Authors: Elena Vrabie and Adriana Spulber

Selling AI products is becoming harder, not easier. In the age of agentic AI, where every startup claims to be transforming workflows and many enterprises are running a pilot, the gap between curiosity and real adoption has never been wider — or more expensive to cross.

To shed some light on what it actually takes to sell agentic AI, we reached out to Elitsa Zaimova-Miller, who has spent over 14 years on the front lines of that gap. As VP of Sales at CloudTalk, one of EMEA's fastest-growing AI calling software companies serving over 5,000 SMEs across 160 countries, she leads the charge on turning voice agent technology into measurable revenue. 

Her career spans 14 years and three continents. She began at CEB (now Gartner) in Washington, DC, advising sales executives on team performance and revenue strategy using the Challenger Sales methodology. At Preply, she launched the B2B division from scratch in 2020, designing its go-to-market strategy and scaling it to over 1,000 enterprise partners and $30M in ARR. Beyond her operating roles, she advises startup founders on ICP definition, sales process design, and scaling go-to-market teams — and volunteers her time coaching girls who are considering careers in sales.

In this interview, Elitsa breaks down what it takes to sell AI voice agents to SMEs, why the knowledge base — not the technology — is usually what breaks a deployment, and what early-stage founders get wrong when they try to scale before they've learned to listen. She also offers a candid take on why 'agentic' may already be losing its meaning, and what CEE-built companies bring to global sales that Silicon Valley often doesn't.

UV: CloudTalk claims AI voice agents can transform sales, support, and engagement without replacing human teams, with strong ROI potential. What typically triggers the move from curiosity to real adoption — and how does that differ across segments?

Elitsa Zaimova Miller: We launched voice agents early last year, starting with existing customers to test real-world impact. Two things shifted in parallel: the technology matured dramatically, and so did customer appetite. By the end of 2024, curiosity had become urgency — companies that had been watching were now moving.

We serve smaller organizations: SMB and lower mid-market. These organizations want to move quickly to grow, but they don't have the resources large enterprises do. That actually works in our favor — they're more open to experimenting with AI because they have to be. They can't afford to staff their way to scale.

The sweet spot is high-volume, lower-complexity customer interactions. A good example: an e-commerce business with customers reaching out 24/7 from all over the world. You don't have the budget to hire people across every language and time zone. A voice agent can handle those conversations in any language, with any accent, and deliver a genuinely good customer experience.

What's shifted culturally is that it's no longer just progressive tech companies experimenting. We have a customer who runs a family-owned hotel chain just outside Venice. They told us: 'Our competitors are doing this, our customers expect it, and we need to stay competitive.' That kind of adoption — in a traditional, relationship-driven industry — tells you something real is happening.

UV: Within the SME segment, where are customers leaning more — sales or support use cases? And where does AI typically hand off to a human?

EM: The most popular use case right now is missed calls. If you're a small global business, you simply can't be available at all hours in all languages. A voice agent fills that gap — handling inbound contacts, answering common questions, troubleshooting simple issues, and escalating when something genuinely needs a human.

We're about to release an AI receptionist that takes this further. Imagine calling an insurance company today: 'Press 1 for billing, press 2 for support' — and then reaching a voicemail. Now imagine a voice agent that greets you by name, looks up your account history in real time, resolves your issue or routes you intelligently, sends a follow-up SMS, and books an appointment. That's the experience we're building toward.

On the sales side, the role of voice agents is more supporting than closing. Sales is complex, relational, and high-stakes — it won't be automated anytime soon. But roughly 40% of what sales teams do is administrative: logging calls, writing notes, confirming appointments, sending calendar invites. Voice agents can handle all of that, freeing the human to focus on what actually drives revenue: the conversation itself.

The handoff to a human happens when complexity rises — emotionally charged situations, negotiations, anything that requires genuine judgment or relationship capital. The voice agent's job is to handle everything else, so your team's time is spent where it actually matters.

UV: Who are typically the economic buyer, technical buyer, and daily user of voice agents in SME organizations — and where do deals most often stall?

EM: In smaller organizations, we usually work directly with the founder. Deploying a voice agent is a core operational decision, and founders take it on themselves. We've designed the product to be set up without deep technical knowledge, and we provide hands-on support for those who need it. In organizations with 500-plus employees, we typically see the CTO or RevOps involved, depending on the industry.

As for where deals stall, there are a few recurring patterns.

The first is education lag. Many buyers are still figuring out how the technology works and what it will take to implement it internally. That creates longer cycles, and it's on us to do a lot of upfront work.

The second is organizational readiness. Deploying a voice agent is a lot like onboarding a new employee: you need to train it properly, and your internal resources need to be organized in a way the agent can actually use. If a voice agent underperforms, it's rarely a technology failure — it's almost always because the knowledge base wasn't structured well enough to support it. This is probably the most underappreciated implementation risk.

