The AI Opportunity in Deeptech and Life Sciences GTM: Deep, Not Wide

Introduction (Klemen)

SaaS go-to-market runs on cheap learning. Ship something incomplete, send a thousand emails, watch what converts, adjust, repeat. The market itself is the laboratory, and the experiments cost almost nothing.

Now take that playbook into deeptech or life sciences. You can't put an unvalidated hardware product in front of a lab, and you can't send a half-formed claim to a clinical researcher. You can't run a hundred outreach experiments across a market of a thousand buyers to find out what lands, because the accounts you burn learning are not replaced. The laboratory is closed. And yet most of the GTM tooling, courses, and benchmarks these industries buy were built on the assumption that it's open.

I've had a running conversation about this with Yacine Cherraoui for months. Yacine works at the intersection of science and business: a chemist by training, an MSc in Sustainability, Entrepreneurship and Innovation from ESCP Business School, today building the go-to-market and revenue operations infrastructure at ATLANT 3D, a company commercializing proprietary advanced manufacturing technology. He serves on the board of La French Tech Munich and mentors early-stage founders across the Franco-German and European ecosystems. He has seen this problem from the lab bench and from inside commercial teams.

We kept arriving at the same conclusion from different directions: in markets like this, the learning has to happen before the outreach, not through it. And that's exactly the work AI just made affordable. So we wrote the conversation down.

---

From the lab to the market

Klemen: You started as a chemist and ended up building the commercial side of a nanotech company. What happened in between?

Yacine: I trained as a chemist before pursuing an MSc in Sustainability, Entrepreneurship and Innovation at ESCP Business School. What drew me to this space is the gap between the two worlds: understanding the science and understanding what it takes to commercialize it. I worked across corporate VC funds and early-stage ventures before moving to the operator side, building GTM and RevOps infrastructure in deeptech.

---

SaaS vs deeptech: the real difference is the cost of iteration

Klemen: We keep saying deeptech GTM plays by different rules than SaaS. Make that concrete. What's actually different?

Yacine: Most of it isn't different, and I'd start there. Enterprise sales cycles, procurement, the need for a repeatable motion: those are the same problems everyone has. The one real difference is the cost of iteration.

SaaS playbooks assume you can learn cheaply: ship something incomplete, watch what happens, adjust. That single assumption underwrites the whole volume logic, the sequences, the statistical conversion rates. In deeptech it works differently. You can't put an unvalidated hardware product in front of a lab, and you can't run a hundred experiments across a market of a thousand buyers to find out what lands.

So the learning has to happen before the outreach rather than through it. That's the actual difference. Most of what people list, long cycles, evidence-gated adoption, follows from it, and compounds with a market that was already small. And it's why a playbook built for cheap iteration tends to use up accounts that aren't easily replaced.

Klemen's take: This is the naive-versus-expert gap in one sentence. The naive approach learns by experimenting on the market, because in SaaS that experiment is nearly free. The expert approach in deeptech moves the experiment off the market and into research, because every test you run on a live account spends something you can't buy back. We see the same pattern putting AI into production: teams that test on ten perfect records learn in production, expensively. Teams that test on the thousand messy ones learn before anything real is at stake. Deeptech GTM has always been forced into the second, disciplined mode. Volume-era tooling just kept pretending otherwise. And the volume era is ending anyway: I now mark AI-generated cold emails in my inbox as spam, and so do your buyers.

---

The opportunity: deep, not wide

Klemen: Everyone pitches AI as a way to do more. More emails, more sequences, more pipeline. What's the actual opportunity in a market like yours?

Yacine: If your market is around a thousand accounts and a handful of deals make the year, volume doesn't work in your favor the way it does in SaaS. Sending more mostly means moving through the market faster. So the useful question isn't "how do we send more." It's "how do we know more before we send anything."

AI helps at three levels: the contact, the account, and the wider policy environment. But differently at each.

Contact level: knowing the person before you contact them. This is where the leverage is largest, because it's where trust is built. What matters isn't which company someone works for. It's which researcher holds which grant, what it's funding, what they published last, what their method implies about the equipment they need.

As of 2026, standard tools are only partly suited to this. Most commercial intent data is account-level, and vendors themselves note it works less well when a category is too niche to generate detectable research activity. That's a fair description of our market.

The information usually exists publicly, though: scattered across grant databases, publication records, and institutional pages rather than packaged as a dataset. An agent can pull those sources, structure them, and produce a short brief per contact. What used to take an afternoon becomes routine. And it changes the conversation: you're opening with evidence you understand their work rather than with a pitch.

Account level: filtering and preparing. Narrowing thousands of accounts down to the genuinely relevant ones is a research task, and AI is well suited to it. Then preparation: the deep dive every rep wished they had time for, now feasible before every call rather than for the three biggest deals of the quarter. What changed isn't outreach. It's the cost of research per account.

