What We Learned Building AI for People Who Don’t Trust AI

We sell AI-powered intelligence to energy suppliers and retailers. These are people who have spent their careers making million-dollar pricing and procurement decisions with spreadsheets, experience, and gut instinct. They are not, as a rule, excited when someone walks in and says “let the AI handle it.”

And they shouldn’t be.

The energy industry has been burned by technology promises before — platforms that looked brilliant in a demo and fell apart in production, automation that worked until it didn’t, and analytics tools that generated noise instead of answers. When your business runs on real-time pricing, regulatory compliance, and risk positions that shift by the hour, “trust the algorithm” is not a compelling pitch.

So we didn’t make that pitch. Instead, we spent years learning what it actually takes to earn trust in an industry that has every reason to be skeptical. Here’s what we found.

Nobody trusts a black box

The fastest way to lose an energy professional’s confidence is to give them an answer without showing the work. Suppliers don’t want a green light that says “price this deal.” They want to see the PTC rate, the matrix price, the headroom calculation, the historical trend, and the competitive signal — and then make the call themselves.

This is something our CEO, Michael Parrella, talks about often. As he puts it: “Let people do what they’re good at and let the machines do the rest. Let the machines get rid of the manual labor and let the humans handle the nuance and the judgment.” (You can hear more of Mike’s perspective in his recent interview on AI and energy leadership.)

So we designed every output in our platform to be explainable. When our system fires a signal that says your competitive position shifted in a utility zone, it shows you exactly what changed, when it changed, and what the data looked like before and after. The AI synthesizes. The human decides. That’s the deal.

Build for the live floor, not the demo room

There’s a pattern in energy technology that Mike nails precisely: “They build for demos instead of live action. And they build in an echo chamber without getting very early feedback from their customers. It’s really easy to make something look smart in a slide deck, but it’s harder to make it run live in PJM with real traders without breaking.”

We learned this the hard way. Early on, we built features we thought were impressive. Our customers told us they were useless — not because they didn’t work, but because they didn’t fit into how retail energy teams actually operate at 7 a.m. when prices need to be set and the market’s already moving.

The fix was simple in concept, painful in practice: we put our team inside the workflow. We sat with suppliers, pricing analysts, and operations staff to understand not just what data they needed, but when they needed it, in what format, and what they did with it in the next thirty seconds. We stopped building features and started building moments of clarity.

Automation without guardrails is just faster failure

The energy industry has already seen what happens when automation runs unsupervised. Algorithmic trading bots that go sideways. Automated processes that compound errors instead of catching them. As Mike observes: “Folks have ran right to automated algo bots and trading with bots and then they go, ‘oh, well, the bot went sideways and put me in a ditch.’ The key here is to always have the right amount of command and control.”

This is why we designed our platform around the principle of “trust but verify.” Our AI generates conclusions, but it frames them as recommendations, not actions. It tells you what to do, not does it for you. The human always has the last word. And every recommendation comes with the evidence trail that got it there.

For energy executives evaluating AI tools, this distinction matters. Ask your vendor: does your AI take action or inform action? If the answer is “it handles everything,” that’s not a feature. That’s a risk.

The user is the hero, not the technology

This is probably the hardest lesson for any technology company to internalize. The natural instinct is to show off the engine — the algorithms, the data pipeline, the scale. But energy professionals don’t care about your architecture. They care about whether they can go into a pricing meeting with confidence, whether they’ll catch a competitive shift before it costs them margin, and whether the tool respects the expertise they’ve built over decades.

Mike’s framing of this is direct: “You have to design for reality and you always have to keep your users as the hero.”

We took that literally. Our Headroom Intel platform doesn’t position itself as the decision-maker. It positions the supplier, the pricing analyst, the operations lead as the decision-maker — and gives them the intelligence to make that decision faster and with more confidence than they could before. The AI is the co-pilot. The human flies the plane.

Trust is earned daily, not sold once

The biggest misconception about selling AI into a skeptical industry is that trust is a sales problem. It’s not. It’s an operations problem. You earn trust the first time your system catches something a human missed. You earn it the tenth time your daily signal is accurate. You earn it the hundredth time someone opens your platform at 7 a.m. and it’s already done the analysis they were about to spend an hour doing manually.

And you lose it in a single moment — one bad signal, one unexplained output, one instance where the system says one thing and reality says another.

So we obsess over accuracy. We obsess over transparency. And we obsess over making our intelligence arrive at the user before they have to go looking for it, because a system that’s only useful when you remember to open it isn’t a system of action. It’s just another dashboard.

What we’d tell other builders

If you’re building AI for energy — or any industry where the stakes are real and the users have seen it all — here’s the short list:

• Show the work. Every AI output should be explainable and auditable. If you can’t explain it, don’t ship it.
• Build where the user is, not where you wish they were. Get into their workflow before you redesign it.
• Recommend, don’t act. Give humans the intelligence and let them make the call. “Trust but verify” isn’t a limitation. It’s a feature.
• Make the user the hero. Your technology should make them look brilliant, not replace them.
• Earn trust every single day. One accurate signal at a time. There’s no shortcut.

The energy industry doesn’t need more AI hype. It needs AI that works, that’s transparent, and that respects the people who use it. That’s what we’re building.