Practical guide
Judgment and synthesis: make a conditional recommendation
Compare competing explanations and options in a fictional support-backlog decision without hiding uncertainty.
Judgment and synthesis means combining evidence, constraints and tradeoffs into a recommendation whose reasoning another person can inspect. This original support-backlog exercise asks the reader to distinguish facts from forecasts, compare options consistently and state what would change the recommendation. The aim is not to manufacture certainty or pretend that an ambiguous workplace decision has one universally correct answer.
Distinguish the situation from its explanation
A measured problem and a proposed cause are different kinds of claim.
The fictional support team has a growing backlog, two absent specialists and a recent product incident. Those facts do not yet establish which factor explains most of the delay. The ticket mix may have changed as well. Start by listing observations separately from causal explanations and proposed remedies.
Ask AI to organize competing explanations, but require each to point to supplied evidence or an explicit gap. Do not let a coherent narrative turn a plausible cause into a finding. A recommendation built on the wrong explanation may solve a visible symptom while leaving the underlying constraint untouched.
Compare options using the same criteria
Use a small, explicit comparison that keeps tradeoffs visible rather than hiding them in an unexplained score.
The exercise offers overtime, temporary help and a narrower service commitment. Compare each against time to take effect, specialist capacity, customer impact, reversibility and evidence needed. These are suggested criteria for this case, not a universal decision formula. The decision owner may need different criteria in a different situation.
Keep unknown entries unknown. If temporary help cannot handle specialist tickets without training, say so. Do not fill a blank with an attractive estimate simply to complete the table. Where the options address different parts of the backlog, explain that difference before ranking them.
Challenge the preferred option
Look for evidence that would make the leading recommendation weaker, not only evidence that makes it sound convincing.
Suppose the initial draft recommends overtime because it can start quickly. Ask whether the delayed tickets require unavailable specialists or whether the product incident is still generating new demand. Either condition could limit the usefulness of extra generalist hours. The challenge is specific to the proposed mechanism, not a generic request to list disadvantages.
Invite AI to draft a counterargument using only the supplied pack, then inspect its factual premises. A persuasive objection with invented ticket data is not useful scrutiny. Keep the valid objection, remove unsupported detail and identify the missing evidence needed to resolve the disagreement.
Show sensitivity to a consequential assumption
Explain which uncertainty could change the choice and how the recommendation depends on it.
In this case, incident duration may matter more than a small difference in hourly cost. Compare a short-lived surge with a continuing specialist bottleneck. These are hypothetical scenarios, not forecasts. Describe how the preferred response changes between them and which observation would tell the owner which situation is developing.
Do not attach precise probabilities unless the exercise supplies a defensible basis. A conditional statement can be more useful: proceed with a reversible short-term measure while confirming specialist demand, then revisit the plan at a named checkpoint. The condition should be observable enough to trigger a real review.
Deliver reasoning the owner can inspect
A recommendation should connect the chosen action to evidence, alternatives and a clear revision condition.
Finish with the proposed action, why it fits the current evidence, the strongest alternative and the facts that would change your view. Identify the next decision and its owner. Keep the source map accessible so the reader can challenge a material premise without reconstructing the entire analysis.
A reviewer can examine whether the worker acknowledged uncertainty, compared options consistently and updated the recommendation after new ticket information. Different defensible recommendations may satisfy those requirements. This exercise supports a task-specific discussion of judgment; it does not justify a fixed label about the person's general decision-making ability.
Sources and scope
NIST addresses AI risk management. Skills England describes workplace AI foundations.
These sources provide background, not endorsement of this exercise. The worked example and suggested review method are original illustrative guidance. They are not customer results, validated benchmarks or evidence of a particular product capability. Adapt the exercise to the task and use qualified review where consequences require it.
Sources: [1] [2]
Sources
- 1.AI RMF Core · NIST
- 2.AI foundation skills for work benchmark · Skills England
Explore this topic
- How to develop and assess causal reasoning in AI-enabled work
Causal reasoning is the ability to judge whether an action contributed to an outcome and what other explanation could account for the same observation. It requires more than noticing that two events occurred together. In AI-enabled work, the skill is visible when a worker states the causal claim, describes the mechanism, chooses a comparison that fits the question, tests rival explanations, and.
- How to develop and assess decision criteria in AI-enabled work
Decision criteria are the explicit tests used to compare options for a particular decision. Good criteria separate non-negotiable conditions from preferences, translate desired outcomes into observable tests, and show how trade-offs will be handled. In AI-enabled work, the skill is not asking a model to invent a scorecard. It is building a defensible comparison that a decision owner can inspect,.
- How to develop and assess evidence weighting in AI-enabled work
Evidence weighting is the ability to decide how much influence each piece or body of evidence should have on a conclusion. It requires more than collecting sources or labelling them credible. The worker must connect evidence to a specific claim, examine quality and applicability, reconcile agreement or conflict, and show how uncertainty changes the recommendation. AI can organize this work, but.
- How to develop and assess exception handling in AI-enabled work
Exception handling is the ability to recognize when an AI-assisted workflow has left its normal operating path, contain the immediate risk, route the case to the right owner, and preserve enough evidence for a safe decision. It is not the same as fixing every unusual output personally. Strong exception handling distinguishes harmless variation from a condition that requires a pause, escalation,.
- How to develop and assess human ownership in AI-enabled work
Human ownership is the ability to keep consequential decisions attached to an accountable person or role when AI contributes analysis, drafting, or recommendations. Ownership is more than placing a human in the loop. The owner needs defined authority, usable evidence, time to challenge the output, and responsibility for the final action. A nominal approver who cannot inspect or change the work.
- How to develop and assess recommendation formation in AI-enabled work
Recommendation formation is the ability to turn evidence, objectives, alternatives, uncertainty, and implementation constraints into a proposed course of action for a named decision owner. A strong recommendation is neither a summary nor a model-generated preference. It explains why this action is proportionate now, what was rejected, what remains uncertain, and what should trigger review or.
- How to develop and assess scenario reasoning in AI-enabled work
Scenario reasoning is the ability to explore how a decision or plan might perform across several plausible futures without pretending to predict which future will occur. The worker identifies consequential uncertainties, builds internally coherent conditions, traces implications, and finds actions that are robust, adaptable, or deliberately contingent. AI can broaden possibilities, but people.
- How to develop and assess trade-off analysis in AI-enabled work
Trade-off analysis is the ability to show what improves, what worsens, who is affected, and which constraints become binding when one option is chosen over another. In AI-enabled work, it means resisting a model's polished single answer and preserving the tensions a decision owner must actually resolve. The output is a transparent comparison, not a disguised recommendation.