Loops Closes $3.5M Seed Round to Build AI Operating System for Commerce

Loops announced the completion of a $3.5 million seed round, with investors including a16z speedrun, NTT Docomo, Connecticut Innovations, Alumni Ventures and Monday Capital. The company says the funds will be used for product development, customer deployment capabilities and international expansion.

The size of the round itself is not striking; what is noteworthy is the product proposition behind it. Loops positions itself as an AI company building agentic commerce systems for brands and retailers. Its core argument is not 'another AI tool,' but that the real pain point for retailers is this: AI is entering e-commerce as separate workflows — one product improves search, another answers shopper questions, a third writes product descriptions, and yet another handles virtual try-on. Each is useful on its own, but retailers must stitch them together themselves.

Loops' answer is to unify AI search, shopping assistant, product information enrichment and virtual try-on into a single system that follows the shopper through the entire journey from discovery to purchase. In the press release, company president and co-founder İlker Selim Zorluoğlu said retailers do not need five AI products that each understand only one side of the customer, but a system that understands the product catalog, the business and shopper intent. This is the company's stated position, not an independently verified conclusion.

Loops says it currently has more than 70 customers and pilot projects, and claims 64% of pilot projects converted to paying customers. These two numbers are the main support for the company's growth narrative, but the source does not disclose the specific split between customers and pilots, nor the statistical basis or time frame.

More striking is the conversion data. In a comparison published by Loops, the purchase conversion rate of A101 sessions involving the platform was about 83% higher than the control group. The company itself notes this is an observational result, not a randomized controlled measurement. This self-limitation matters: it means the source does not explain how the control group was selected or whether other variables were excluded, so the number cannot be treated as causal proof of effect.

Turkish supermarket chain A101 and fashion retailer Mavi are Loops users; Mavi uses it for AI search and shopping assistance. These customer relationships are factual information disclosed by the company, but the conversion rate and customer-scale numbers above all come from Loops' press release and its own statements; no independent third-party verification appears in the source.

Loops currently offers product modules including AI search and shopping assistant, marketplace assistant, product information enrichment and virtual try-on. According to the company's description, these modules share the same product knowledge, business context and customer journey: better catalog data improves search, search provides context for the assistant, and customer conversations reveal where product information is weak.

The logic of this design is to let each part reinforce the others rather than operate in silos. Shoppers describe their needs in natural language, and AI search bypasses keywords, filters and the retailer's internal catalog logic; the built-in shopping assistant answers questions as they arise and carries the service conversation across connected channels; product information enrichment rewrites descriptions, attributes and catalog data so products are readable to both customers and AI systems; virtual try-on attempts to close the last gap between browsing and purchase.

What needs to be distinguished: the module list and product positioning are the company's own statements, while 'each part reinforces the others' is the company's causal explanation; the source contains no independent effect measurement to support this mechanistic claim.

Loops' starting point was a concrete frustration: a co-founder, while buying a gift for his two-year-old daughter in Bali, messaged a retailer and waited hours without a reply, ultimately not making the purchase. The founder then discovered a problem bigger than slow customer service: shoppers are already telling AI in natural language what they want, while most e-commerce still runs on keyword search, filters and separate customer service queues.

Nazir built Loops' first prototype over a weekend, and that prototype later grew into a company. The company says it has handled millions of shopper experiences. This origin story is the company's own account, used to explain why the product is built around a 'continuous journey' rather than point tools; the 'millions of shopper experiences' claim likewise has no third-party verification.

In the press release, a16z Speedrun investment partner Fareed Mosavat said Loops' pilot conversion and early customer expansion show that brands want a platform that can cover more of the shopping journey. This is an investor's public endorsement and reflects his own judgment, not an independent audit of Loops' performance.

Notably, the source does not disclose whether the $3.5 million round has actually closed, nor the valuation or equity terms. For a seed round, the absence of this information is not unusual, but it means the statement 'completed the funding' remains at the level of a company announcement in the source.

Loops plans to extend its vertical AI operating system model to the financial services and energy industries. The company's argument is that the durable opportunity in enterprise AI belongs to systems deeply embedded in a single industry, carrying the knowledge, integrations, agents and controls needed to run real workflows, rather than a general model with a thin wrapper. Its playbook is the one already validated in commerce: start from a specific problem with clear value, work closely enough with customers to understand how the industry actually operates, then productize the repeatable parts.

At the same time, Loops says it is building task-specific small language model capabilities, on the grounds that domain knowledge and deployment constraints make focused models more suitable. But the source does not disclose these models' specific capabilities, training methods or release timing.

Expansion into financial services and energy is currently only a plan; the source gives no timeline or specific progress, nor any evidence that the commerce model can be replicated in these industries.