Building a $2M ARR product after his $8M ARR product failed overnight

Some businesses do not fail. They get switched off by someone else's decision.

Ruslan Leteyski is Bulgarian, based in Sofia, and started as a UI and UX designer who taught himself to code because he could not afford to hire anyone. In 2017 he built Checkout X, an alternative checkout for Shopify stores, and bootstrapped it to $8M in annual recurring revenue. Then Shopify changed its platform, locked down the checkout, and the product died. Years of work vanished because someone else changed their mind.

He built and micro exited another Shopify app, Vanga AI, after that. Then in 2024 he cofounded Zipchat, an AI sales agent for e-commerce stores. It handles any conversation between a brand and a customer across any channel, combining every conversation into one place with a single agent that only needs to be trained once.

Zipchat currently handles 700,000 conversations a month, growing about 10% week over week, and is about to cross $2M ARR.

Here is his account of building it, including a genuinely honest admission that there was no clean growth story behind any of it.

Zipchat is Ruslan's fourth business, and he is direct that it did not begin with any kind of revelation. He wanted to start something new, he had experience in e-commerce, ChatGPT had become the center of attention, and his reasoning was simple: build a ChatGPT wrapper, learn about AI, and hopefully make some money along the way.

The building process itself was straightforward. The team built the most minimal version of the product possible, launched it, started running ads on the Shopify App Store, gathered feedback, and gained momentum from there.

Zipchat runs on Ruby on Rails, Hotwire, and Postgres, leveraging RubyLLM for the AI implementation. On the model side, they started with OpenAI, moved to Anthropic, and now work with all the major labs.

The stack itself has not changed dramatically, but the development process has. With AI writing most of the code now, the team has shifted its entire development philosophy to match. Peer reviews are gone, and readability and maintainability are no longer enforced the way they once were. Instead of a Head of Engineering supervising a team of developers, each developer now effectively acts as the Head of Engineering for their own agents.

Ruslan did not initially realize that building a wrapper would be as difficult as it turned out to be, nor that AI would go on to disrupt the shopping experience as thoroughly as it has.

Financially, he was fortunate enough not to have to worry about money for a while, which let him focus on building something genuinely interesting. But the costs were significant. His partners required small salaries, GPT-4 access was expensive at the time, and the team ran ads on top of it all. The result was running at a loss of roughly $20k a month for almost a year.

The team ground through it and eventually began breaking even. Revenue based financing, loans structured against MRR, helped significantly once they secured it. They stayed deliberately lean until hitting $50k MRR: Ruslan built the product, Carlo handled sales and support, and Luca ran ads. They were reluctant to hire anyone until they could break even and genuinely afford to.

Zipchat sells plans based on AI replies. The more usage a merchant generates, the more revenue grows, and merchants who go over their limit are charged additional usage fees on top.

To expand customer revenue, the company needs merchants to grow their own usage, so they employ a land and expand model. Merchants typically start using Zipchat for website chat, then over time add it to search, product pages, social media, email, and other surfaces. The business model has not been heavily iterated on until now, though the team is beginning to run A/B tests to see what performs best going forward.

Ruslan is unusually blunt on this point: there is no growth story here. No growth hack, no viral moment, no single channel that cracked things open. It has been a slow grind, and the team keeps grinding. Their marketing assets are long term plays where every action compounds over time:

Cold outreach through emails, calls, and conferences has helped build awareness. The team buys data from sources like Store Leads and Similarweb, and sends more than a million emails a month. Alongside the awareness work, they have also had some success with AppSumo, Facebook ads, and Google ads.

The only real strategy Ruslan names is to never stop shipping and selling: work on the product, and talk to customers.

AI agents are doing the work of full GTM teams. Pipeline, content, customer service. The startups using them aren't waiting for headcount approvals.

Get the free Practical Guide to Agentic GTM for Startups and see exactly where to start.

Unlock the Playbook

The biggest challenge Zipchat still faces, in Ruslan's own words, is having the least creative business idea of the moment. More than 100 competitors have launched in the Shopify App Store alone, with likely thousands more worldwide. Even if Zipchat is better than everyone else in the category, each competitor only needs to acquire two customers for that fragmentation to compound and work against them.

Even so, Ruslan chose this business because he wanted to learn, and he considers it a genuinely good space to be in. All of e-commerce is being disrupted right now, and he sees real value in being among the people figuring out how that disruption should actually happen, rather than watching it happen from outside.

Ruslan's clearest advantage this time around is having cofounders who are as obsessed with winning as he is. In all of his previous businesses he was a solo founder and had to figure out everything alone. He did not want to repeat that, so this time he found people who are, in his own framing, a thousand times better than him in their respective fields.

Beyond that, his other real advantage has simply been talking to customers and solving their actual problems. His advice is to largely ignore startup books, generic advice, and the competition. What matters is the team you have and the customers you serve.

Pick an idea space, not a single idea. Choose something you are genuinely interested in and want to learn about. Even if that specific attempt fails, it will give you inspiration for the next one.

Do not think. Just do. Launch the product, try to sell it to someone, and repeat. If you cannot sell it, ask why not. If you sell it to one person, figure out how to sell it to ten, then a hundred. It really is that simple, and the only way to learn is by doing.

All it takes is a bit of luck. His previous business was more successful than this one, even though his execution this time has been considerably better. You might get lucky on the first attempt despite having no experience, or you might not get lucky for a long time despite doing everything right. The only way to get lucky is to keep playing.

Ruslan wants to build something that makes a real impact, something people across the world use at scale, something that genuinely makes the world tick. He does not have a precise definition of what that means, but he suspects a billion AI conversations would probably qualify.

You can follow along on X and LinkedIn, and check out Zipchat.