Loops Raises $3.5M for an 'AI Operating System for Commerce': What Connecting AI Search, Shopping Assistance and Product Data on One Knowledge Layer Asks of Ecommerce Teams
Starting from Loops' seed round, backed by a16z speedrun and NTT Docomo, we look at what happens when retailers add AI one feature at a time, why tying AI search, shopping assistance and product enrichment to shared product knowledge matters, and how to read the early numbers.
Key Takeaways
- On October 12, 2026, retail AI company Loops raised a $3.5 million seed round from a16z speedrun, NTT Docomo and others. It is pitching an "AI operating system for commerce" that runs AI search, a shopping assistant, product enrichment and virtual try-on as one system on shared product knowledge
- The design philosophy matters more than the round size. Retail AI tends to arrive as separate products, each understanding only a slice of the customer. Connecting product data, search and conversation in a single loop also shapes how external AI agents see a retailer's products
- The published results are largely observational, and different figures have appeared for the same customer. The lesson for ecommerce teams is less about any one vendor and more about their own design: where the source of truth for product knowledge lives, whether conversation logs flow back into product data, and how impact is measured
Retail AI has arrived one feature at a time
Online shopping has become very good at showing people thousands of products and surprisingly bad at helping them choose one. Loops was built around the idea that AI can collapse those experiences into one continuous journey.
markets.businessinsider.comAI that improves site search, a chatbot that answers questions, generative AI that writes product descriptions, virtual try-on. Line up the AI products ecommerce teams have evaluated in recent years and most of them are single-purpose tools that cover one part of the buying journey. This announcement takes aim at that way of stacking things up.
San Francisco-based Loops announced a $3.5 million seed round on October 12, 2026. The investors are a16z speedrun, NTT Docomo, Connecticut Innovations, Alumni Ventures and Monday Capital. According to SiliconANGLE, a16z speedrun led the round. The money will go to product development, customer deployment capacity and international expansion.
The founders are CEO Ari Nazir and President İlker Selim Zorluoğlu, with Yusuf Bahadır as CTO. The company traces its origin to a trip to Bali, where Zorluoğlu tried to buy a gift online for his two-year-old daughter. He messaged the retailer with questions, waited hours without a reply, and never bought it. Nazir is said to have built the first prototype over a single weekend.
Loops had already announced an investment from a16z speedrun in December 2025. That release was datelined Istanbul and described the company as serving more than 50 brands across Türkiye, MENA and Europe.
Where the cost of stitching shows up
The core of Loops' argument comes down to this remark from co-founder Zorluoğlu.
The biggest mistake we see is treating AI as a collection of features. A retailer does not need five different AI products that each understand one slice of the customer. It needs one system that understands the catalog, the business and what the shopper is trying to do.
Why is a collection of features a problem? Each single-purpose AI product ingests product data on its own and maintains its own index or knowledge base. When the search vendor's index, the chatbot's FAQ and the description generator's output all live separately, three different "understandings" of the same product exist side by side. Updating material or size information in one place offers no guarantee it reaches the other two.
The other loss is that information surfacing in conversations has nowhere to go. Suppose a shopper asks the assistant whether a garment is machine washable and it cannot answer. That is a signal that the product data lacks a care-instruction attribute. If the chatbot and product data management are separate products, though, the signal stays buried in conversation logs.
Loops puts this loop at the center of its product. As the announcement describes it, better catalog data improves search, search gives the assistant context, and customer conversations reveal where product information is weak. On its website, the company explains that it structures catalog, attributes, sizing, brand information, FAQs and policies into a single semantic layer, and runs both search and the assistant from there.
Unite.AI's analysis offers the example of a shopper looking for an outfit for an outdoor wedding within a budget. Matching the word "wedding" is not enough. The system has to interpret the occasion, price limit, style preferences and product characteristics, then keep that context when the shopper asks about fit or alternatives. If search and the assistant are separate products, the assistant does not know what the shopper saw in search.
The same article notes that Loops' search can rank results around the retailer's revenue and margin goals. That is a design choice that lets business objectives influence which suitable items appear first. It makes sense for sellers, but how to reconcile it with the neutrality shoppers expect from a recommendation remains a question for the retailer deploying it.
The product data layer also decides how outside AI sees you
What makes this announcement relevant to agentic commerce, where AI agents find and buy products on a shopper's behalf, is how enrichment is positioned. Loops says it rewrites product descriptions, attributes and catalog data until products are legible to customers and AI systems alike. Its website also makes the point that unstructured catalogs effectively do not exist for AI assistants.
In other words, product data cleaned up for an on-site shopping assistant also becomes the material that external AI such as ChatGPT or Gemini uses to understand those products. A question the on-site assistant cannot answer is likely one an external AI agent cannot answer either. Internal conversation logs work as an inspection tool for gaps in the product data that outside AI relies on.