The third is expectation management. Voice agents get better through iteration — scripts, prompts, language. Customers who expect perfection from day one are setting themselves up for frustration. We try to reframe this early: this is a new team member, not a plug-and-play tool.

UV: What does your team do in the first 90 days to drive adoption and value realization?

EM: It depends on the customer. For smaller businesses, we offer a three-month package that lets them run a real test — not a pilot in the traditional sense, but an actual deployment where they should start seeing impact within weeks. For us, three months is time to prove value, not to decide whether to buy.

For more complex use cases — outbound agents that need to collect information, trigger follow-up actions, or integrate with other systems — setup takes three to four weeks. After that, the following two months are typically when customers start seeing meaningful results.

Our structure reflects this: the sales team focuses on the close, and once deployment begins, a post-sales team works closely with our technical team to support the customer. We want customers to consume 90 to 95% of their purchased minutes and see clear value within three weeks of going live. Once setup is done well, adoption tends to take care of itself.

UV: Are you seeing strong adoption in specific industries beyond e-commerce and hospitality?

EM: We're industry-agnostic by design, but patterns emerge.

Insurance is a strong fit, especially in the US — the volume of routine paperwork and status-check calls is enormous, and a voice agent actually delivers a better experience than the IVR menus most companies still use. B2C sales organizations are another — immigration services, for instance, where customers need to track documents, visa steps, and appointment schedules. Lots of structured, repeatable interactions that voice agents handle extremely well.

For outbound, B2B companies doing event marketing — inviting people to webinars, confirming registrations, following up — are a strong fit.

The one that surprised me most was the universities. I expected them to be slow adopters, and I was wrong. They use us cyclically: during enrollment season, they're handling enormous volumes of student outreach — confirming documents, scheduling activities, answering questions. Then usage drops sharply over the summer. It's essentially seasonal staffing, but without the hiring and training overhead.

UV: Have cyclical or seasonal usage patterns changed how you think about packaging?

EM: Absolutely. Early on, we noticed certain customers with hundreds of thousands of minutes in one month and near-zero the next. When we dug into it, the explanation was always the same: they were using voice agents to handle seasonal peaks they couldn't staff for.

That insight changed how we think about value delivery. For seasonal businesses — whether it's a toy retailer in November or a university in March — the ability to scale instantly and scale back just as fast is the core value proposition. We now design for that explicitly rather than treating it as anomalous behavior.

UV: How has your pricing and packaging model evolved?

EM: It's been constant experimentation. If you look at what competitors charge for voice agents, the strategies are all over the map: per-minute, per-resolution, outcome-based packages. We started with pay-as-you-go, realized it created too much friction, and moved to bundles.

The biggest lesson was: simplicity is what matters the most. Our customers are already learning how the technology works; they shouldn't also have to decode a pricing structure. We eliminated hidden fees, concurrency charges, and LLM model surcharges. The package tells you exactly what you're getting and what you'll pay.

We review pricing quarterly. The product is still evolving, and the pricing has to evolve with it — but the principle stays constant: fair for the customer, sustainable for us, no surprises.

UV: What go-to-market advice would you give early-stage founders selling AI products — particularly around ICP selection, first customers, and when to build a sales function?

EM: The number one thing founders get wrong is trying to perfect the product before talking to the market. Stop building in isolation. Build something good enough, go sell it, and then alter it based on what you learn. Get your first ten customers yourself. If you're spending 90% of your time building, flip that ratio — spend 50% building and 50% in front of customers.

No one will be more passionate about your product than you are. I hear a lot of founders say, 'I just want to build — how do I outsource sales?' That's a red flag. If you're not willing to sell it yourself, you're not really a founder; you're an engineer who wants to stay inside. Selling doesn't mean being a natural extrovert. It means being genuinely curious about whether your product solves a real problem, and having enough conviction to go find out.

For getting first customers: start with your existing network. Think about former colleagues who might be willing to test something. Use LinkedIn. Build a list of ten companies you want to close and chase every warm connection into those organizations. Partnerships are valuable long-term, but they take time to mature — they're not a substitute for direct selling early on.

Treat early GTM like a pattern recognition problem. Every conversation gives you a data point. Listen, alter, listen, alter. After ten deals, you'll have enough signal to start thinking about what a repeatable process actually looks like.

UV: When should a founder stop customizing everything for each customer and start standardizing?

EM: There's a concept I heard from one of the co-founders of Booking.com, who advised us at Preply: 'milking the cows.' Milking the cows is the most boring thing you can do. But if you don't do it, the cows die.

