When AI hits the corner store: Bulgaria’s nFuse takes on a $5 trillion market

After a decade of massive investment in digitalization, the FMCG (fast-moving consumer goods) industry has hit an unexpected barrier: small retailers are overwhelmingly refusing to use official B2B apps. Although the fragmented retail segment (small neighborhood stores and kiosks) is valued at over $5 trillion globally, only 15% of orders in this segment go through digital platforms. The remaining 85% still rely on flyers, phone calls, and visits from sales representatives.

It is precisely in this technological gap that the Sofia-based startup nFuse, founded by Stoyan Ivanov and Stefan Radov – former executives at Coca-Cola – positions itself. Instead of trying to change the habits of the “person behind the counter,” they adapt the technology to them.

We have secured a $2 million investment, co-funded by Eleven Ventures and LAUNCHub Ventures. A significant portion of the funds will be directed toward product development and R&D, including expanding AI capabilities, integrating payments, and developing predictive intelligence on the platform,” the company told Economic.bg.

An Army of AI Agents

The traditional approach of major brands has been straightforward: they build complex portals with catalogs and analytical tools, after which sales representatives try to “educate” small shop owners on how to use them.

The industry designed software for people at headquarters who want charts and tables. But on the ground, the priority is speed. The retailer doesn’t want to learn a new interface just to stock ten cases of beer,” commented Stoyan Ivanov.

The result of this misaligned focus is a statistical failure: between 80 and 95% of B2B e-commerce projects in the sector fail to meet their targets for active users.

nFuse finds a way to solve this problem by entering the channels that merchants already use – WhatsApp and Viber. The key innovation, however, isn’t in the chat itself, but in how information is processed beneath the surface. The startup isn’t offering a new app; their platform is an “invisible” layer beneath the otherwise familiar interface of chat apps. In short: the merchant sends a photo of an empty shelf, a voice message, or a short text, and nFuse’s AI model converts this unstructured input into a precise order in the manufacturer’s ERP system.

We’re not a chatbot with menus. nFuse uses LLM models (Large Language Models) that analyze free-form text, voice messages, and photos in real time. The retailer writes or speaks naturally: “I need two cases of beer and chips,” and our AI agents interpret the request, match it against the distributor’s catalog – which may contain thousands of SKUs (stock-keeping units – ed.) – and generate a structured order,” the founders explain to Economic.bg.

The technological challenge here lies in the precision of the artificial intelligence. The system must recognize slang, poor sound quality, and blurry photos, matching them with specific SKU codes from inventory. The company claims that this model increases conversion rates to 70% – a dramatic jump compared to the industry standard.

nFuse’s technological architecture does not rely solely on off-the-shelf solutions like GPT-4 or Gemini. The company has developed a hybrid model:

We use leading LLM models as a foundation, but we’ve built our own layer of 30+ specialized AI agents optimized for the FMCG context. They are trained on specific industry data – product catalogs, retail jargon, and regional variations in product names. The system learns from each buyer’s context – if a salesperson corrects a mistake once, it doesn’t happen again.”

How to Bypass IT Bureaucracy

One of the biggest hurdles for large corporations is the cumbersome integration of new software, which often takes a year and a half. nFuse claims to shorten this process to two months through a combination of ready-made connectors to major ERP systems (such as SAP or Oracle) and a specific API-first architecture. This integration layer “translates” unstructured chat requests into standardized, warehouse-ready orders, eliminating the need for major IT projects on the client’s side.

Who pays the bill?

To ensure widespread adoption, nFuse relies on a strategy where the end user (the retailer) bears no financial burden.

The manufacturer or distributor pays, not the retailer. For the retailer, the service is free, which is key to high adoption rates,” the company explained.

Monetization does not follow the standard SaaS model with a fixed fee, but is tied to transaction volume and actual results achieved, such as revenue growth or the number of successfully processed orders. The cost per order drops below $1, compared to 5 to 20 times higher costs through traditional channels, the company claims.

With the capital raised, nFuse plans not only to expand geographically into Europe, EMEA, and the U.S., but also to completely transform commercial relationships. The next step is to integrate payments and microloans directly into the chat, turning Viber and WhatsApp into full-fledged financial tools for small businesses.

In a world overflowing with complex apps, the Bulgarian startup is banking on a paradox: the most powerful technology is the one that remains invisible to the user.

Translated with DeepL.