Chapter 3 · Lesson 2
Compare Alternatives With Criteria
Use multiple drafts to explore trade-offs without letting AI make the final judgment for you.
The first idea looks reasonable, but you cannot tell whether it is actually the strongest direction.
Generate meaningfully different options
Ask for alternatives based on different strategies, not cosmetic rewrites. A campaign could prioritise speed, trust, or distinctiveness. A workflow could optimise for control, cost, or automation. Named strategies make trade-offs visible.
Choose criteria before choosing a winner
Define what matters before seeing the options, such as evidence quality, effort, reversibility, risk, and expected impact. Otherwise a fluent explanation can quietly change the standard to favour whichever answer it just produced.
Use AI as an analyst, not the owner
AI can organise comparisons and reveal assumptions, but it does not know your unspoken priorities or carry responsibility for the decision. Inspect its scoring, challenge missing criteria, and make the final choice yourself.
Build the mental model
Comparison is useful when options differ strategically and the criteria exist before the sales pitch.
Create option diversity deliberately
Assign each option a different optimisation target such as speed, control, or learning. Three cosmetic variants do not reveal meaningful trade-offs.
Use evidence and uncertainty in scoring
A score should cite its basis and confidence. Treat missing evidence as missing, not as an average score that quietly rewards an unsupported option.
A team is choosing how to produce a recurring customer insight report.
Give me three approaches and recommend the best one.
Create three approaches optimised respectively for speed, analyst control, and scale. Compare them against setup effort (20%), evidence traceability (35%), correction cost (25%), and monthly operating effort (20%). Cite the assumption behind every score, mark missing evidence, and show how the ranking changes if traceability receives 50%. Do not choose for us.
The options differ by strategy, weights are explicit, scores require evidence, and sensitivity analysis shows whether the recommendation is robust.
Avoid false precision
Weighted matrices organise judgment; they do not turn uncertain assumptions into facts. Use ranges or confidence labels when exact scores would be invented.
Run a sensitivity check on the two most important criteria. If a small weight change reverses the result, the decision is fragile and deserves discussion.
Build a decision table for a real choice
Define three meaningfully different options, four criteria, evidence for each score, and one sensitivity test.
Options optimise for: [...]. Criteria/weights: [...]. Evidence required: [...]. Missing data: [...]. Sensitivity test: [...].multiple choice
What should happen before asking AI to recommend one of three options?
Choose the answer that best fits the principle from this lesson.
Compare strategies against criteria chosen before the recommendation.