The Assumption Nobody Thought to Test.

The Valley of Death has a well-documented address: the gap between R&D funding and commercial revenue. The last three autopsies examined the conditions that create it. This edition moves inside it — specifically, the assumption risk that accumulates quietly while the technical curve advances and the commercial curve waits. It is the least visible risk category in deep tech. At TRL 6–7, it is frequently the largest one.

The Assumption

Technical progress is the primary de-risking activity for deep tech founders. Every milestone on the TRL scale represents risk removed. By TRL 7, the hard work is done. What remains is commercial — and commercial can start in earnest once the technology is ready.

This assumption feels rigorous. It has a framework behind it. It produces measurable milestones. It gives teams a shared language for progress that investors, grant bodies, and technical peers all recognise and reward.

It also has a structural blind spot.

The Reality Check

Commercial engagement in deep tech is not a downstream activity. It is a parallel curve — and it has its own compounding logic.

From the earliest stages, before performance benchmarks are anywhere near customer expectations, conversations with potential customers are already doing essential work: mapping pain points, locating the decision architecture inside target organisations, and — critically — beginning to define what “good enough to buy” actually means in operational terms. These conversations are not premature. They are the only way to calibrate what the technical curve is supposed to be optimising toward.

The terrain looks familiar at TRL 6–7. Regular customer contact, mutual interest, a pilot in discussion. What is not yet visible is the delta between what the customer meant and what the founding team heard. That gap does not announce itself. It accumulates quietly across every conversation where the harder question — what result would make you sign? — was present but not asked.

Commercial conversations without explicit assumption-testing map the relationship, not the risk.

This is not a story about founders who ignored their market. It is a story about a gap that opens gradually, across technically capable teams, in the space between sustained engagement and secured understanding. The instrument for measuring that gap is rarely calibrated — because nothing in the TRL framework requires it.

The Forensic Analysis

GE launched Predix in 2015 as the operating system for the industrial world — a platform connecting turbines, pipelines, and manufacturing equipment into a unified data layer generating predictive intelligence at scale. The vision was coherent. The technology was real. The pilots ran. By 2018, GE Digital was reporting losses of approximately $400M. By 2019, the Predix workforce had been reduced from roughly 1,500 to under 500. GE’s stock, trading at $32 in 2016, fell below $7.

None of that was caused by a failed laboratory result.

The pilots succeeded — on the terms the people who commissioned them had defined. C-suite champions inside GE and inside client organisations were buying a vision of digital transformation. That vision was legible, fundable, and generated significant internal visibility for the people who sponsored it. What it did not generate was a clear answer to a different question entirely: what does the plant manager need this to do on Tuesday morning?

The operational user — the factory floor manager, the turbine maintenance engineer, the plant supervisor — was rarely the design partner. Their pain points were specific, workflow-dependent, and resistant to platform logic. The Predix interface that made sense in a boardroom presentation did not map onto the daily operational reality of the people whose adoption would determine whether the technology actually worked at scale. That gap was present from the beginning. It was not measured.

The commitment stage revealed the second problem. Pilots were interpreted by the startup side as procurement signals. They were interpreted by the client side as useful experiments — bounded in scope, budget, and obligation. The corporate champion who had authorised the pilot frequently lacked both the political capital and the structural mandate to force operational deployment across the organisation. The person who could say yes to a pilot and the person who could say yes to a rollout were not the same person. In most cases, they had never been in the same room during the engagement.

The scalability problem followed directly. Because no standardised operational use case had been secured — because success criteria had been defined at the vision level rather than the workflow level — every new client engagement required bespoke customisation. Each pilot was, in effect, a new R&D project. The commercial curve was not advancing. It was repeating.

The hidden beneficiary dynamic runs through all three stages. Corporate champions on both sides were rewarded for initiating innovative projects. Completion, integration, and long-term operational adoption generated no equivalent reward. The startup interpreted sustained engagement and funded pilots as evidence of commercial momentum. The structural incentives of the people sustaining that engagement pointed elsewhere.

This is not a story about naive founders or cynical clients. GE was not a startup. Predix was not a garage project. The teams involved were experienced, well-resourced, and technically capable. The gap that opened was not between competence and incompetence. It was between two different definitions of what the pilot was supposed to prove — and the assumption, on one side, that both parties shared the same definition.

They did not. Nobody checked.

The conversion gap is not unique to GE. McKinsey’s technology research consistently finds that fewer than one in ten organisations moves from active pilot to scaled deployment — a figure that has remained stable across multiple technology waves, from analytics to cloud to AI (McKinsey Technology Trends Outlook, 2024). The deep tech hardware sector lacks equivalent longitudinal data, and direct comparison carries risk. But the directional signal is consistent enough to warrant the question: if conversion from pilot to scale is this rare even in software — where iteration is fast, costs are low, and integration complexity is comparatively manageable — what should the prior be for hardware, where none of those conditions hold?

Open Questions

Measurement

– How many of the conversations logged as “customer engagement” included an explicit discussion of success criteria — in the customer’s own terms?

– What was measured at each TRL milestone? What was not measured?

– If commercial readiness had been tracked as a parallel metric from TRL 4 onward, what would that curve have looked like?

Incentive

Internal innovation champions inside large organisations often have a structural interest in running pilots that differs from the startup’s interest in converting them. Being the person who “brought in the startup” generates internal visibility. The pilot produces data that strengthens their position — regardless of whether a commercial agreement follows. The startup interprets sustained engagement as a buying signal. It may be something else entirely.

– Who inside the customer organisation benefits from a pilot that never converts — and would they describe it that way?

– Which stakeholders in this ecosystem have an interest in technical milestones being the primary measure of progress?

Temporal

– At what point did the gap between technical maturity and commercial maturity first become measurable? Was it visible then?

– How long does a typical sales cycle in this sector take from first qualified conversation to signed agreement — and when was that clock started?

Comparative

– In sectors where commercial engagement is structurally required earlier — medtech design partners, defense co-development contracts — does the failure pattern look different, or the same?

– Which teams in this space advanced both curves in parallel? What made that possible?

Counterfactual

– At the moment the pilot ended, did the startup and the customer use the same word — “success” or “failure” — to describe the result?

– If they used different words, when did their definitions diverge — and was anyone in the room when it happened?

– Who in the customer organisation has the authority to define what operational success looks like — and have they been in any of the conversations?

– If the person funding the pilot cannot describe the conditions under which their organisation would deploy at scale, who can — and what would it take to get them in the room?

Destruction Desk
We perform autopsies on innovation’s failed assumptions.

This newsletter was edited by Manfred Lueth.

You received this email because you signed up for this newsletter from DestructionDesk.com. To stop receiving this newsletter, unsubscribe or manage your email preferences.