One of the most common questions we receive from founders pitching Lumino Capital is a variation of the same question: "How do you evaluate the technical feasibility of what we are building?" It is a fair question, and it gets at a fundamental challenge in deep tech investing. The tools that work well for evaluating software startups - looking at early user traction, measuring the speed at which a team ships features, assessing market size and competitive dynamics - are poorly suited to companies whose primary challenge is not product-market fit but the prior question of whether the underlying science will work at commercially relevant scale.
This article attempts to lay out the framework we use at Lumino Capital to evaluate scientific risk in early-stage deep tech companies. It is not a complete guide to deep tech due diligence - there are entire books on that subject - but it reflects the core mental models we have developed over six years of investing in hard technology, and we share it in the hope that it is useful both to founders preparing to pitch and to investors thinking about how to approach this category.
Distinguishing Scientific Risk from Engineering Risk
The first and most important distinction in evaluating a deep tech company is between scientific risk and engineering risk. Scientific risk is the risk that the underlying physical or biological phenomenon the company is relying on does not behave as the founders believe it will. Engineering risk is the risk that a phenomenon that has been demonstrated in the laboratory cannot be reliably reproduced at commercially relevant scale, cost, and quality.
This distinction matters enormously for how you evaluate and price a company at the seed stage. Companies carrying significant scientific risk require extraordinary technical confidence from investors - you are essentially betting that the science is right before it has been proven. Companies carrying primarily engineering risk are making a different kind of bet: the science is established, but executing the engineering challenge is genuinely hard and the outcome is uncertain.
In our experience, the most dangerous deep tech investments are those where founders (and investors) have miscategorised scientific risk as engineering risk. This happens when a team has demonstrated a phenomenon in a highly controlled laboratory setting and interprets this as proof that the underlying science is validated, when in fact the demonstration conditions were so far from commercial deployment that the extrapolation remains scientifically uncertain. Battery chemists know this phenomenon well: the history of battery science is littered with laboratory results that looked extraordinary in a small-format coin cell but could not be reproduced in a practical format at useful temperatures and cycle life.
The Technology Readiness Level Trap
NASA's Technology Readiness Level scale, a nine-level framework for characterising the maturity of a technology from basic principles (TRL 1) to proven in operational environment (TRL 9), is widely used in the deep tech community as a shorthand for technological maturity. We have found it useful as a starting point, but deeply insufficient as an investment framework, for two reasons.
First, the TRL scale conflates scientific validation with engineering development in ways that obscure the most important risk categories. A technology at TRL 4 (validated in laboratory) may be carrying significant residual scientific risk if the laboratory conditions used for validation were not representative of commercial use conditions. A technology at TRL 6 (demonstrated in relevant environment) may be well de-risked scientifically but face a very difficult engineering scaling challenge. The TRL rating gives you information about where you are in the development process, but not about the nature of the remaining uncertainty.
Second, the TRL scale does not capture the cost of the development journey. Two companies might both be at TRL 4, but one might require a $5 million seed round and 18 months of work to reach TRL 6, while the other might require $50 million and five years. The capital efficiency of the development pathway is at least as important as the current TRL in determining whether a seed investment makes sense.
Our Evaluation Framework: The Five Questions
After considerable iteration, we have settled on a set of five questions that we believe are the most important for evaluating scientific and technical risk in a seed-stage deep tech company. These questions are not independent - the answers interact with each other in important ways - but they provide a structured way to think through the key risk categories.
1. Is the core phenomenon well-established?
The most important scientific question in any deep tech investment is whether the underlying physical or biological phenomenon has been reliably demonstrated under conditions that are at least directionally similar to the commercial use case. We look for published evidence in peer-reviewed literature (not conference proceedings or press releases), independent replication of key results, and a clear mechanistic understanding of why the phenomenon works - not just evidence that it does.
We are particularly cautious about technologies that rely on effects observed at nano- or micro-scale where the behaviour at macro-scale is assumed but not demonstrated. Nanomaterials with extraordinary laboratory properties have a long history of failing to maintain those properties when synthesised at scale. Biological assays conducted in cell culture have repeatedly failed to predict behaviour in animal models or humans. Scale dependence is one of the most common sources of scientific risk in deep tech, and it is one that founders often underestimate.
