A climate pilot succeeds because a founder, a champion, and a handful of engineers work around every gap by hand. Scale fails when that heroics-dependent setup meets deployment six, where nobody has the founder's phone number. Crossing the chasm means converting tribal knowledge into documented, standardized, and automated systems before volume arrives.
Quick Answer: A pilot runs on heroics — a founder's cell number, a champion smoothing customer friction, configs hand-tuned by whoever happened to be on-site. Scale needs the same result from a team that has never met the founder. The work between deployment one and deployment ten is "de-heroing": turning what one person knew into what a documented, standardized, and eventually automated system does on its own.
The Chasm Is Organizational, Not Technical
The gap between a climate pilot and repeatable scale is rarely a product gap — it's a dependency gap. A pilot works because specific people compensate for everything the system doesn't yet do, and that compensation disappears the moment volume outpaces the founding team's calendar.
Geoffrey Moore's Crossing the Chasm described a related but distinct problem: early adopters buy a vision, the early majority buys a working reference. Climate hardware has a second, quieter chasm layered on top of Moore's — even after the market says yes, operations has to say yes too. A utility, a fleet operator, or a building owner signing deployment four doesn't want to hear that the last install took eleven site visits from your VP of Engineering.
This is where climate tech's hardware-and-software mix bites hardest. As explored in our piece on the hardware-software two-clock problem, hardware iterates in months and years while software iterates in days — and pilots quietly borrow against that mismatch. A founding engineer patches firmware overnight to cover a hardware quirk nobody has time to redesign. It works, once, because someone was watching closely enough to catch the failure and fix it before the customer noticed.
The soft costs behind that gap are well documented outside of climate tech specifically. The National Renewable Energy Laboratory (NREL) has spent over a decade benchmarking U.S. solar deployment costs and consistently finds that installation, permitting, and inspection variability across sites — not the hardware itself — is the stickiest driver of cost spread between an efficient market and an inefficient one.
Climate hardware pilots reproduce that pattern at the company level. The panel or battery is standardized, but the site work around it isn't, so cost and risk both concentrate wherever the process is least documented.
Three symptoms tell you the chasm is organizational, not technical:
- Deployment success correlates with which employee showed up, not with the product spec.
- Post-mortems produce individual heroics stories ("Sarah caught it at 2 a.m.") instead of process gaps.
- Nobody can answer "what would we do differently with our tenth customer" without naming a person, not a system.
If any of those are true of your last three deployments, you're not scaling a product yet — you're scaling a team of specific people, which doesn't scale past the size of that team. Our complete guide to energy and climate product management covers the adjacent forecasting and stakeholder problems that compound this one; this piece focuses narrowly on the deployment mechanics themselves.
Learning Pilot or Selling Pilot? Diagnose Which One You're Running
A learning pilot exists to reduce your uncertainty about the product or the market; a selling pilot exists to prove repeatability to a reference customer who will talk to your next ten prospects. Treating one like the other is the single fastest way to either over-engineer a throwaway or under-engineer a template you'll ship five more times.
The confusion is understandable — both look identical from the outside. Both involve a signed contract, an install date, and a customer expecting the thing to work. But they optimize for opposite things, and the PM who doesn't name which one they're running ends up hand-tuning a config that was supposed to become a repeatable playbook, or documenting rigor into a site you were only ever going to visit once.
| Dimension | Learning pilot | Selling pilot |
|---|---|---|
| Primary goal | Reduce product/market uncertainty | Prove repeatability to a reference customer |
| Success metric | What you learned, even from failure | Clean install, on-time commissioning, happy reference |
| Hand-tuning the config | Expected — that's the point | A warning sign the "repeatable" story is fake |
| Customer's expectation | Partnership, tolerance for rough edges | Vendor performance, near-zero rough edges |
| Right next step after it ends | Revise the hypothesis or the product | Copy the install runbook to site two |
| Who should run it | PM plus a small technical team | Whoever will actually run deployment two through ten |
A learning pilot is designed to be interrogated using something like JTBD — what job is this customer actually hiring the pilot to do, and what "forces of progress" pushed them to sign up despite the risk? Our complete guide to jobs-to-be-done walks through how to structure that interrogation instead of just collecting anecdotes.
A selling pilot, meanwhile, is really a stakeholder-management exercise wearing an engineering costume. If your pilot lives inside a utility's sandbox or demonstration program, the champion who got you in the door is often not the person who signs deployment two — our guide to navigating utility and regulator stakeholders covers how that internal handoff typically works, and why a selling pilot that only satisfies the champion, not the procurement or engineering function behind them, tends to stall at exactly one site.