Macro level: watching policy and funding. This influences the entire industry and shapes who becomes a buyer in the first place. For example, the European Chips Act came into force in 2023, aiming to double Europe's share of global chip production to 20% by 2030. It reads like macroeconomics. But it funds pilot lines and competence centres, organisations with equipment budgets that didn't exist three years ago, and research infrastructure projects that resolve to named organisations, their location, and their exact share of the funding, published openly in CORDIS. INFRACHIP is one example: 14.5 million euros, running to 2027, bringing together universities, research organisations, and industry clusters across Europe, with a stated objective that includes capital investment in state-of-the-art equipment and moving results from lab to fab. One policy decision travels from a production target down to named organisations with published budgets and a stated technical goal.

Life sciences works the same way, published in the same place. ISIDORe II launched in June 2026 and runs to 2029: a 10.5 million euro Horizon Europe programme that assembled around sixty research organisations, coordinated by ERINHA in Brussels, into a single instrument for infectious-disease research, from structural biology to clinical trials, giving scientists funded access to state-of-the-art facilities. It's live right now. If your buyer is a lab, a funded consortium like this is a list of buyers who just acquired budget and a mandate, names, locations, and shares included.

And these signals point forward, not back. The Commission proposed a Chips Act 2.0 in June 2026, which means the next wave of funded organisations is already forming while this piece goes out. Intent signals tell you who is already searching. Funding and policy point to who is about to exist.

Which is why the premise is worth revisiting. A small market has generally been treated as a disadvantage: too few accounts for a numbers game. But depth was always the right approach here; it simply wasn't affordable. That's the part that changed, and it's where a lot of the opportunity now sits.

Klemen's take: We've built exactly the kind of agent Yacine describes at the contact level: it gathers whether a researcher holds a grant, the grant's title, the abstract. That data exists in no prospecting tool at any price, and referencing it changes reply rates because the email stops being a pitch and starts being a peer noticing relevant work. One caveat from production experience: none of this works as "give the AI a login and let it figure things out." Every workflow that survives contact with real data has deterministic plumbing underneath: code pulls and structures the context, AI reasons on top of it. The depth game rewards engineering discipline, not tool subscriptions.

There's a line HubSpot used in its Fall 2025 Spotlight: most businesses make 100% of their decisions with 20% of the data. It's a marketing line, not a research statistic, but it describes deeptech GTM better than it describes SaaS. The other 80% of the data exists. It's in grants, papers, policy feeds, and funding databases. It was just never affordable to collect per account. Now it is.

---

For teams selling science

Klemen: What's the one thing you wish more deeptech and climate founders understood before taking hard science to market?

Yacine: Approach go-to-market with the same rigour as the science and the product.

Klemen: Where do you start?

Yacine: It starts with documenting the process. Written-down workflows make the GTM strategy and its operations explicit and the inconsistencies visible. And they also define where AI can actually be applied. You can't automate or delegate a process nobody has defined.

Klemen: And when you walk into a company where a system already exists, however improvised?

Yacine: This hits early-stage deeptech hardest. At that stage there's often no GTM or commercial function yet: whatever process exists was built by the team around real constraints that are often never written down. So it's worth understanding why the current setup looks the way it does before replacing it.

Klemen: Who should own "building with AI" in a small deeptech company? Everyone, or one person?

Yacine: One person, or one function depending on size. But named either way. Without a clear owner, people naturally build their own versions, and the same problem tends to get solved several times over in parallel. Name one owner, give them controlled context and reusable, validated skills, and the team compounds off that work rather than each person starting from scratch.

Klemen's take: I've watched the compounding version of this work. I mentor a RevOps leader whose team kept improvising with AI individually, everyone prompting their own way. He built one validated skill encoding how their migrations actually work, shared it with the whole team, and tasks that took 30 hours dropped to two or three, at the same quality. One owner, shared context, reusable skills. That's the pattern, and it applies to a five-person deeptech commercial team even more than it applies to an agency.

---

Closing note (Klemen)

The thread through this whole conversation is one economic fact: the cost of knowing dropped. Knowing the researcher before you write to her. Knowing which accounts are genuinely worth the effort. Knowing which funding programme is about to create your next buyer. That work was always the right way to sell science. It just didn't fit in anyone's day.

It does now, and that reframes what a small market is. A thousand accounts was never enough for a numbers game, which is why deeptech teams kept borrowing playbooks built for someone else's market. But a thousand accounts is exactly the right size to know properly. The teams that point AI at depth rather than volume aren't compensating for a small TAM. They're finally playing the game their market was always shaped for.

---

Yacine Cherraoui works at the intersection of science and business, helping deeptech and impact-driven ventures scale. He is currently building the go-to-market and revenue operations infrastructure at ATLANT 3D, serves on the board of La French Tech Munich, and mentors early-stage founders across the Franco-German and European startup ecosystems.

Klemen Hrovat is co-founder and CRO of Sellestial, an AI expert partner helping companies build agents, custom apps, and production-grade AI operations on top of HubSpot and beyond.

Sources: European Chips Act, European Commission. INFRACHIP, CORDIS project 101131822. ISIDORe II, CORDIS project 101293306.