Founders tend to try something, decide it isn't working, and swing completely to the other side. That's the wrong instinct. Move the pendulum in small, deliberate steps, grounded in data. Test one variable at a time — if your outbound email isn't converting, is it the subject line, the message, or the persona? You can't know if you change all three at once.

When we built the outbound motion at Preply, that's exactly what we did. Start with one version. Check the open rate. If it's low, change the subject line — nothing else. Check the response rate. If it's low, change the body — nothing else. That compounding effect of small, consistent, data-driven changes is what eventually creates scale. It's not glamorous, and it's not fast. But it works.

UV: When should a founder transition from founder-led sales to a structured sales team — and what profiles work best for early-stage AI companies?

EM: Two questions: when and who.

On when: hire when you see traction. You've closed ten to fifteen companies. You have a directional sense of where you should be going. Now you need execution bandwidth. The mistake I see constantly is founders hiring to solve a problem they haven't diagnosed yet — bringing in a salesperson hoping they'll figure out what's not working. That's not how it works. You need to find the path first, then hire someone to run it hard.

On who: Don't hire a VP of Sales with fifteen years of experience and expect them to do prospecting. It won't happen. What worked well for us was hiring two full-cycle account executives simultaneously. The reason for two: if you hire one and it doesn't work, you're testing two variables at once — the person and the market fit. Two costs more but gives you a cleaner signal faster.

In terms of profile: hire someone genuinely excited about AI, who uses it daily, and who wants to figure things out on their own. They don't need a technical background. They need drive, resilience, and intellectual curiosity. You can teach almost anything. You cannot teach someone to care. Don't hire a resume — experience tells you how someone responded to a problem in the past, and not how they would solve your problem. Hire conviction.

UV: Should founders use words like 'agentic' in their sales pitch? Does riding the AI hype help or hurt?

EM: Your pitch should follow your customers, not the other way around. Honestly, I'm skeptical of pitches in general — they tend to be too commercial, too one-directional. I tell my team: don't build a pitch. Go and listen. How are your customers describing their problems? What language are they using? If they're saying 'agentic' every day, mirror it. If they've never heard the word, don't introduce it.

The technology is genuinely remarkable — voice agents that switch languages mid-conversation, that can tell a joke, that remember your history, and take action during the call. The gap between where this was two years ago and where it is now is extraordinary. But that's a reason to be specific and concrete about what it actually does, not a reason to lean harder on buzzwords.

I see too many companies putting 'AI' or 'agentic' in their name or pitch without it meaning anything substantive. Customers who don't fully understand the technology — which is still most customers — are less likely to buy something they can't picture. Try to focus on educating your customers, teaching them the power of AI rather than intimidating them with buzzwords they may not fully understand. 

UV: Is 'agentic' just a buzzword?

EM: Personally, yes — it's become one. People have used it so broadly that it's lost precision. That doesn't mean the underlying technology isn't real or significant; it absolutely is. But if I'm talking to a customer, I'd rather explain what the product does in plain language than reach for a term that might make them feel like they're missing something.

Out of more than eight billion people, roughly 2.5 million are using generative AI tools. Think about who's buying those tools and whether they actually understand them. The companies that will win long-term are the ones that help customers understand what they're getting, not the ones that make them feel embarrassed for asking questions.

There's also a retention dimension here that founders often underweight. It's not just about selling — it's about whether your customers are genuinely happy after they buy. If you're closing deals on hype and underdelivering on substance, your churn will tell that story. Your existing customer base is the truest signal of how good your business actually is. How you describe the product for investors is one thing; how you implement it and deliver value for customers is something else entirely.

UV: You've sold in the US and now lead sales at a CEE-founded company selling globally. What advantages or challenges does a CEE background create when selling AI to Western European and North American customers?

EM: The first thing I'd challenge is the framing of 'CEE company selling globally.' CloudTalk is a global company that happens to have been founded by Central Europeans. There's a meaningful difference. If you want to serve US customers, you need a team that lives and breathes the US market — not people trying to cover it from across the Atlantic. We have a team in Toronto serving the Americas, many of them originally from Latin America. Our EMEA team is based in Barcelona but genuinely international — South Africa, the Czech Republic, Mexico, Brazil, and Hungary. The founders happen to be Slovak. Headquarters are less important than proximity to your customers.

That said, there is something distinctly Central and Eastern European about how our founders and leadership operate. We are direct. We are execution-focused. In CEE, the default is: let's identify the problem, figure out the fastest path forward, and move forward. I've worked in US companies for most of my career, and I've learned to code-switch — but I genuinely value that directness. It's a competitive advantage when speed and clarity matter.

The real challenge for CEE companies going to the US isn't cultural — it's cost. Building a credible US presence is expensive, and you need the right people on the ground. But if that's where your customers are, you don't have a choice. Follow the customer, not the map.