Our longstanding partners at Commodities People know how to get commodity professionals talking, and Digitalisation in Commodities Online was no exception, with 4 riveting panel discussions and tons of audience engagement. Gen10 were an event partner, and we're grateful to the team at Commodities People for hosting us.
Our longstanding partners at Commodities People know how to get commodity professionals talking, and Digitalisation in Commodities Online was no exception, with 4 riveting panel discussions and tons of audience engagement. Gen10 were an event partner, and we're grateful to the team at Commodities People for hosting us.
The event's theme was digitalisation as the foundation for smarter decisions, faster execution and operational resilience. AI ran through every session, with each panel taking a different angle on it.
Our CEO Richard Williamson spoke in one of those conversations: “Cleaning the house before adding AI: Pragmatic tech reinvestment in commodities trading”. The panel discussed a question we hear from trading firms all the time: what needs to be in order before AI can deliver real value? You can watch the full session below.
What the Session Covered
Chaired by James Gracey of JDE Peet's, the 45-minute discussion included a live panel with three audience polls and a Q&A. Richard was joined by Uche from Vitol, who works on ETRM and AI implementation.
Three themes shaped the conversation: where to reinvest, how to simplify legacy CTRM stacks and clean data, and how to rebuild the architecture for a competitive edge. In the opening poll, most of the audience named back-office efficiency as their main objective for the next 12 months.
That result matches much of what we see at Gen10. The back office is where much of the repetitive work in a trading business sits, so AI can add real value here. That's the thinking behind NaNi, our AI assistant. She works inside CommOS, so she can create a draft booking from the text of a confirmation email, compare contracts, and check everything in a shipment against the Letter of Credit.
Five Takeaways From Richard
Prepare Your Data First (44:28)
Richard's closing remarks put it plainly: “The work is not on AI investment. It should be on preparing your data.” In his view, the industry needs to ensure their data is understandable to AI if they expect to see results. That is what we have built at Gen10. Drawing on 25 years of expertise across over 100 commodities, we have created a semantic layer that sits on top of each client's own data, workflows and processes, so the AI can truly understand it.
A Data Lake Takes You Far, but Not All the Way (20:58)
Richard sees the value in data lakes providing analytics and visibility, particularly for firms with several CTRMs across multiple geographies or assets, and adding a semantic layer helps AI make sense of them. But for agentic work, such as creating contracts, making allocations and processing invoices, the AI needs to sit inside the access control and workflow rules of the system where the work happens. A data lake alone can't get you there.
Orchestrate Your Agents Inside the CTRM (30:23)
Once agents start acting rather than just answering questions, they need governing. Richard talked through the practical questions: which LLMs can the agents use, what are they meant to be doing, and how many tokens are they spending? An orchestration layer coordinates the connections between your systems and models, keeps the agents under control and lets you monitor token spend. His view was that the best place for that layer is inside your CTRM, where access control and workflow rules already apply.
Get the Best From Deterministic and Probabilistic Systems (30:55)
LLMs are probabilistic, and a CTRM can't afford to guess. Richard's approach is to put tests in place, and to design agents that write and run queries to fetch numbers from your data rather than calculate them. That plays to each system's strengths: the AI model handles the language and your data system supplies the numbers. Done properly, you cut the risk of hallucination.
Protect Your Positions (40:41)
An audience member asked how to stay safe when AI touches sensitive information, and whether a local server is the only answer. Richard advised against putting positions into public AI tools. Providers may promise not to use your data, but a stronger route is a ring-fenced AI inside your CTRM environment, where it cannot train on your data. This can be cloud-hosted, and for smaller firms in particular, he said local servers bring single points of failure, key-person risk and the cost of GPUs.
Uche added a practical checklist: be clear on what the AI can do and who is accountable when it gets something wrong, give it only the data it needs at that point, make sure it respects the permissions people already have, and start with read-only access. If it can change anything in a live system, that system should enforce the same limits and approvals it would for a person. As you get comfortable, you can monitor it and widen its access.
This is also why we built NaNi inside CommOS, with Conductor.AI as the governance layer. Access is role-based, every action leaves an audit trail, there are controls on usage cost, and a person approves anything that moves money, position or risk.
Where to Go Next
If you're thinking about how to adopt AI safely in your own business, our whitepaper, Charting a Safer Course at Full Speed , is a free 25-page guide to the practical side of adopting AI. It explains why AI needs clean, connected data, how CommOS, NaNi and Conductor.AI fit together as one governed architecture, and how to choose between point solutions and an integrated foundation.