NOTE 002

How to break into industries that run on relationships

How to break into industries that run on relationships

Executable systems for earning trust, finding the right people, and turning one credible relationship into recursive market access.

01 / OPERATING METHOD

Turn one relationship into the next.

Turn one relationship into the next.

Map the system, find a credible way in, earn trust through useful work, then use what you learn to improve the map and open the next door.

SWIPE TO EXPLORE →

SWIPE TO EXPLORE →

02 / ANALYSIS

A physical AI company eventually has to walk into somebody else’s world.

The factory already has machines, maintenance schedules, procurement rules, integrators and operators who have seen new vendors come and go. Whatever gets deployed has to work inside that system.

The obvious answer is to hire someone who already knows the place. A fifteen-year automotive operator knows who matters, which vendors have credibility, how plants buy and what has already failed.

Hire that person when the problem is important enough and the expertise is available.

“If you’re not connected to the technology and what you’re doing with it, everything else that stems from that can become misguided.”

“If you’re not connected to the technology and what you’re doing with it, everything else that stems from that can become misguided.”

Early operator · Growth-stage industrial energy · Commercialization + strategy

The harder case comes with expansion. Each new industry, geography or deployment adds another operating system to understand. The number of systems grows faster than the number of twenty-year insiders a company can hire.

I ran into this in advanced nuclear. Joining Applied Atomics without a nuclear career meant working across utilities, regulators, governments, investors and suppliers in an industry built on long-standing relationships.

Established industries expose small credibility gaps quickly: which organizations matter, which titles carry weight, which acronyms everyone assumes you know. None of that is expertise, but getting it wrong marks you as an outsider.

For two weeks, I increased learning density: each conversation improved preparation for the next. The goal was simple: remove enough information disadvantage to ask better questions and make useful decisions.

The early work was reconstruction: who mattered, how utilities evaluated projects, which markets were serious and who could actually move something. Research built the first model; good conversations changed it.

That reconstruction is getting cheaper. AI can build a first map, network tools can expose paths through existing relationships and conversations can be captured and compared. The advantage is not instant expertise. It is getting to better questions sooner.

01 / Build a working map

01 / Build a working map

Start with the company outcome. “Learn automotive” or “understand utilities” can absorb months without producing much useful direction. A bounded problem forces the map to become specific.

For a robot company entering a new plant, map the buyers and operators, equipment, integrators, maintenance practices, safety requirements, incumbent vendors and deployment constraints. An energy company will map different actors. The test is the same: do you know where the important unknowns are?

The map also needs a dictionary. Industry fluency means knowing how different people describe value. Engineering may care about performance, the CFO about dollars, the operator about how many buttons they have to click. Learn their language, then explain the technology in those terms.

“Learning an industry’s language means understanding how different people inside it describe value. The engineer, CFO and operator each care about completely different things. You need to shift as needed to make the message land."

“Learning an industry’s language means understanding how different people inside it describe value. The engineer, CFO and operator each care about completely different things. You need to shift as needed to make the message land."

Early operator · Growth-stage battery technology · Commercialization + customers

The first map does not need to be comprehensive. It needs to be good enough to start testing.

The loop can run quickly. A few conversations kill assumptions and change which customer is worth visiting. A site visit exposes an installation constraint. The deployment plan changes. By the next pass, the map is different.

02 / Get into the system

02 / Get into the system

Once the map gets decent, get off the internet.

The useful entry point is not always the executive buyer. It may be an integrator, plant manager, technician, supplier or former operator who can show where the map is wrong. One good relationship can save weeks.

Relationship-driven industries are often denser than they look. The company may already be one or two credible introductions from the person it needs. Network tools make those edges easier to see.

The introduction still consumes somebody’s reputation. That decision needs judgment.

How you enter matters. Learning creates a different conversation than selling. Being new gives temporary permission to ask basic questions and admit what you do not understand. If curiosity is obviously a pretext for a pitch, that window closes.

“There’s a window where being new gives you permission to ask basic questions. If people think those questions are just a pretext to sell them something, that window closes quickly.”

“There’s a window where being new gives you permission to ask basic questions. If people think those questions are just a pretext to sell them something, that window closes quickly.”

Early operator · Growth-stage battery technology · Commercialization + customers

SWIPE TO EXPLORE →

SWIPE TO EXPLORE →

Places matter. If eight of the twelve people who keep appearing in the research will be at the same working group in Detroit, go to Detroit. Book the coffees. Take the plant tour. Talk to the person who owns the machine after the vendor leaves.

Use research to concentrate field time. A 30-person working group with half the relevant people can be worth more than a 5,000-person conference.

Then the work looks pretty ordinary. Show up, listen, follow through, come back when there is a reason to.

“A lot of trust gets built through small tactical executions. Help with the deck. Find the missing information. Get the right people together. Eventually people start pulling you into more consequential work.”

“A lot of trust gets built through small tactical executions. Help with the deck. Find the missing information. Get the right people together. Eventually people start pulling you into more consequential work.”

Senior operator · Late-stage defense technology · Strategy + operations

Conversation creates the opening. Work makes it durable. One defense operator described earning trust through small tactical executions: help on a deck, find the missing information, get the right people together. Do that well and the next request gets more consequential.

In one early market map, commercial opportunity was doing too much of the ranking. A market could contain plausible customers and still be a poor place for a first deployment if policy, local institutions or existing relationships created too much friction.