2. What are the known failure modes?
The most technically credible founders are those who have deeply internalized the ways in which their technology could fail, and have a clear plan for detecting and addressing each failure mode. We are instinctively sceptical of founders who present only upside scenarios. Our question is always: what would a world-class scientific critic say about your approach, and how do you respond?
The most common failure modes we have encountered in our portfolio categories include: chemical degradation mechanisms in energy storage materials; decoherence sources in quantum systems; off-target effects in biological applications; surface passivation degradation in semiconductor devices; and supply chain constraints for critical materials. In each case, the founding team should be able to discuss these issues with the same depth as a domain expert, because they are the central technical challenges that will determine whether the company succeeds.
3. What is the commercialisation pathway?
Deep tech companies need to develop a commercial product, not just a scientific demonstration. We evaluate the gap between the current state of technology and the performance and cost requirements of the first commercial application, and we assess whether this gap can be bridged with the resources being raised at seed.
A mistake we see frequently is a mismatch between the technology's current state and the chosen first commercial application. The first commercial application should be the one where the technology is closest to meeting performance requirements, not necessarily the one with the largest market. Companies that target the highest-performance application first - because it is the most exciting or because the market is largest - often find themselves in a development programme that is much more capital-intensive and time-consuming than they anticipated, while leaving easier wins on the table.
4. What is the intellectual property position?
Defensible intellectual property is not sufficient for a great deep tech investment, but the absence of it is usually a serious concern. We look for patent portfolios that cover genuinely novel and non-obvious innovations, and we pay particular attention to whether the company's core IP protects the commercial product they are building or merely the laboratory demonstration. A patent on a laboratory-scale synthesis process may provide little commercial value if the commercial product is manufactured via a different process.
We are also interested in whether the company is building know-how that cannot be captured in a patent - proprietary process knowledge, accumulated experimental data, materials characterisation expertise - since this is often more defensible than patents in markets where large incumbents are willing to design around existing IP.
5. Who would know if this is wrong?
Our final question is perhaps the most important, and it is one that we apply to every technical claim we encounter. We ask: who in the world has the expertise to evaluate this claim, and what do they think? In every major domain in which we invest, there is a small community of experts who would immediately recognise a serious technical error in a company's claims. Accessing the opinion of these experts - through formal technical advisory relationships, informal conversations, or detailed literature review - is a critical component of our due diligence process.
We have found that the most dangerous companies are not those making obviously implausible claims, but those making claims that are superficially plausible but contain subtle technical errors that only domain experts would recognise. These companies often attract initial investment from generalist investors, then struggle to raise follow-on funding once specialist investors conduct more rigorous technical diligence. We have made it a priority to ensure that our technical due diligence is at the level that specialist follow-on investors will apply - not because we doubt the founders, but because we want to be confident that the company can successfully raise future rounds from the most rigorous capital in the market.
When to Invest Despite Residual Scientific Risk
The framework described above is not intended to suggest that we only invest in technologies where all scientific questions have been answered. If we required zero residual scientific risk, we would never invest in a seed-stage deep tech company. The question is not whether scientific risk exists, but whether it is the right kind of risk at the right price.
We are prepared to invest with significant residual scientific risk when: the founding team has the expertise and experimental infrastructure to efficiently resolve the key uncertainties; the capital requirements to reach the next meaningful de-risking milestone are well-matched to a seed round; the commercial opportunity, if the science proves out, is large enough to justify the risk; and the downside scenario, if the science fails, is a company that fails gracefully rather than one that burns through capital without generating useful information.
The last criterion is one that we think is underappreciated. In the best deep tech investments, even a negative scientific result has value - it generates knowledge that advances the field and clarifies the development path for subsequent companies. We look for companies where the experimental programme is designed to generate clear, interpretable results at each stage, so that the investment thesis can be updated as evidence accumulates. This requires founders who are scientifically rigorous and commercially honest - who are willing to update their plans based on experimental evidence rather than defending the original thesis against disconfirming data.
That combination - scientific rigour, commercial ambition, and the intellectual honesty to follow the evidence wherever it leads - is what we look for above all else in the founders we back. It is rare, and when we find it, we back it with conviction.