Turn AI policy into decisions people can apply.

Separate actual AI capability from automation, product language, and marketing claims.

1. 1.1

   AI, automation, and rules-based software

2. 1.2

   Prediction, generation, classification, and recommendation

3. 1.3

   Where AI already appears in company work

4. 1.4

   Capability, limitation, and risk

Understand why AI output can be useful, variable, plausible, and wrong at the same time.

1. 2.1

   Inputs, instructions, and context

2. 2.2

   Pattern recognition is not human understanding

3. 2.3

   Probability, variability, and confidence

4. 2.4

   Why the same request can produce a different answer

5. 2.5

   Five questions before trusting a result

Connect data choices and evaluation methods to product quality and affected people.

1. 3.1

   How data shapes system behaviour

2. 3.2

   Training data and learned patterns

3. 3.3

   Quality, coverage, and bias

4. 3.4

   Validation, test sets, and benchmarks

5. 3.5

   Evaluation tied to real work

Choose the simplest technical approach that fits the decision, workflow, and consequence.

1. 4.1

   Models, capabilities, and boundaries

2. 4.2

   How large language models generate

3. 4.3

   Retrieval and grounding

4. 4.4

   Agents, tools, and actions

5. 4.5

   Selecting the right approach

Turn an AI idea into a testable product decision with clear value, boundaries, and ownership.

1. 5.1

   Define the problem before selecting a model

2. 5.2

   User value and failure consequences

3. 5.3

   Cross-functional feature alignment

4. 5.4

   Build, buy, or do without AI

5. 5.5

   Pilot with success and stop criteria

Match evidence and human review to the importance and reversibility of the decision.

1. 6.1

   Verification that fits the consequence

2. 6.2

   Hallucination, plausibility, and false confidence

3. 6.3

   Meaningful human oversight

4. 6.4

   Working with uncertainty

5. 6.5

   Escalation and stopping conditions

Recognise foreseeable harm early and select controls that address the actual failure mode.

1. 7.1

   Privacy and confidential information

2. 7.2

   Security, prompt injection, and unintended disclosure

3. 7.3

   Bias, fairness, and affected people

4. 7.4

   Intellectual property and transparency

5. 7.5

   Controls, logging, and monitoring

Connect legal and policy responsibilities to the way the organisation actually uses AI.

1. 8.1

   Provider, deployer, user, and affected person

2. 8.2

   AI literacy in the EU AI Act

3. 8.3

   Policies, approvals, and accountable owners

4. 8.4

   Evidence, records, and continuous updates

Apply one repeatable decision structure to a realistic cross-functional AI proposal.

1. 9.1

   Define the decision and affected people

2. 9.2

   Test value, data, capability, and evidence

3. 9.3

   Set safeguards, ownership, and escalation

4. 9.4

   Cross-functional scenario and action plan