The screen changed. Market size and customer demand stayed in the model, alongside institutional resistance, available trust and the company’s ability to matter locally. That was enough to push some obvious markets down the list and elevate smaller ones where an existing partner could provide a credible entry point. The partner network then became part of the entry strategy: map the relationships, borrow credibility where appropriate and let introductions compound.

For first-of-a-kind deployment, the surrounding system belongs in the market map. Policy, local credibility, partners and the density of useful relationships can change which market is actually executable.

For a new technology, credibility can also come from something working somewhere else. An MVP, prototype or validation in an adjacent sector gives the next customer evidence that somebody has already absorbed part of the first-mover risk. In relationship-driven industries, being able to say that another credible company has tried the technology can matter as much as another performance claim.

03 / Let reality wreck the map

03 / Let reality wreck the map

The first few conversations should break the model. The assumed buyer does not own the budget. Three people name the same integrator. A site visit turns a product problem into an installation problem. A technician surfaces a failure mode executives never mentioned.

Those corrections are the work.

Information gets better with access. Early conversations repeat what is online. Later ones explain why the last deployment failed, who can quietly stop a project and what has to be true before a facility takes the risk.

Capture enough to compare. Link calls, people, site visits and decisions. Use AI to surface recurring names, contradictions and changes in assumptions.

More capture also creates more junk.

Private does not mean shareable. Repeated does not mean true. A confident synthesis can still rest on weak sources. Some information should stay where it was shared.

Review field intelligence before it becomes company knowledge.

CAPTURE → REVIEW → ROUTE → RETAIN

The first few conversations should break the model. The assumed buyer does not own the budget. Three people name the same integrator. A site visit turns a product problem into an installation problem. A technician surfaces a failure mode executives never mentioned.

Those corrections are the work.

Information gets better with access. Early conversations repeat what is online. Later ones explain why the last deployment failed, who can quietly stop a project and what has to be true before a facility takes the risk.

Capture enough to compare. Link calls, people, site visits and decisions. Use AI to surface recurring names, contradictions and changes in assumptions.

More capture also creates more junk.

Private does not mean shareable. Repeated does not mean true. A confident synthesis can still rest on weak sources. Some information should stay where it was shared.

Review field intelligence before it becomes company knowledge.

CAPTURE → REVIEW → ROUTE → RETAIN

Ask four things: Is it reliable? Can it travel? Who needs it? Does it change anything?

Route only what matters. A founder may need a brief before a customer meeting. An engineer may need one detail that changes an integration assumption. A repeated objection may change qualification. Everything else can stay in the notes until there is more evidence.

In another market, public planning documents and upcoming public discussions made two infrastructure constraints unusually important before the first serious customer conversation. That preparation meant the meeting could start with a working hypothesis about the customer’s physical system rather than basic discovery.

What came back moved in several directions. It changed market prioritization, informed preparation for similar customers and surfaced a technical consideration relevant to deployment in comparable environments. Field intelligence rarely belongs to only the person who collected it.

04 / Make the learning travel

04 / Make the learning travel

Physical AI companies accumulate knowledge in awkward places: deployment, sales, founder relationships, field operations. Six months later, a new person can still start from scratch because those lessons never entered the operating system.

Repeated field learning should eventually change the machinery of the company.

Repeated learning should change behavior. Recurring installation issue? Change the deployment plan. Same customer bottleneck? Update qualification. Integrator consistently helps? Make it visible before the next market. Repeated workaround? Route the pattern to product with source context.

Some of this belongs in playbooks. Some belongs in CRM structure, onboarding, account plans or deployment checklists. Some should become context available to internal AI systems. The format depends on who needs to use it and how often.

A single anecdote should not become a rule. Neither should gossip, stale information or a conclusion whose source cannot safely travel. Repetition helps, evidence helps, and sometimes the right answer is to keep the observation provisional.

Even with CRM and regular commercial cadence, the market model stayed fragmented across documents, messages, meeting notes and people’s heads. Each piece was small. Together they changed how the market was understood.

The hard part was keeping the model current and routing new information to where it mattered. Both still depended on individual operators. The artifacts existed; some connective tissue did not.

A new operator entering the same system six months later should inherit the useful parts. They should know which assumptions were tested, which relationships have history, what failed, what keeps recurring and where the company is still guessing.

05 / Where this gets interesting for physical AI

05 / Where this gets interesting for physical AI

Industry expertise remains valuable because physical systems contain things documents miss. Machines behave differently in production. Facilities have local workarounds. Experienced operators notice what newcomers do not know to ask.

The change is in how much reconstruction has to happen before a capable outsider can become useful.

A small team can now arrive with a reasonable first model, find people who can correct it, use relationships deliberately and carry field learning into the next deployment. That creates room for operators who can move across customer, product, deployment and partnerships without arriving with every answer.

That does not make technical understanding optional. One early operator at a late stage energy company described understanding the technology as the starting point for everything else they did across the company. Without enough grounding in how the system actually works, the connections drawn across commercial, deployment or organizational problems can be confidently wrong.

There are limits. Safety-critical judgment cannot be synthesized into existence. Tacit knowledge can take years. Some relationships require reputation. Some lessons only make sense standing next to the machine when it fails.

Those limits are part of the operating problem. The useful question is where deep inherited expertise is required, where it can be borrowed from customers, advisors and partners, and where a strong cross-functional operator can learn fast enough to own the work.

As physical AI companies enter more facilities and industries, they will answer that question repeatedly. The good ones will leave more behind each time: better maps, relationships, deployment assumptions and operating rules.

THE OPERATOR NETWORK

Built with people actually building frontier companies.