CLIENT OVERVIEW

Koch Energy Services (KES) and Koch Canada Energy Services (KCES) market natural gas throughout North America. They provide a full line of services, including purchasing, sales, transportation, scheduling, storage, hedging, asset optimisation and market analysis. KES’s commercial origination, trading, and operations teams have the expertise necessary to provide innovative energy solutions across the natural gas and electricity value chains. KES manages more than seven billion cubic feet per day, has more than 20 billion cubic feet of leased storage space and is active across more than 100 pipeline systems.

To learn more about Koch Energy Services, visit https://kochenergyservices.com/.

Reimagining trade capture in the chat era

Physical gas traders operate at the centre of constant activity: multiple chat windows open, prices moving, and counterparties negotiating in real time. Yet once a deal is struck, every transaction must be entered perfectly into internal systems. Manual entry has long been a bottleneck, introducing fat-finger errors, reconciliation work, and compliance risk.

Koch Energy Services, one of North America’s largest physical gas traders, partnered with ipushpull to close that gap. Together, they built an LLM-enabled agent that identifies and books over-the-counter (OTC) gas trades directly from chat messages into Koch’s system of record.

The system parses trader chat in real time, extracts 12–13 booking fields per order, and handles complexities such as multiple transactions in a single thread, roll conventions, weekend adjustments, and intercompany trades. It is a practical, embedded solution that delivers straight-through processing without requiring traders to change how they work.

In addition to the difficulties around collecting information from sell-side organisations in response to their enquiry, the trades are typically described using a mix of email and Microsoft Excel spreadsheets across multiple files and formats, which leads to a manual and often inefficient workflow at the sell side. On the sell side, trade data is typically manually extracted from the files to represent the trade in internal pricing systems before any decisions are taken around the pricing of risk and any other assessment of the trade value.

The number of manual touchpoints within the process and the lack of data connectivity between the buy and sell sides lead to an inefficient and costly process, with operational risk being a significant concern among all participants. Client service can suffer too, due to the unstructured nature of the workflow process.

From manual entry to automated precision

For Koch, chat had already become the backbone of trader communication. But every transaction still required manual entry into spreadsheets and internal systems, slowing execution and leaving room for error. The new model automates that translation.

Working in a secure Microsoft Azure environment, Koch and ipushpull trained the LLM using historical data: real chat transcripts matched against confirmed trades. Over time, the model learned to interpret the shorthand that characterises energy-market dialogue, distinguishing routine chatter from confirmed transactions.

Accuracy climbed from the 70s to the high 80s after only 20–30 training examples and continues to improve through field-by-field performance tracking. Traders quickly discovered that adding explicit confirms in chat boosted precision toward near-perfect accuracy.

Measurable efficiency and control

Since deployment, the system has processed more than 40,000 trades directly from chat. Trade-entry times have fallen by 30–50 percent, while booking accuracy has reached the high 80s and continues to rise.

The benefits cascade across the workflow. Each trade is now linked to its originating chat message, creating a direct audit trail and eliminating hours of manual reconstruction for compliance and surveillance teams. The integration has also reduced downstream reconciliation work, cut back-office escalations, and delivered faster straight-through processing from execution to confirmation.

Human-in-the-loop

Automation at Koch is not about replacing human oversight; it is about amplifying it. Traders still validate and submit the order ticket generated by the agent, keeping control of the process.

Adoption spread organically. Once traders saw peers saving time and reducing errors, usage grew spontaneously across desks. This behaviour shift reinforced model performance: clearer phrasing produced higher accuracy, while the system’s growing precision encouraged further adoption.

Compliance, analytics and governance

Each trade is now auditable from chat to confirmation, with data stored and searchable in Koch’s systems. Field-by-field accuracy metrics are tracked to refine the model and target improvement areas.

For compliance and risk teams, chat-to-deal linking simplifies trade reconstruction and shortens surveillance timelines. The same analytics infrastructure provides transparency for model tuning and governance reporting.

As AI use expands, both firms are preparing for the next wave of requirements: stronger identity management, permissioning, and agent guardrails to ensure that automation remains secure and accountable in regulated environments.

Expanding across markets

The initial rollout focused on Koch’s physical natural gas operations, but the framework is built for scale. The same principles, LLM interpretation of domain-specific language, chat-based capture, and structured data integration, apply across other commodities and OTC markets.

Both companies view AI as an emerging utility. The differentiator now lies in orchestration: choosing the right models for each task and embedding them effectively into workflows.

The next development phases include agent-assisted workflows, richer analytics, and eventually agent-to-agent communication, where intelligent systems coordinate transactions and validations across organisations within permissioned environments.

A step toward the future of trading

The Koch–ipushpull partnership demonstrates how focused collaboration can deliver measurable, scalable innovation. By combining Koch’s trading expertise with ipushpull’s AI and workflow engineering, the firms have created a model that strengthens control, improves accuracy, and accelerates trading without disrupting the natural flow of communication.

In turning unstructured chat into structured data, Koch and ipushpull have shown what the future of AI-assisted trading looks like: precise, auditable, and built around the way markets already work.

Read more:
Automatically capture, interpret and book natural gas trades from chat apps to ETRMs, regulatory systems and downstream workflows.