Separate observation, inference and recommendation
5 min read
Lesson objectives
- Apply separate observation, inference and recommendation to a case with defined evidence.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
A report says, “Sales fell after we changed the call script, so the new script is unpersuasive and we must remove it.” Treat this as a chain, not one fact. “Sales fell” is an observation needing a defined measure, period, and source. “The script is unpersuasive” is an explanation that may or may not be right. “Remove it” is a recommendation that also depends on the costs of removal and the available alternatives. Separating these statements exposes the link for which evidence is missing. Sales might truly have fallen while stock ran out in some locations or fewer customers called. Replacing a script would not remedy a stock shortage.
For a consequential briefing, make four columns: statement, type, current support, and verification needed. For an observation, request the underlying record and measurement rule. For an inference, request the bridging assumption and a plausible rival. For a recommendation, request the decision criterion and effects of the options. A phrase such as “the numbers show” does not make an inaccurate number reliable. “Probably” may signal a reasonable but uncalibrated interpretation rather than a direct observation. Also distinguish evidence that a person made a complaint from evidence that the complaint's proposed cause is correct.
Suppose orders completed from qualified calls were 24% in one week and 18% in the next. That is a descriptive comparison only if “qualified call” and “completed order” meant the same thing in both periods and data-cleaning rules remained comparable. The difference is six percentage points; the relative decline is 6/24=25%, not “six percent.” Even a correct calculation cannot identify the cause. Check the number of qualified calls, customer mix, product availability, agent experience, the exact date of script adoption, and whether orders had enough time to complete. Each factor could change the numerator, denominator, or proposed persuasion mechanism. The questions arise from this causal story; they are not a generic checklist substituted for analysis.
Consider what the report calls evidence. System records can support a recalculation of orders but may contain miscoded calls. Five agents saying the script “feels cold” can suggest a mechanism, yet cannot measure its effect on conversion. Recorded calls might reveal that agents rarely used the script, changing the hypothesis from weak wording to inconsistent implementation. If feasible and fair, assigning comparable calls between two scripts and tracking completions and costs would provide a stronger comparison. A vivid unsuccessful call should not stand in for the distribution, and staff testimony need not be discarded: it may be valuable evidence about implementation and a source of hypotheses.
Turn the explanation into a testable prediction. If the script is a principal cause, we would expect a difference among customers with comparable products and stock conditions who actually encountered the different scripts. State what would count against this: similar declines in an unchanged channel would weaken the script-only account. A decline only where the script was used would strengthen it, but differences between channels would still need examination. This is a disciplined inference, not a promise of certainty.
Lesson summary
Separate observation, inference and recommendation: the lesson links the check to a worked case and practice.