Name it in the deployment plan, in writing, before day one. A pilot with an undeclared goal will get judged against whichever standard is convenient after the fact — usually the wrong one.
The De-Heroing Framework: Four Things That Must Stop Depending on You
De-heroing is the discipline of moving every pilot-critical task through three states — documented, standardized, then automated — before volume makes the founding team's personal attention the bottleneck. It applies to four recurring areas: install, commissioning, configuration, and support.
As Atul Gawande argued in The Checklist Manifesto, expert domains from aviation to surgery reduce catastrophic failure not by hiring more heroes but by writing down what the heroes already know, so a checklist — not a memory — carries it forward. Climate hardware deployment is exactly this kind of expert domain in miniature.
The maturity ladder works the same way for each area:
- Documented — someone writes down what actually happened at the last site, including the workarounds, not just the intended process.
- Standardized — the workaround becomes a rule that applies to any site meeting certain conditions, replacing "ask the engineer who did it last time."
- Automated — a tool, script, or config generator applies the rule without a human remembering to.
Not everything needs to reach step three by deployment ten. Most pilots-to-scale programs fail simply because nothing ever reaches step one.
| Area | Deployment-1 reality | What must exist by deployment 10 |
|---|---|---|
| Install | Founding engineer flies out, adapts on-site | Written runbook + trained install partner network |
| Commissioning | Ad hoc testing, "looks good, ship it" | Standard test protocol with pass/fail thresholds |
| Configuration | Parameters hand-tuned per site by memory | Config schema with validated inputs, not free text |
| Support | Customer texts the founder directly | Documented escalation tiers, a support team that doesn't include the founder |
This is also where the customer journey becomes a diagnostic tool rather than a marketing exercise. Mapping the emotional and operational curve from contract signature through the first ninety days of live operation exposes exactly where the founding team is quietly propping up a step the customer thinks is "just how the product works."
Toyota's standardized-work discipline — the backbone of the Toyota Production System that Taiichi Ohno helped build — runs on the same insight: a process can only improve once it's written down consistently enough that a deviation is visible as a deviation, not absorbed as "that's just how Dana does it."
Sequencing the ladder without stalling deployments
A PM doesn't get to pause deployments while de-heroing catches up — the work has to happen alongside live sites. A workable sequence looks like this:
- Pick the pillar causing the most founder-hours per site right now, not the one that feels most technically interesting.
- Document it during the very next deployment, treating the site visit itself as the source material rather than reconstructing it from memory afterward.
- Standardize only after the second site confirms the rule generalizes — a rule written from one data point is a guess wearing a checklist's clothing.
- Automate last, once the standardized version has survived two or three sites unchanged.
Running all four pillars through this ladder at once is rarely realistic with a small team. Picking the single pillar consuming the most senior-engineer time per deployment and clearing it first tends to compound fastest into calendar relief.
When Hand-Tuned Pilot Configs Become a Scaling Trap
Hand-tuned configuration is the most common de-heroing failure because it's invisible while it's working — the pilot performs beautifully, and nobody notices that the performance lives in one engineer's head, not in the product. It only becomes visible once a second, third, and fourth site each need their own bespoke pass.
Consider a pattern common to distributed energy deployments — solar-plus-storage, demand-response controllers, and building-energy-management systems all share a version of it. A dispatch or forecasting algorithm ships with default parameters, but the pilot site's real load profile, interconnection agreement, and weather exposure never quite match the defaults. A founding engineer manually recalibrates the model after commissioning, tightening thresholds and overriding a few edge cases by hand until the numbers look right.
That recalibration is genuine, valuable engineering work. The trap is that it never gets captured as a general rule — only as a one-off adjustment to one site's config file. Our piece on AI energy-demand forecasting and megawatt-scale errors walks through how these hand-tuned assumptions compound: a forecasting model quietly tuned to one site's load shape can be wildly wrong at the next site, and because the tuning was never generalized, nobody can tell whether the new site needs a different correction or the same one that worked before.
By deployment four or five, the same engineer becomes the bottleneck for every new site, because they're the only one who knows which of a dozen hand-adjusted parameters actually matters. Two things typically have to happen to break the trap:
- Convert the tuning knowledge into an explicit
config schema— named, bounded, validated parameters instead of a free-form file only one person can safely edit. - Separate site-specific inputs (load profile, grid code, interconnection limits) from tuning logic, so a new site supplies inputs and the system derives its own starting configuration instead of inheriting a copy-pasted file from the last install.
