The AI landscape in commodity trading is moving fast enough that Commodity Technology Advisory's report, AI in Energy and Commodities, is already in its second edition after just a single year.
Comparing the two editions shows how much has changed already: agentic AI (AI systems that can act autonomously to carry out multi-step tasks) was a roadmap goal a year ago, but now is already gaining substantial traction. The industry is also moving toward a world where companies can orchestrate AI workflows across their entire operations. And a shift toward usage-based, token-driven AI pricing has made the true cost of running AI far more visible to businesses, in some cases turning monthly bills that were previously around $100 into several thousand dollars. That visibility is turning AI spend into a genuine budget line, one businesses need to actively manage and justify, not simply absorb as background overhead.
Despite all that change, one theme has stayed constant across both editions: data quality and structure remain a deciding factor in the return on AI investment.
The data gap remains
Across nearly every use case the report examines, from trading decision support, to document automation, to natural language interfaces, the same theme keeps surfacing: the quality and structure of the underlying data is what ultimately determines how well AI performs.
The report notes that a significant portion of analysts' time is still spent on data preparation: collecting information from multiple providers, aligning time series, and reconciling discrepancies. When these preparatory steps become overloaded, decision-making on the trade desk slows, delaying trades and ultimately hitting the bottom line, regardless of how capable the AI on top of it might be.
The report also notes that a significant proportion of hallucinations and unreliable AI outputs can be traced back to deficiencies in the underlying data. When AI gets something wrong, the fault often lies upstream, in the data it was working from, not in the model doing the reasoning.
Where AI is actually delivering value today
The report is refreshingly candid about where AI has and hasn't matured. Information services and reporting are singled out as the most established use case, with AI now considered close to a standard requirement for many trading organisations. Process automation in back-office and operations functions is identified as the second major growth area.
By contrast, the market is more cautious about front-office automated decision-making, long-term price forecasting, and fully natural-language-operated CTRM systems. On forecasting specifically, it notes that discussions at this year's Price Forecasting Summit in Amsterdam centred on fundamental models and scenario-building rather than pure AI pattern recognition, since AI-driven methods hold up well for short-term forecasts but less so over longer horizons. This lines up with what we've long argued: AI should support trading decisions, not replace the judgement of experienced traders.
Governance and cost remain real concerns
The report highlights ongoing industry concerns around data security, governance, and the "black box" nature of some AI systems, echoing guidance from the Committee of Chief Risk Officers on the need for clear data lineage, access controls, and human oversight in any AI-enabled workflow. It also points to a genuinely new pressure this year: with AI costs becoming more visible under token-based pricing, the report predicts organisations will increasingly need to justify AI spend against measurable value.
Where Gen10 fits
We were glad to see Gen10 named in the report among vendors making genuine progress in this space, specifically for supporting both conversational and agentic workflows natively within our platform. It's a useful external check on something we've said consistently: AI is only as good as the data behind it.
CommOS, our Commodity Management System, has always been designed to provide a single, clean, live source of operational data. This has always been an important business need, but now AI has made it critical. That's exactly why CommOS is such a strong foundation for NaNi: it's working from data that's already structured and reliable.
On top of that, the AI capabilities built into NaNi sit behind their own governance and orchestration layer, so the same access controls, audit trails, and human oversight the report calls for are part of how NaNi operates from the start.
The report's broader conclusion is one we'd agree with. AI is becoming a permanent part of the commodity trading technology landscape, but its success depends on patience, strong governance, and a data foundation that's genuinely ready to support it.