Acne Studios scarves. Cosimo yields scarves from brands like Acne Studios, with similar product traits.
• “More like this” on red pleated dress: Other shopping apps return dresses that look as identical as possible, in color and form. Cosimo lets you pick what you liked about it — the pleats, the bardot neckline — and find more of that, without insisting on features that don’t matter to you.

This isn’t just better search. This is what becomes possible when every product in a catalog shares the same structured language.

What’s next

We’re just getting started.

The same problem we described at the top — product data that costs one engineer, one retailer, one integration at a time — doesn’t just affect shopping experiences. It’s the problem every retailer will face as AI agents become a primary discovery channel. The infrastructure behind Cosimo is our answer to that, and we’ll have much more to share soon.

For now: try Cosimo and let us know what you think.

We’re adding brands every week, and if yours should be one of them get in touch.

— Mahmoud

You can subscribe here to stay tuned on updates. Our team is mostly engineers who don’t love writing, so we promise not to spam.