Each pool has a direct link to the P&L and carries both financial and quality metrics. That combination prevents the organisation from winning on speed while losing trust or compliance.

For large groups, apply the rule at each level. The group may hold three to five pools, and each division may also hold three to five pools. The logic does not change.

If you cannot answer these questions in writing, the pool is not ready. Pick another pool or close the gaps and come back.

In the era of AI you should add AI-specific targets that preserve Momentum. Examples include model cost per transaction, model latency at the ninety-fifth percentile, agent override rate by human reviewers, evaluation scores on golden datasets, and incident counts with time to recovery. If quality falls while cost or speed improves, you are not gaining Momentum. You are borrowing against the future.

Write it so that any leader can read it once and know what to do. Include rules about data usage and privacy, model error tolerance by use case, latency budgets, cost budgets, where a human must be in the loop, what gets logged, and what triggers incident escalation. In the era of AI you should also specify what decision trails must contain, which prompts and responses must be stored, where data may reside, and how often model drift must be checked. If boundaries are vague, teams will optimise locally and create hidden risks. If boundaries are clear, teams will innovate inside known guardrails and audits will move faster because controls are inside the workflow.

Every bet in the portfolio has a hurdle rate, stage gates, owners, and kill criteria. Capital moves based on evidence. If a bet cannot produce the evidence, it should stop. If it does produce the evidence, it should scale.

The Factory should offer connectors to core systems, prompt and agent libraries, tool-calling patterns, and evaluation harnesses with golden datasets. It should enforce version control for prompts and agents, provide a registry for approved components, and define where a human stays in the loop and why. The Factory should also run a showback rhythm where teams present shipped workflows and evidence, not slide decks. Compliance will move faster when controls and audit trails are part of the runtime from the first build. In the era of AI the Factory is the difference between isolated cleverness and compounding performance.

Hold the cadence steady. The organisation will learn what matters and will start to move in one direction with force.

Illustrative visual.

If a team cannot produce the pack, they are not ready for more money.

Controls belong in the workflow, not in a binder. Build data lineage, access policies, and retention rules into the data layer. Require model and agent approvals with evaluation against accuracy, robustness, and fairness. Enforce versioning so that you can rebuild a decision trail. Set cost and latency budgets in the runtime. Provide kill switches and incident playbooks with clear severity levels. Log prompts, inputs, outputs, and human overrides at the level required by your regulators and customers. If you do this from the start, audits will be faster because the story matches the logs.

Boards should insist on two things. First, leadership fluency so that the right pools and bets are chosen. Second, a system that makes those choices repeatable and safe. The board dashboard should show the value pools, the financial and quality targets, the current deltas, the risk posture, and any incidents. Oversight improves when evidence is visible and when controls are inside the workflow.