Neither fix eliminates the need for engineering judgment entirely — some sites will always need a specialist's eye. But it shrinks the specialist's job from "personally tune every site" to "review the handful of sites the automated defaults flag as unusual," which is the difference between a bottleneck and a safety net.
A quick self-check for whether you're already inside this trap: if the answer to any of the following is yes, the config layer needs attention before the next site is signed.
- Does a new site's config file start as a copy of the last site's, edited by hand rather than generated from that site's own inputs?
- Would a new hire need to ask a specific person which parameters are safe to change?
- Has a "quick tweak" made during commissioning ever gone undocumented anywhere outside that one engineer's notes?
Turning Pilot Learnings Into a Repeatable Playbook
Every pilot generates two kinds of knowledge: what was true only at that one site, and what will be true at the next nine. The de-heroing framework only works if a PM can reliably tell those apart, deployment after deployment, instead of relying on memory or a scattered thread of Slack messages.
This is a capture problem before it's a process problem. A learning that lives in one engineer's head, one closed support ticket, or one commissioning report nobody re-reads doesn't make it into the next site's runbook — it just gets re-learned, expensively, at deployment six.
Prodinja's Reflection journal is built for exactly this gap. It lets a PM capture what a specific pilot actually taught them — a hand-tuned parameter that turned out to be universal, a commissioning step that broke on the second try — and tag each entry against the situation it came from. Those one-off notes then accumulate into the material a real playbook gets built from, instead of evaporating once the pilot ends.
The habit matters more than the tool. A PM running five simultaneous pilots who reflects systematically after each site visit will out-scale a PM running one flawless pilot who never writes anything down. De-heroing is, at its core, an information-architecture problem: the pilot's lessons have to outlive the person who learned them.
Key Takeaways
- The pilot-to-scale gap is organizational, not technical — it's measured in how many tasks still depend on a specific person, not in product maturity.
- Diagnose learning pilot versus selling pilot before day one — they optimize for opposite things, and an undeclared goal gets judged by the wrong standard later.
- De-heroing runs on a three-step ladder — documented, then standardized, then automated — applied to install, commissioning, configuration, and support.
- Hand-tuned configs are the most common trap because they work invisibly right up until a second and third site each need their own bespoke recalibration.
- Checklists and standardized work, not more heroics, are what the highest-reliability industries actually rely on — the same discipline Gawande documented in aviation and surgery applies to climate hardware rollouts.
- Capture pilot learnings as you go, distinguishing site-specific quirks from generalizable rules, so the tenth deployment benefits from what the first one taught you instead of repeating it.
Frequently Asked Questions
How many deployments does it take to cross the climate tech pilot-to-scale chasm?
There's no fixed number — it depends on how much of the install, commissioning, and support process was already documented before the pilot. Teams that start de-heroing during the pilot itself, rather than after it, often see the chasm narrow by deployment three or four instead of ten.
What's the difference between a learning pilot and a selling pilot?
A learning pilot exists to reduce your own uncertainty about the product or market and tolerates rough edges; a selling pilot exists to prove repeatability to a reference customer and should look close to production-ready. Confusing the two leads teams to either over-polish a throwaway or under-document a template.
Why do hand-tuned pilot configurations block scaling?
Hand-tuned configs work because one engineer's judgment quietly compensates for gaps in the product, but that judgment never gets captured as a reusable rule. Each new site then needs its own bespoke recalibration, making the engineer — not the product — the bottleneck on deployment volume.
Is a checklist really enough to replace an engineer's judgment on-site?
Not entirely — a checklist replaces the routine parts of judgment, not all of it. As Atul Gawande's research on high-reliability fields shows, checklists free experts to focus their judgment on the genuinely unusual cases instead of re-deriving the routine ones from scratch every time.
What should a PM document first when de-heroing a pilot process?
Start with whatever task currently requires calling a specific person by name to get unstuck — that's the highest-leverage thing to write down first. Documenting the founder's mental checklist for commissioning usually pays off faster than documenting anything customer-facing.
Does de-heroing mean the founding team should stop visiting deployment sites?
No — it means their visits stop being load-bearing. A founder or lead engineer can still show up to a tenth site out of curiosity or relationship-building; the test is whether the deployment would succeed without them there, not whether they choose to attend.