How Do Startups Sell to Industrial Enterprises Without Proof?

How can a startup win industrial enterprise customers without a track record?

Show up in person, do real work before anyone pays, and stay narrow. Traction, a physical AI company that monitors industrial machines, followed this path. Its CEO, Igor Marinelli, visited 88 facilities in the first year, hired a sales development rep (SDR) before any other role, and mapped customer sites for free. It took about two years from founding to the first "yes." The company now reports over $100 million in ARR.

What does Traction sell, and why is it a hard sale?

Traction builds monitoring infrastructure for critical industrial assets such as chillers, HVAC units, compressed air systems and robotic palletizers. It combines proprietary hardware with AI. The sensors cover vibration, ultrasound, magnetic field and current, and the company describes them as non-invasive. The AI first classifies the condition of the asset, then other models judge how urgent it is. Marinelli compared the first step to Shazam identifying a song. The urgency step uses LLMs with context such as OEM manuals, data sheets and a memory layer, because what counts as critical differs between plants.

The company pitches it as observability for the physical world. Software teams already know the idea from logs and uptime tools. Traction applies it to machines, with failures grouped into mechanical, electrical and operational causes.

The sales difficulty is structural. Marinelli described a chicken-and-egg problem. The system needs a lot of data to work, collecting data requires the hardware and software to be ready, and the customer has to hand over some of its most important equipment before seeing proof. A minimum viable product does not work here. As Marinelli put it, repeating advice he credited to a YC partner, it has to be a minimum awesome product. A missed failure on real machinery is hard to recover from.

How did the founders run discovery?

They went to the plants. Traction was founded in September 2019, and Marinelli said that in the first year he visited 88 facilities. He was on the road constantly, talking to operators and founders.

His method was to look for overlap. He compared problems across pharmaceutical lines, food and beverage plants and paper production, and asked whether solving one customer's problem would solve the next customer's too. He said the team avoided generalism and resisted adjacencies. That produced a product that is domain specific (machine health) while staying sector agnostic. Much of the equipment, an HVAC unit for example, is the same whether it sits in a home or a factory.

His motivation was personal. His father was a plant manager at a paper manufacturing facility and complained about the technology he worked with. His co-founder's father spent about 35 years in a cement factory. Founders who know the buyer's daily frustrations tend to ask better questions in discovery.

Why was an SDR the first hire?

The founders were all technical, and Marinelli said they assumed they could figure out sales themselves. They still hired an SDR first, because in their industry building the product and building pipeline have to run in parallel. The sales cycle was long, with roughly two years from founding to the first yes, so waiting to start pipeline would have left a gap.

The lesson for hardware and AI founders is about timing. If the product needs years of refinement before it earns trust, someone has to be generating conversations the whole time. Founders can run the technical conversations, but outreach volume is a separate job.

Are on-site visits really cheaper than marketing?

Marinelli says yes. He said he reached that conclusion after arguing it through several board meetings. Traction ran 2,500 on-sites in the last year. His comparison was the rising cost of inbound traffic from Google against the cost of a rental car, a flight and a hotel. Adding it all up, he argued, an on-site moves a deal further along.

The reason is what happens during the visit. Traction does a risk assessment and maps the whole facility. It identifies where observability is needed and which assets need protection. The customer gets a concrete picture of its own plant, and the vendor has already done part of the work before any contract exists. Marinelli said the point is to surprise the customer by showing up and mapping their struggles.

Where do paid pilots and mutual action plans fit?

The portion of the interview available here does not detail Traction's pilot terms or its use of mutual action plans, so treat this section as general practice, not a description of Traction's contracts.

A paid pilot is a limited, priced deployment with defined success criteria. It filters out buyers who are not serious. It also gives the customer's champion something to defend internally. A mutual action plan is a shared document listing steps, owners and dates for both sides, from installation through the decision. For a hardware product, the pilot criteria usually cover things like which assets are instrumented, what failures must be detected, and how much lead time counts as success.

These tools address the same trust gap that Traction faced. The free risk assessment lowers the cost of saying yes to a conversation. A scoped pilot lowers the cost of saying yes to deployment.

What does the sales approach look like in practice?

Does this playbook transfer to other startups?

Parts of it do. Founder-led discovery, early pipeline generation and doing real work before the contract apply to most enterprise sales with high trust barriers. The on-site volume is harder to copy. It suits a product where the buyer's physical environment is the sales material, and where deal sizes justify travel costs. A software startup selling to a different kind of buyer should test whether in-person work pays back before committing to it.

Also note the timeline. Traction reached its first yes after about two years, and the company went through YC in 2021 with no revenue. Patience and runway are part of the playbook.