AI Strategy and Value

Review questions

Subject 1 Integrated Summary: From Value Strategy to a Decision Roadmap

AI strategy starts with an outcome the organization wants to improve, not a tool it wants to buy. This subject therefore distinguishes a tool, value, and organizational capability. A tool is a specific product or model; value is the desired change in an outcome or decision; capability is the repeatable ability to turn a need into responsible use, then measure, improve, or stop it. Capability brings together data, roles, process, technology, controls, and measurement. A “no AI for now” decision can therefore be correct when process or data improvement addresses the problem more directly or when readiness conditions are not met.

Reference frameworks help organize the decision but do not make it for the organization. The NIST AI RMF Core uses Govern, Map, Measure, and Manage, with governance working across the lifecycle. ISO/IEC 42001 presents an organization-level AI management-system perspective; ISO/IEC 38507 adds a governing-body and management perspective; and the OECD AI Principles connect innovation with human-centred values and beneficial outcomes. In this diploma these sources widen the strategy and governance lens. Applying general principles is not treated as equivalent to global legal compliance.

Once the problem is framed without a preselected solution, we build a value hypothesis. A testable hypothesis states the problem, baseline, intervention, target, value mechanism, guardrail, and continue-or-stop rule. In the teaching example there were 1,000 requests per month, 30% requiring rework, and 12 minutes per affected request. The calculation is one thousand multiplied by thirty percent, multiplied by twelve minutes, then divided by sixty, giving 60 rework hours per month. After improving the form, the rate fell to 20%, giving 40 hours, a difference of 20 hours. That difference is not automatically a cash saving, and it does not prove AI caused the change if the improvement happened before AI was introduced. Disciplined measurement separates the target, the observed result, and the causal claim.

When several use cases compete, strategy treats them as a portfolio. Strategic fit, value, feasibility, data readiness, and risk manageability can be compared, and teaching weights can make the trade-off visible. A weighted score is not a probability of success. A result such as 3.85 out of 5, or 77% of the theoretical maximum, describes assumed ratings and weights only. An average must also not hide a decisive condition; that is the role of gates. No right to use the data, or no credible handling path for a sensitive error, may stop a use case regardless of its score. Capacity and dependencies then shape sequencing: building a shared data component can be more strategic than starting a high-value standalone case if that component unlocks several later uses.

Finally, strategy becomes a decision roadmap. The roadmap connects the objective with the portfolio, capabilities, operating model, measures, and review points. An operating model is not simply a committee. It means knowing who permits an experiment, who decides data is fit, who accepts residual risk within their authority, who approves production use, and who can stop the system. The roadmap needs leading capability indicators, such as the share of experiments that start with a baseline, guardrail, and decision owner, as well as lagging value indicators such as cycle time with rework rate. If data, assumptions, or reference frameworks change, the review point is used to re-examine the value hypothesis and portfolio sequence. Scaling is not an automatic reward after an experiment; it is a new decision based on evidence and conditions.

Review questions

Test your understanding with these questions. Marking and saved results are available with a free account.

Question 1Which step represents the more mature start to an AI strategy? Buy the highest-rated assistant and then look for uses. Define a business outcome and baseline, then compare AI with alternatives and identify required capabilities and controls. Set a target to use AI in 60% of processes. Postpone every experiment until data is perfectly error-free.
Question 2Which description best captures governance as a cross-cutting function in the taught NIST AI RMF logic? A one-time legal review after the system is built. Connect responsibilities, decision rights, risk, and measurement from the start through use and change. Choose the most technically accurate AI model. Publish general principles and let teams interpret them without defined decision rights.
Question 3What is the best step before buying an AI classifier? Buy the classifier because 28% is high. Fix categories and the knowledge base, measure again, then define the remaining value AI might add. Train the model on current data so it learns the inconsistency. Replace employees with full automation to eliminate judgment differences.
Question 4Which metric is closer to a value outcome than an adoption metric? Number of AI licences purchased. Percentage of employees who opened the assistant once. Reduced request rerouting while classification error stays within the agreed limit. Number of use cases listed in an idea register.
Question 5Which problem statement is the most solution-neutral? We need a predictive model because competitors use one. We need AI to improve efficiency. The team cannot detect delay signals early enough to change the plan, causing repeated rescheduling. Our data is imperfect, so we need a smarter model.
Question 6How many monthly rework hours does the baseline represent? 36 hours 50 hours 60 hours 3,600 hours
Question 7Which value hypothesis gives management a clearer decision rule? We will deploy AI because it is modern and strategically important. After stabilizing definitions, test reducing rework from 20% to 15% while critical-field error does not exceed baseline, and scale only if both conditions are met. We expect major savings and will measure them later. If users like the tool, we will adopt it.
Question 8What is the most disciplined conclusion from this sequence? AI caused rework to fall from 30% to 19%. The form change was first associated with an 8-point drop, then the AI stage with a further 3-point drop; causal attribution needs a stronger comparison. The form change caused all improvement because AI added only 3 points. Nothing can be learned from before/after measurement.
Question 9Why use a decision gate before relying on a weighted score? Because weighted scores are prohibited in strategic decisions. To prevent a high average from compensating for a decisive unacceptable condition, such as no right to use the data or no handling path for a sensitive error. To increase every use-case score before management review. To convert the score into an exact probability of success.
Question 10What is the weighted score for use case X? 3.25 out of 5 3.50 out of 5 3.75 out of 5 4.00 out of 5
Question 11Which sequence uses capacity and resolves R’s dependency logically? Cycle 1: R only; cycle 2: P+Q. Cycle 1: P+Q; cycle 2: R. Cycle 1: P+R; cycle 2: Q. Cycle 1: P+Q+R; cycle 2: nothing.
Question 12How should a 77% weighted portfolio score be interpreted in the teaching model? The use case has a 77% probability of success. The result reflects the assumed ratings and weights and still needs qualitative interpretation and gates before a decision. The use case must be implemented because it exceeded 70%. The result proves the use case is economically better than every alternative.
Question 13Which sequence reflects the integrated strategy roadmap taught in the lesson? Tool → budget → training → adoption campaign. Objective → portfolio → capabilities → operating model → measures and decision points. Committee → policy → vendor list → launch. Adoption metric → training → more use → annual report.
Question 14What is the most important operating-model gap? The council has no marketing name. Decision rights are undefined: who permits experiments, accepts risk, moves to production, and stops use. Not using one identical model for all use cases. The CEO does not review every system output.
Question 15Which metric is the better example of a leading capability indicator? Total annual savings after systems are in operation. Cycle time after six months of scaling. Share of experiments that began with a documented baseline, guardrail, and decision owner. Number of public articles published about AI.
Question 16What is the better strategic decision at the review point? Scale according to the original plan because commitment to the plan matters more than changing it. Re-examine the value and data hypothesis, update the baseline, then resequence the roadmap before deciding to scale. Cancel all AI initiatives because the data changed once. Keep the old baseline so comparison remains easy.