Enterprises mostly buy from names that carry decades of enterprise IT credibility, a long list of high-level executives, and solid references. A pre-seed startup building a multi-agent AI platform has none of the things listed above. 

And yet we closed enterprise deals. We entered a government pilot with Ukraine's Ministry of Justice and are now preparing for European expansion. Here, I want to share the lessons I learned about selling AI to enterprises before anyone had heard of you, and what I think investors should look for when evaluating early-stage AI companies doing the same.

The brand problem is real, so you'd better stop pretending it isn't

One of the common mistakes you, as an AI founder, can make is trying to outrun the trust deficit with the product. You think the demo wows them and they'll overlook the 10-person team, the zero case studies, and the website that went live three months ago. They won’t. 

Regulated enterprise sectors like legal, finance, or government don't just buy software. First of all, they buy risk mitigation.

Every vendor they onboard is a potential liability: a data breach, a failed integration, a compliance gap, or downtime due to low reliability. You have to keep in mind that when you have no brand, you're a high-risk vendor by default. The thing is that at this point, your job isn't to prove your product works but to lower their risk perception to the point that they're willing to find out.

Tactic 1: Sell the pilot instead ot the product

Our first paying client was Silpo, one of Ukraine's largest retailers. We didn't walk in cold but went through Fozzy Venture Studio, an internal accelerator run by Silpo's parent company. The accelerator ran for about five months, then came another three to four months of testing and configuration before the agent went live.

What made that process work? Right sequencing. We launched in co-pilot mode first, which means every AI-generated response had to be approved by a contact center manager before it reached the customer. That one decision built the initial trust: the team could see the responses, correct them, and understand the logic. 

And almost immediately after that, we switched to fully automatic mode for a defined set of topics. That list of topics has been expanding ever since, and with it, the share of customer inquiries resolved by AI has grown.

The reason Silpo cared wasn't automation for its own sake. They are extremely focused on customer experience. Their main concern was whether the AI would handle customers the way a good human operator would, or it would generate the kind of friction people associate with chatbots. That framing shaped everything about how we built and configured the agents.

Tactic 2: Your reference isn't a customer. It's a use case

When you're early, you don't have a Gartner Magic Quadrant ranking or 200 enterprise logos on your website. What you do have is a specific, credible use case that an enterprise buyer can immediately map onto their own pain.

For us, the Ministry of Justice pilot was strategically important not just as a contract, but as a proof point. We got into that project through GovTech Lab, a program designed to connect government institutions with startups. That deployment taught us a lot and it comes is sales conversations for a reason. A system handling citizen legal consultations has real stakes: wrong answers have real consequences. That meant government-certified infrastructure, local models, and proper response verification built in from the start.

When you can describe that to an enterprise buyer, the indirect message is clear: “If our system could meet those standards, it can meet yours.” Programs like GovTech Lab are an amazing entry point for startups into complex projects. They let you test your capabilities against the real requirements of high-stakes institutions. 

The lesson I'd like to give you here: pick your first deployments not for the revenue they bring, but for the story they let you tell to the next buyer. One strong reference use case in a regulated, high-stakes sector is worth ten pilots in low-stakes environments.

Four things I would do again:
• enter through accelerators, venture studios, or public-sector innovation programs rather than traditional procurement channels
• secure the right to publish the case study before signing the contract
• establish and document baseline metrics before launch so results can be measured credibly
• and build a reusable security and compliance package covering data handling, model deployment, verification, and risk management.

Together, these steps turn an early deployment into more than a customer relationship - they create a reference case that makes every future enterprise sale easier.

Tactic 3: Use the language of the person who signs the contract

AI founders love talking to engineers and product teams, and I get it – those conversations are exciting. But engineers don't sign enterprise contracts. CFOs, COOs, and General Counsels do, and they don't care about your model architecture.

Right now, about 25% of Silpo's incoming requests are fully automated. Their leadership sees that number not in percentage points but in headcount and capacity. Their contact center operators spend their time on complex requests instead of answering "How do I buy a croissant with a discount?". 

When senior decision-makers understand the behind-the-scene of the AI and humans interaction, the question of cost disappears. It becomes a question of efficiency and company development. The lesson I’d like to give you here: Help them understand it.

That shift in framing from "what does the product do?" to "where did your team save time?" is the entire game. Lots of AI founders pitch enterprises like they're presenting at a developer conference. The vocabulary and framing are wrong, and the buyer checks out before the second slide.

Tactic 4: Name the institutions, not just the logos

When you have no brand, you borrow credibility from institutions that already have it.

Being selected for Ukraine's GovTech Lab wasn't just a pilot opportunity, it was a signal we could use in our next sales conversations. An independent institution evaluated our technology against real government requirements and decided it was worth running. 

For AI founders operating in Europe right now, there are real opportunities to accumulate these signals: accelerator programs, public sector pilots, regulatory sandbox participation, and research partnerships with universities. 

The lesson I’d like to give you here: don't hand off the agent and go dark. Stay involved: requirements, data mapping, integration, everything after launch. I believe that is actually what makes a difference  for enterprise buyers. They see you’re solving the problem, not just selling software.

Most AI startups build their products based on their own vision and hypotheses, not on client requirements. They build blind. In our case, through accelerators and startup programs, we chose not to focus on short-term revenue, but on deeply understanding the problem and the client's actual requirements. That strategy helped us more than anything else. And it's the advice I give to every early-stage AI founder I talk to.

For investors evaluating early-stage AI startups targeting enterprise, this is one of the clearest signals to look for: not just "do they have customers?" but "what institutional validation have they accumulated, and how deliberately?"

Share