Your starting point
Prerequisites: Use the ideas and vocabulary from lesson 53; revisit its worked examples if needed.
Definition: A recommendation proposes what to do, not merely what the data shows. It combines evidence with goals, constraints, values and uncertainty. A defensible recommendation makes those additional judgments visible rather than presenting them as automatic outputs of a metric.
Distinguish evidence from a decision rule
A lower cost ratio may support considering a campaign, but the decision also depends on quality, capacity and acceptable risk. State the objective and constraints. Different stakeholders can reasonably choose different actions from the same descriptive evidence when their goals differ. Do not imply that the spreadsheet has resolved value judgments that were never encoded or discussed.
Choose a proportionate action
When evidence is limited, a small reversible pilot may be more defensible than a permanent large change. Define what would count as success and what unintended consequence should be monitored. Identify conditions that would cause you to stop or revise the action. A recommendation without a verification plan can turn a tentative interpretation into an unchecked policy.
Communicate alternatives and ownership
Present at least one reasonable alternative and explain the trade-off. Name who would decide, implement and review the result in a real setting. Keep sensitive or consequential judgments out of automated approval flows. The academy provides practice in analytical reasoning, not authority to make decisions about real people or organizations from synthetic examples.
Two worked examples
A bounded campaign pilot
A synthetic comparison shows a lower recorded cost per conversion for A. A proportionate recommendation is to test a small additional allocation under a consistent audience and outcome definition, while monitoring conversion quality. “Move the entire budget to A immediately” assumes future stability and causal transfer that the table does not establish.
A service improvement
A support histogram shows a long tail of response times. A recommendation might be to inspect the ten longest valid cases and pilot a handoff change, measuring total resolution time and rework. Automatically penalizing the slowest agent would ignore case complexity and the incomplete causal explanation.
Where this becomes useful
An executive report is useful when the reader can see not only what action is proposed, but why it is proportionate to the evidence and how the organization would learn whether it worked.
Make an attempt
Use the complete workbench with Customer support or Marketing campaigns. Write one recommendation, a reasonable alternative, a success criterion and a stop-or-revise condition. Link each part to a verified finding or explicitly stated judgment.
Misconception check
Claim: A recommendation is just the highest or lowest number in a table.
Inspect the correction
Action requires goals, constraints, alternatives and uncertainty to be considered alongside the metric.
Independent application
Prepare a short recommendation memo from a synthetic analysis. Include a verified finding, objective, proposed action, alternative, limitation, success criterion and review point. Distinguish observed facts from judgments about acceptable trade-offs.
Completion records practice and self-review. It is not certification of statistical competence.
