Four decision rules for AI business strategy. Use these prompts to explain why an option fits the scenario and why the alternatives fall short.
Domain 1 · 24%
Choose the right tool
AI Fundamentals and Literacy
- Rules: Use deterministic automation when the logic is known and the same inputs must produce the same outcome.
- Predictive ML vs. generative AI: Predict a label, score, or forecast with ML; use generative AI when the task calls for new text or other content.
- Prompting, retrieval, fine-tuning: Clarify instructions and examples first. Retrieve current, authorized facts for knowledge gaps. Consider fine-tuning when a repeatable behavior still needs improvement and you can evaluate it.
- Agents: Grant only the tools and permissions required. Put approval before high-impact actions.
- Data and monitoring: Check completeness, recency, and representation; compare live performance with a baseline.
Ask: Is this a rules, prediction, knowledge, or behavior problem?
Domain 2 · 28%
Prove the value
AI Strategy and Business Value Creation
- Start with the outcome: Name the workflow, owner, baseline, target, and measurement window before choosing technology.
- Prioritize: Compare impact, feasibility, data readiness, risk, and fit with the business strategy. Use a simpler alternative when it meets the need.
- Build, buy, or partner: Compare differentiation, time to value, internal capability, control, and the cost of switching later.
- ROI: (Attributable benefits − total costs) ÷ total costs. Include recurring usage, integration, human review, and monitoring. Count saved hours once; book cash savings only when spending actually changes.
- Evidence: Use adoption and quality as early indicators; verify lasting outcomes such as cost per resolved case, retention, or customer satisfaction.
Ask: What improves, compared with what, and at what full cost?
Domain 3 · 24%
Govern by consequence
AI Governance and Responsible AI Leadership
- Classify risk: Consider who could be harmed, how severe and reversible the outcome is, and how much autonomy the system has.
- Assign accountability: Name an owner with authority to approve, pause, or escalate. Reassess when the use case, audience, or autonomy changes.
- Test fairness and reliability: Look beyond averages to subgroup outcomes. Investigate bias in objectives, data, decisions, and feedback loops.
- Use layered controls: Least-privilege access, trusted sources, evidence checks, input/output safeguards, and human review address different failure modes.
- Protect data and rights: Check permitted use, privacy and IP obligations, auditability, vendor terms, and routes for human review or recourse.
Ask: What could go wrong, who owns it, and what happens before harm?
Domain 4 · 24%
Earn the right to scale
Business Readiness, Leadership, and AI Transformation
- Assess readiness: Check leadership alignment, culture, data ownership and access, infrastructure, and governance. Remove the blocker before funding scale.
- Run a bounded pilot: Choose a representative workflow, baseline, success and stop criteria, and a feedback route.
- Prepare people: Explain role changes, train by role, empower champions, and provide a safe way to report errors.
- Prepare operations: Name the owner, fund support and monitoring, test production load, and rehearse fallback and rollback.
- Scale in stages: Reuse standards and patterns while testing local conditions, support capacity, and continuing value.
Ask: What evidence says this will keep working outside the pilot?