The O&S philosophy
The work is not to add AI to every task. The work is to redesign the workflow so evidence, judgment, and accountability move together.
Human-owned
Humans own judgment, accountability, trade-offs, stakeholder context, and final decisions.
- Strategic choices with meaningful consequences.
- Ambiguous business calls where context is incomplete.
- Ethical, legal, people, or brand-sensitive decisions.
- Final recommendations and actions.
Deterministic tools
Use non-AI tools where exactness, repeatability, and auditability matter more than interpretation.
- Calculations, reconciliations, joins, and transformations.
- Version control, data validation, and source retrieval.
- Scheduled checks and rule-based alerts.
- Anything that should produce the same answer every time.
AI-supported
Use AI where the task involves synthesis, framing, summarisation, pattern-finding, or language.
- Turning messy source material into candidate narratives.
- Finding assumptions, contradictions, and missing evidence.
- Drafting options, scenarios, and stakeholder explanations.
- Translating technical analysis into commercial language.
Do not automate blindly
Avoid AI automation where evidence is weak, verification is hard, stakes are high, or the workflow would hide accountability.
- High-impact decisions without human review.
- Sensitive personal or commercial data without controls.
- Workflows where a confident wrong answer is worse than no answer.
Decision Workflow Stack
The five layers
Human / Tool / AI allocation is the centre, but practical systems need five layers around it.
1. Evidence
Source data, metric definitions, commercial context, reliability, caveats, and source-of-truth rules.
2. Allocation
What stays human-owned, what belongs in deterministic tools, what AI supports, and what may be automated safely.
3. Trace
The question, reasoning, assumptions, confidence, recommendation, decision owner, action taken, and outcome.
4. Governance
Review, accountability, reversibility, escalation, data sensitivity, and auditability.
5. Learning
Corrections, overrides, outcomes, and reusable context that improve the next run.
Quality bar
- Decision: Does this improve a real business decision?
- Evidence: Are sources, assumptions, and caveats visible?
- Allocation: Is it clear what humans, deterministic tools, and AI each own?
- Trust: Is there review, accountability, and reversibility?
- Learning: Does the workflow get smarter from corrections or outcomes?