Trust falls at 200 megawatts.

For three years I was the CMO of a Polish startup applying AI to solar photovoltaic diagnostics. Solar Spy. We built a model that could read inverter telemetry, panel-level performance data, and a few thousand hours of historical fault logs, and tell an operations director which of his 600,000 panels was about to underperform. The model worked. The product worked. The marketing did not work, for the first eighteen months, and I want to tell you why.

There is one slide in every AI-for-heavy-industry sales deck that never works, and almost every AI marketer keeps putting it in. It is the accuracy slide. The "92.4% predictive precision" slide. We had it. It cost us deals.

Who the buyer actually is

The operations director at a 200 megawatt solar farm is not a tech buyer. He is a man (it was always a man) who has been awake at 3am for the last fifteen years restarting inverters, calling Chinese manufacturers about warranty claims, and arguing with insurance adjusters about hail damage. He has read every failure log of every system he has ever touched. He does not buy AI. He buys the model he has personally read the failure log of.

Three things I learned

Ⓐ The accuracy slide makes him suspicious. "92.4% precision" reads to him as "7.6% of the time, you are going to wake me up at 3am for a panel that is fine." He cares about the failure modes, not the success rate. The deck never showed the failure modes. We were optimising for the wrong number.

Ⓑ The pilot is the product. Every deal we closed, we closed because we ran a six-week pilot on his own farm, on his own data, and he watched the model be wrong and corrected it. The trust was not in the slide. The trust was in the time he had personally spent inside the model's mistakes. We started pricing pilots seriously eighteen months in. The whole company changed.

Ⓒ The reference customer is unbeatable. One operations director who has run the system for a year is worth more than ten Gartner mentions. Industrial people trust other industrial people. Not analysts. Not journalists. The entire AI press industry is solving for the wrong audience.

What we changed

We threw away the accuracy slide. We replaced it with a "here is the failure log of the model in the last 90 days, organised by site" slide. Deals started closing. We replaced "thought leadership" content with field notes from an actual operations director. Pipeline started moving. We replaced our keynote at the AI conference with a panel at the PV conference. The conversion rate on conversations doubled.

Trust does not scale in heavy industry. It is hand-built, one operations director at a time, one failure log at a time. The companies that figure this out are the ones still standing in 2030.

If you are marketing AI into heavy industry and your deck still has the accuracy slide, take it out tomorrow. It is costing you deals. If you are an operations director reading this and an AI vendor cannot show you the failure log, do not sign the contract. They are not ready.