Loops is not alone in treating product data as the foundation for the AI era. Fanplayr's Catalog Intelligence and Commerce's (formerly BigCommerce) AI enrichment come from the same concern. The difference is that Loops sells enrichment not as a standalone product but as the layer underneath that makes search and conversation accurate. For the broader picture, see making product data agent-ready.
Marketplace support follows the same logic. Loops' Marketplace Assistant answers product questions on marketplaces such as Türkiye's Trendyol using the same product knowledge and brand voice. Even as sales happen across the website, app, WhatsApp and marketplaces, the design keeps one source of truth for product knowledge.
How far can the published numbers be read?
The figures cited as results make good headlines but need careful reading. Here are the main published numbers alongside where they come from.
| Metric | Reported figure | Source | What to keep in mind |
|---|---|---|---|
| Customers | 70+ (customers and pilots combined) | Press release | Split between paying customers and pilots not disclosed |
| Pilot-to-paid conversion | 64% | Press release | Cohort size, period and contract values not disclosed |
| A101 purchase conversion | Roughly 83% higher than the comparison segment | Press release | The company states this is observational, not randomized |
| A101 Ekstra conversion | Loops sessions at 2.55x the standard baseline | Loops case study page | No explanation of how this differs from the 83% figure in definition or period |
| Tchibo average order value | 13.6% higher | SiliconANGLE report | Not mentioned in the press release |
| Pricing | Not disclosed | None | Not stated in the announcement or coverage |
The headline example is A101, a major Turkish supermarket chain. The press release says sessions involving Loops showed roughly 83% higher purchase conversion than the comparison segment, while noting itself that this is observational rather than a randomized measure. SiliconANGLE, by contrast, wrote that A101 increased its conversion rate by 83% after integrating Loops, wording that reads as causal.
Unite.AI flags this clearly. Shoppers who choose to engage with an assistant may already have higher purchase intent. The comparison is an encouraging signal worth investigating, but it does not establish that Loops alone produced an 83% improvement. On top of that, the case study page on Loops' website says that for A101 Ekstra, A101's online arm, Loops sessions converted at 2.55x the standard baseline. That deployment went live in July 2026, and no explanation is given for how it differs from the roughly 83% figure in the press release in definition or period.
There are also caveats about competition. Forbes points out that the big AI models already let consumers search ecommerce sites in natural language, and that rivals such as Algolia and Bloomreach carry valuations running into the billions. Established site search providers are expanding into conversational features too, as with JD Sports adopting Algolia as its agentic commerce foundation. The all-in-one pitch is not exclusive to startups.
What NTT Docomo's participation means
For readers in Japan, NTT Docomo's presence among the investors stands out. Loops' website also displays the NTT Docomo logo as a backer, next to a16z speedrun.
The announcement, however, says nothing about entering the Japanese market or about a business partnership with NTT Docomo. International expansion is listed as a use of funds, but no target regions are named. What can be confirmed today is the investment itself, and any further meaning will have to wait for future announcements.
Questions to ask when designing your own AI stack
Whether or not a retailer ever considers Loops, the announcement is useful for reviewing one's own AI rollout.
The starting point is where the source of truth for product knowledge lives. Check whether search, chat, description generation and marketplace support each reference different data. If they do, fixing data in one place will not carry over to the others. Whether you buy an all-in-one product or own a product data foundation and feed each AI tool from it, a single reference point is the prerequisite.
Next, confirm there is a path from conversation logs back into product data. Questions the assistant could not answer, products that triggered repeated follow-ups, phrasings that returned zero search results. Whether product data owners review these records regularly and add attributes will widen the accuracy gap over time. Loops describing itself on its website as "one system that runs in a loop" reflects how central it makes this cycle to its value.
How to measure impact should also be settled before rollout. Vendor comparisons along the lines of "chat users convert at X times non-users" are affected by self-selection, as the A101 example shows. Where possible, agreeing before signing on a holdout test that shows the feature to only some visitors makes the post-launch decision much easier.
Finally, recognize that both the all-in-one and the best-of-breed approach carry costs. Salih Yılmaz, General Manager of A101 Ekstra, says thousands of trucks deliver the company's products every day but it does not build the trucks itself, and that AI orchestration is the same. Nazir likewise says he wants customers to buy a business capability they can depend on rather than assemble and maintain it themselves. The flip side is that an all-in-one platform deepens dependence on a single vendor and makes it harder to swap out individual functions. With pricing undisclosed, a total-cost comparison is not possible for now.
Conclusion
At $3.5 million, Loops' round is small by AI commerce standards. It still deserves attention because it puts front and center, as a product design philosophy, a critique of how retail AI has piled up feature by feature.
The next things to watch are whether results from randomized comparisons appear, whether major deployments follow outside Türkiye, and how site search leaders such as Algolia and Bloomreach tell their own all-in-one story. Whichever path a retailer takes, keeping one source of truth for product knowledge and feeding what conversations teach back into product data is also preparation for being chosen by external AI agents.
