Loops Raised $3.5 Million to Make Online Shopping Less Like a Spreadsheet

There is a special kind of online shopping failure where you know exactly what you want, the retailer technically sells it, and the website responds by asking you to choose between 47 filters that all mean “maybe.” You type “a gift for a two-year-old who likes animals,” and the store offers a 900-word product description, three broken images, and a button labeled “compare.”

Loops would like to retire this experience. The San Francisco startup announced a $3.5 million seed round on October 12, backed by a16z speedrun, NTT Docomo, Connecticut Innovations, Alumni Ventures, and Monday Capital. Its product connects AI search, product discovery, shopping assistance, customer operations, product enrichment, and virtual try-on into one system for brands and retailers.

That is a lot of nouns to fit under one roof. It is also a surprisingly coherent answer to a real retail problem: every company wants an AI shopping experience, but most of them are assembling one feature at a time like a child building a spaceship out of refrigerator magnets.

The catalog is the part AI keeps pretending is already solved

Loops’ basic thesis is that online retail is fragmented in a very specific way. A retailer might buy one tool for keyword search, another for chatbot support, another for generating product descriptions, and another for virtual try-on. Each tool knows one slice of the customer and none of them knows what happened five minutes ago.

Loops puts the retailer’s product catalog, business rules, and customer context underneath all of those experiences. A shopper can describe what they want in ordinary language, receive product suggestions, ask follow-up questions, get help across connected channels, and—if the retailer has the relevant feature enabled—see a virtual version of the product on themselves.

The plumbing is the point. If the product data is incomplete, the search is bad. If the search is bad, the assistant confidently recommends nonsense. If the assistant recommends nonsense, virtual try-on merely gives the nonsense better lighting. Product enrichment—the work of filling in attributes, descriptions, and structured data—sounds like the least glamorous part of the pitch because it is the least glamorous part of the pitch. It is also where the whole system earns the right to be useful.

This is the same reason AI shopping agents are becoming an infrastructure fight, not just a chatbot novelty. Whoever understands the shopper’s intent, the retailer’s inventory, and the eventual transaction gets to sit in the valuable middle. The middle is where software goes to become a business model.

A failed birthday present becomes a venture thesis

Loops says its origin story began in Bali, when co-founder İlker Selim Zorluoğlu tried to buy a gift for his two-year-old daughter, messaged a retailer, and waited hours for a response. He never completed the purchase. The founders then built a prototype over a weekend, which is exactly the sort of sentence investors enjoy because it compresses customer pain, founder urgency, and a flattering amount of sleep deprivation into one paragraph.

The more interesting detail is what happened afterward. Loops says it now has more than 70 customers and pilots, has handled millions of shopper experiences, and converted 64 percent of pilots into paying customers. Those are company-reported figures, so they deserve a sensible eyebrow rather than instant worship. Still, they suggest the product is being tested in actual retail environments rather than merely demonstrated to a room full of people who already believe every website should have a personality.

At Turkish supermarket chain A101, Loops says sessions using its platform showed roughly 83 percent higher observed purchase conversion than a comparison segment. The company explicitly calls that observational rather than randomized, which is a small but meaningful sign that someone in the room has met a statistician. Fashion retailer Mavi is using Loops for AI search and shopping assistance, letting customers search by style, occasion, or fit instead of trying to guess the exact taxonomy invented by a merchandising committee.

Retail AI has been steadily moving in this direction. Replenit’s retail decision engine tries to help brands choose who to contact and when, rather than firing another discount at everybody who once looked at moisturizer. Amazon turned the product page into a talking shopping show. Deliverect gave restaurant menus a revenue strategy, because apparently fries now require an autonomous agent.

Loops is betting that the next step is not another clever feature but a shared layer that lets all the features remember why the customer showed up.

“Operating system” is doing a lot of lifting here

There is a moment in almost every enterprise AI announcement when a useful product is asked to become an operating system. This is often how a search tool acquires a roadmap, a roadmap acquires a platform strategy, and a platform strategy acquires the confident facial expressions of three people standing beside a screen.

Loops begins in commerce, where the value is at least legible. Better search should help people find products. Better product data should help machines describe those products accurately. Better assistance should reduce the number of customers abandoned between “I have a question” and “fine, I’ll buy it somewhere else.” That is not random feature confetti. The pieces can reinforce one another.

The ambition becomes harder when Loops expands beyond commerce into financial services and energy, as the company says it plans to do. Those industries have different data, different workflows, different regulatory expectations, and different meanings of “please recommend something.” A shopper can survive a bad virtual try-on. A bank customer probably does not want their loan eligibility handled with the same cheerful improvisation.

The company is also building specialized small-language-model capabilities for tasks where domain knowledge and deployment constraints matter. That is the sensible version of the operating-system dream: use the right model and controls for a bounded job. The less sensible version is putting one universal agent in charge because the pitch deck has run out of empty boxes.

The retailer wants magic. The retailer also wants a rollback button.

Loops has a good shot at being useful because it is aiming at a buyer who is already tired of assembling the stack. A retailer does not necessarily want five AI vendors, five data integrations, five dashboards, and five account managers explaining why the customer journey is technically omnichannel but emotionally a scavenger hunt.

But a unified system also concentrates the failure. If one shared intelligence layer misunderstands a catalog attribute, the error can travel from search to assistant to support to conversion analytics before anyone notices. The more surfaces Loops owns, the more important it becomes to show retailers what the system knows, what it inferred, and what it is allowed to change.

That is where the company’s operational discipline will matter more than the phrase “agentic.” Retailers need permissions, logs, human review, reliable integrations, and an easy way to stop the machine from promoting a discontinued sandal to the front page because one person in Ohio bought it at 2:14 a.m. The future of commerce may be conversational. The incident-response process should remain refreshingly boring.

Verdict: a promising little rocket with catalog-shaped homework

My verdict: promising little rocket.

Loops has a real customer problem, a product that hangs together, early deployment evidence, and a founder story that starts with a missed gift rather than a mystical encounter with a trillion-dollar market map. I mean that as both a joke and a compliment. The company is trying to make digital shopping behave more like a competent store associate: understand what the customer means, know what is actually in stock, answer the follow-up, and stay present long enough to help the decision happen.

The risks are equally clear. “AI operating system” is a large coat for a young company. Retail integrations are tedious, product data is feral, and every conversion claim eventually meets the awkward human questions of attribution, seasonality, discounts, and what shoppers were going to buy anyway. The expansion beyond commerce may be smart later, but it should not distract from proving that this first market can support a durable business.

Still, I like the shape of the bet. Loops is not asking retailers to admire a chatbot. It is asking them to stop making customers translate their intentions into the dialect of a database. That is a worthwhile ambition, even if the database remains unconvinced. If the company can keep the system grounded in trustworthy catalog data, measurable outcomes, and a human override that is easy to find, it may build something more valuable than a flashy shopping assistant: a less exhausting way to buy the thing you already meant to buy.