AI Energy-Trading Decision Boundaries Adviser

The mandate

The committee has not agreed where forecasting assistance ends and delegated trading begins. Teams are testing learned price, demand and congestion signals, portfolio recommendations and tool-using agents that can prepare orders, yet proposals describe accuracy and back-tested return without consistently exposing regime change, position limits, approval latency or the operator’s ability to interrupt an action.

The adviser will reserve three days each month for model-and-control evidence, a chair or committee session, and preparation or individual challenge with traders and risk leaders. Four committee meetings are included; a material incident or investment query receives acknowledgement within one business day and a written view within three, but the adviser never joins the trading approval chain.

The appointment runs for nine months through a board decision on the next automation stage. The committee chair may propose one renewal of up to three months if a named live-shadow test remains incomplete; only the full board may renew, and an unrealised performance target does not extend the term automatically.

This position has no line authority, dealing permission, risk delegation, board vote or executive responsibility for positions, limits, models, systems, disclosures or capital. Traders, risk officers and technology executives own those decisions; advice must identify its evidence and uncertainty and cannot be cited as authorisation for a market action.

Concurrent non-competing energy and quantitative appointments are permissible after disclosure. A role with a trading counterparty, material interest in a data or execution vendor, contingent return-linked fee, or access to another participant’s confidential strategy creates a conflict requiring recusal, restricted information or termination.

Why the board wants this voice

The board understands market risk and energy assets but lacks a member who has governed learned decision systems from forecast through executable order controls. Management debates have separated model performance from the operational authority the tooling gradually acquires. The chair wants an independent practitioner who can make autonomy increments explicit before successful pilots become de facto delegations.

What you will own

• Press management to classify each AI use as research, forecast, recommendation, order preparation or execution and name the accountable human at every boundary.
• Test model evidence across market regimes, sparse events, delayed fundamentals, revised data, transaction costs and decisions not taken.
• Challenge back-tests for leakage, survivor bias, unrealistic liquidity, execution assumptions and optimisation against the same period used for selection.
• Shape authority gates around position, product, horizon, confidence, loss, market condition, approval and immediate kill-switch behaviour.
• Probe model and agent observability for inputs, rationale, tool calls, order changes, overrides, rejected actions and post-trade attribution.
• Evaluate build, vendor and partnership choices for data rights, model change, audit access, latency, resilience and exit from proprietary dependencies.
• Frame the board’s automation decision with approved use cells, prohibited actions, residual uncertainty, dissent and evidence required before expansion.

Candidate qualifications

• Governed machine-learning forecasts or algorithmic decision systems in gas, power, commodities or another tightly controlled trading environment.
• Designed boundaries between research signals, trader recommendations and executable orders with tested limit, approval and kill controls.
• Identified leakage or unrealistic execution in a profitable back-test and can explain the subsequent investment or deployment decision.
• Presented model and market-risk evidence to a board committee without converting historical performance into assurance about future regimes.
• Managed vendor, exchange, broker and data-provider conflicts while protecting confidential strategy and independent challenge.
• Advised on progressive automation where accountable traders retained authority and realised outcomes were attributed after launch.

Non-negotiables

• Can attend all four Doha committee meetings and both designated trading-floor or governance reviews during the nine-month term.
• Will disclose trading interests, counterparty work, data and execution vendor economics, and any performance-linked remuneration.
• Accepts no dealing permission or operating role and will not represent advice as approval of a position or model.
• Brings production market-decision governance; generic energy strategy, forecasting research or enterprise AI experience is insufficient.

1. 49 words maximum. Describe one profitable trading model you constrained because its authority or back-test evidence was unsafe.
2. 49 words maximum. Which current counterparty, market, vendor or investment relationship must this board assess for conflict?
3. 49 words maximum. Confirm the Doha cadence and name the evidence required before a recommendation agent may prepare an executable order.