AI promises to transform energy retail but without trustworthy data and traceable calculations, even the most convincing answers can be dangerously wrong.

Ask an AI tool why your margins fell last quarter and it will produce a convincing explanation. It might even sound like something your own team would say. But can you trust it? That’s the challenge facing energy retailers. AI can produce plausible answers without producing correct ones. And in an industry where pricing, settlement and contract decisions directly affect profitability, that distinction matters.

I call it the plausibility trap and, at Gorilla, we’ve experienced it ourselves.

The problem isn’t a lack of ambition

Despite the hype, most retailers remain cautious about AI and understandably. Legal, security and IT concerns are slowing adoption, while operational teams are already stretched. AI is making progress in customer service, communications and marketing but its greatest potential lies in the operational core, where decisions directly affect margins.

And the commercial stakes are significant. Gorilla’s 2026 Energy Margin Intelligence Index found that just 31 per cent of energy retail leaders say their organisations can accurately and consistently quantify margin leakage. Meanwhile, the majority estimate that between six and ten per cent of revenue is at risk through limited visibility and slow decision-making.

AI could help retailers close that intelligence gap but introducing it into an environment where the underlying numbers are already difficult to reconcile risks making a familiar problem considerably more dangerous. Retailers are right to proceed carefully but waiting isn’t a strategy.

We learned the hard way

At Gorilla, AI is embedded in how we develop our products, yet when we introduced AI capabilities, we encountered an uncomfortable problem. Where underlying data or calculations had gaps, AI confidently filled them. The answers sounded convincing but they weren’t always correct.

Fixing this required multiple iterations, largely focussed on strengthening the deterministic calculations beneath the AI. Today, answers are traceable and auditable, with uncertainty clearly identified. Users can verify the evidence rather than simply trust the explanation. That experience taught us something important: AI is only as trustworthy as the foundations supporting it.

AI can help build those foundations

There’s a misconception that retailers must prepare their data before introducing AI. Our experience suggests otherwise. We now use AI to continuously interrogate our product, identifying questions it cannot reliably answer. Each exposes a gap we can address before customers encounter it.

That changes the relationship between AI and data readiness. Rather than waiting for every system, dataset and calculation to be perfected, retailers can use AI to expose inconsistencies, challenge assumptions and identify where the underlying information needs attention. The technology becomes part of the process of building trust, not simply the reward for having achieved it.

The same principle applies to energy retail. AI can help identify weaknesses in data and calculations, while accelerating the work needed to resolve them. Once those foundations are reliable, AI becomes far more valuable; exploring scenarios, testing decisions and recommending actions grounded in verifiable information.

Crucially, those foundations are also what get AI past sign-off. The greatest opportunities lie in the operational core, precisely where legal, security and IT teams are right to demand the highest standards. When every number can be traced to a deterministic calculation, every recommendation can be checked and uncertainty is made explicit, the conversation changes. AI stops being a leap of faith and becomes something that risk and governance teams can meaningfully assess.

The next twelve months matter

Retailers should use this time to strengthen their data foundations, involve risk and security teams early, and experiment in controlled environments. The goal shouldn’t be deploying AI everywhere. It should be making AI trustworthy enough to use where it matters the most. The question isn’t simply whether AI can produce an answer, it’s whether the business can explain, verify and stand behind it. Those retailers that succeed won’t necessarily use the most AI, they will be the ones who can trust what it tells them.

‍