In this conversation, Yoni Tserruya, the co-founder and CEO of Lusha, shares why he believes AI is not overhyped, but misunderstood. While many founders rush to layer generative tools onto their go-to-market motion, he argues that true transformation starts with something far less flashy: clean, unified, signal-rich data. From redefining the role of the modern seller to envisioning a future of agent-led growth, Tserruya makes the case that durable advantage in 2026 and beyond will belong to companies that build strong foundations before chasing AI features.
Every founder is told AI will transform their go-to-market motion. Yet many teams feel stuck. Is AI overpromised?
AI is not the problem; the foundation is. What I see is companies layering AI on top of broken systems and expecting transformation. But you cannot automate chaos. AI doesn’t fix fragmented CRMs, decaying data, and disconnected signals. Instead, it accelerates these issues. Generative AI can write better emails, it can summarize calls, and it can build sequences faster. But if you are targeting the wrong accounts with outdated contact data at the wrong moment, you are just scaling irrelevance.
AI amplifies whatever it sits on. If the base is weak, the output is noise. If the base is strong, the output is leverage. GTM teams need to build strong foundations in order for AI to flourish. The question is what you are using it to optimize for: volume or relevance.
What exactly is broken in those foundations?
It’s relevancy. When I look at most GTM teams, there are two things missing. The first is clean, current, contextual data. Most companies are still working off static lists, stale CRM records, and contact data that was accurate eighteen months ago. That is the base layer, and without it, nothing else works. The second, and this is where I think the real opportunity sits, is predictive infrastructure. Knowing who is in your database is one thing. Knowing which of those ten thousand accounts are actually in-market right now, based on real behavioral signals, that is a different capability entirely. That is where the gap lives. Most teams have invested heavily in execution tools and almost nothing in the data layer that tells those tools where to aim.
What does a strong data foundation actually mean? For early-stage founders building their GTM engine, for example.
For an early-stage founder, I would keep this simple. A strong data foundation starts with three things: knowing exactly who your ICP is with real specificity, having reliable contact data for those people, and building a signal framework before you scale outreach. That last part is the one founders skip. A signal is anything that increases the probability that an account is in-market right now. At Lusha, the signals that work for us include things like a new VP of Sales being hired, a company opening a new office, or a company integrating a technology that complements what we do. Any of those signals means something is changing inside that organization, and change is when buying decisions get made. If you start building that signal logic early, even manually, you will outperform teams spending ten times as much on outbound volume.
When data is fragmented, AI cannot see clearly. When data is unified, AI becomes strategic. In this case, accuracy, freshness, and compliance matter a lot. If 25% of your data decays every year and you do not fix it, your models degrade, your forecasts weaken, and your reps lose trust in the system. The foundation is everything.
What is the biggest mistake you see founders making with AI right now?
They chase features instead of fundamentals. It is tempting to add a copilot, launch an AI assistant, and ship a flashy demo. But if the underlying data is generic or inconsistent, you are building on sand. The hard questions every founder should ask include: What truth do we own? What signals do we collect that others cannot? How do we verify them? How do we fuse them with our customer’s internal data to create something predictive?
If your answer is vague, your AI advantage will be temporary. Durable advantage comes from owning infrastructure and not from wrapping generative AI around commoditized inputs.
How does this change the role of the modern seller?
AI will replace average execution, but it will not replace human judgment. All repetitive tasks should disappear. These tasks include manual research, guessing which accounts to call, updating fields across tools, and chasing incomplete information. Machines should handle that. What remains for humans are the high-leverage skills: empathy, strategy, negotiation, and trust. But the bar rises. When AI provides context and prioritization, you cannot show up unprepared. You cannot rely on volume to compensate for weak targeting.
The future is the human-AI team. The machine predicts who to talk to and why, while the human builds the relationship and closes. The human feedback then trains the system, improving the next prediction. That loop is where performance compounds.
You talk about moving from product-led growth to agent-led growth. How does that connect to data foundations?
It connects directly. We built Lusha on the conviction that buyers want to self-serve. That PLG insight was right, and it only gets more true over time. But I think PLG is evolving into something new, which I call agent-led growth. We are heading toward a world where autonomous agents, not human users, evaluate and purchase software on behalf of organizations. An agent tasked with finding the best sales intelligence solution will assess options, run comparisons, and maybe even initiate a contract. And that agent will have zero tolerance for friction. No “contact sales” form. No opaque pricing. No complicated onboarding. The platforms that win in that world will be the ones built for machine consumption, open APIs, clean data structures, and frictionless access. The data foundation piece connects directly because agents need structured, reliable, signal-rich data to make decisions. If your data layer is a mess, you are invisible to them.
Your data must be accessible, reliable, structured, and trustworthy
Many teams feel overwhelmed with the tools and signals. Where should they start?
They should start by simplifying. Map your data flow: where does raw data enter, how is it verified, enriched, scored, and routed, and where is feedback captured? If you cannot clearly explain that flow, no AI will fix it.
Then prioritize accuracy and connectivity. Clean your CRM, deduplicate, enrich consistently, and connect signals to workflows so intent does not sit idle in a dashboard. Only then should you layer automation and generative capabilities on top. AI should serve as a catalyst at the end of the process, not as a temporary fix at the start.
If you had to give one piece of advice to founders scaling their GTM in 2026 and beyond, what would it be?
Invest in your data infrastructure before you invest in your AI stack. I know that sounds counterintuitive when everyone is rushing to deploy agents and automation. But the quality of your AI output is entirely dependent on the quality of your data input. The analogy I keep coming back to is this: imagine a sales rep with a perfectly personalized outreach sequence and an Ivy League work ethic, but she is working off a list of companies that are the wrong size, in the wrong stage, with contacts who left eighteen months ago. All that talent goes nowhere. AI is the same. The most sophisticated model in the world will underperform if the data it is operating on is stale, incomplete, or irrelevant. Get the foundation right first. Then the AI compounds on top of something real.
Yoni Tserruya is the co-founder and CEO of Lusha, a B2B sales intelligence platform used by over 800,000 sales professionals and 223,000 organizations worldwide.
