‘Every trader hates using their system’ — this is a sentence we’ve heard time and again. For firms, the current tech landscape for commodities — CTRMs, ERPs, chartering software, etc. — provides some important benefits. Data exists in a secure and auditable system of record, operations and trades are recorded, and some reporting work is fully automated. But the reality is that for most traders, operators, and analysts, these tools simply do not ever automate a large share of what they need to do each day. Business-critical work is still carried out in manual processes, by email, phone, and paper. What’s more, the technology used by firms today typically adds more work, not less — in addition to trading, scheduling voyages, and managing positions, you also need to capture trade tickets, book movements, and reconcile PnL when you close book each day.
Why is 90% of what could be automated on a commodity trading desk still manual, and why do users associate the technology they use with pain and frustration?
Traditional CTRMs and ERPs work well as systems of record. They store data for you, calculate the key commercial, risk, and financial metrics, and provide some reports that a risk manager or accountant can use to understand the state of the business. They disappoint, or fail completely, when firms try to use them as a workflow automation layer.
There are two main reasons for this. Systems of record rely on standardized, structured inputs, and the raw inputs of physical commodity trading — confirms, contracts, shipping documents — are natively unstructured and usually non-standard. That means there is inevitably a human doing manual work to create clean, structured data before any automation can begin. The second reason is that commodity trading firms frequently operate processes that cannot be codified straightforwardly with traditional business logic. This is not to say that their processes are necessarily complicated — rather that they elude definition by means of hard-coded if-then statements. Instead, they are rules of thumb — frameworks for how to operate that require institutional experience and judgement to carry out appropriately, and where it is not possible to anticipate in advance every edge case that may arise. This means that when commodity traders do implement workflow tools in their CTRMs and ERPs, they often fall short and users need to work around them to handle exceptions.
What can AI do to solve these problems for commodity trading firms, and how? Our thesis is that commodity trading is one of the industries that is most suited to reap the benefits of the AI revolution. Let’s start with the general principle: AI can actually do the manual work that you want automated. It can read and draft trading documents of any type and format, translate that to structured records, interpret and explain business data, and manage communications between people on the desk and trading counterparties. It is able to operate using rules of thumb, apply judgement in edge cases, and escalate to a human in the loop where needed. These are new fundamental capabilities that unlock automation for the 90% of use cases that traditional software and business logic have been unable to touch.
How do we provide these new capabilities to a commodity trading firm safely, effectively, and cheaply? We believe there are four pillars:
AI for commodities needs to be specialized and dedicated for the domain. Generalist AI productivity apps are great if your job is to produce PowerPoint decks, but do not translate well if your job is to get crude oil from a port on one side of the world to a terminal on the other. When we talk about specialist AI for commodities, we mean that every lever available — model, agent guidance, tool construction, memory and learning architecture — is designed with commodity trading in mind.
AI for commodities needs to sit outside, and integrate with, existing systems of record. A CTRM is good at securing data and calculating VaR. It should continue to do those things, and agents should take over the workflow layer, reading and capturing data in systems of record when they need to. This lets firms keep the benefits of their current systems, and adopt a more automated, flexible enterprise architecture.
AI for commodities needs to be responsive to external events. Most AI apps today serve single users, and begin working when asked a specific question. A physical commodity trade lifecycle involves multiple people inside and outside the firm collaborating to execute tasks, and timeliness of action is critical to doing so successfully. AI for commodities therefore needs to be inherently a multiplayer, event-oriented solution.
AI for commodities needs to be safe and governable. Businesses need role-based access management, proper delegation of authority, as well as tracking and auditability for any process — even those carried out by humans. Agentic workflows need to be governed by exactly the same principles, and enforcement needs to be fully deterministic.
Emporic was founded to be a solution to bring the AI revolution to commodity trading. See what we can do for your business.