Turning a finding into a recommendation
A finding is a statement about what is true. A recommendation is a statement about what to do, and it requires things the analysis does not contain: an objective, a set of options, and a judgment about risk.
Presenting a finding as though it were a recommendation, or a recommendation as though it were a finding, is the commonest failure at the interface between analysis and decision.
What does the step actually require?
An objective, the feasible options, the expected outcome of each under the finding, and the risks. The finding narrows the options; it does not choose between them.
That is why an analyst who says "the data shows we should do X" is usually skipping the parts where judgment and non-analytical constraints live.
How should the two be separated in a document?
Explicitly, with the finding stated first and the recommendation clearly labelled as a judgment built on it. A reader can then disagree with the recommendation while accepting the finding, which is a productive disagreement.
Blending them makes the finding contestable on grounds that are really about the recommendation, and the analysis loses standing it should have kept.
What should a recommendation include?
The action, the expected effect with its uncertainty, the cost, the risk if the finding is wrong, and what would be monitored to detect that. Five elements, and the last is what turns a recommendation into a plan.
The monitoring element also protects the analyst. A recommendation with a stated tripwire is one that can be adjusted rather than defended.
How much should uncertainty change the recommendation?
Proportionally. A finding with a wide range supports a reversible, small-scale action; a narrow one supports a larger commitment. Recommending a large irreversible move on a wide interval is the mismatch to avoid.
That framing also gives a route forward when the analysis is inconclusive: recommend the test rather than the rollout.
What about recommendations against the analyst's own prior?
They are the most valuable ones and the hardest to make. An analysis that consistently confirms what the team already believed deserves scrutiny, because confirmation is what a poorly specified analysis produces by default.
What should the analyst not do?
Advocate. An analyst who becomes an advocate for a recommendation loses the standing that made the finding credible, and every subsequent analysis is read as motivated.
Presenting the strongest version of the case against is a better use of the role and it is what a decision-maker actually needs.
How should disagreement be handled?
By locating it. Disagreement about the finding is an evidence question and can be resolved with more analysis. Disagreement about the recommendation is a judgment question and usually cannot.
Naming which one is in play shortens most of these conversations considerably.
Frequently asked questions
Should every analysis end with a recommendation?
No. Some findings are informational and forcing a recommendation onto them manufactures a decision nobody needed. Saying "this is what we found and it does not change the plan" is a legitimate ending.
Who owns the recommendation?
Ideally the person accountable for the decision, developed with the analyst. Analyst-owned recommendations tend to be either too cautious or too detached from constraints the analyst cannot see.
How do I recommend under high uncertainty?
Recommend the smallest reversible step that would reduce the uncertainty, with the decision criterion stated. That is nearly always available and it is more useful than a confident guess.
What if leadership wants a single number?
Give the point estimate with the range in the same sentence. Dropping the range to satisfy a preference transfers the uncertainty to the decision silently, which is worse for everybody including the person who asked.
Should I present options or one recommendation?
One recommendation with the main alternative and why it was not chosen. A menu of equal options pushes the analytical work back onto the decision-maker.
How do I avoid being ignored?
By tying the recommendation to a decision that is already on the table and to a timeframe in which it can be acted on. Excellent analysis delivered after the decision is made is a filing exercise.
What if the recommendation is rejected?
Record it with the reasoning, and monitor what you said you would monitor. A rejected recommendation that turns out to have been right is only useful if there is a record of it.
Does this apply to routine reporting?
Routine reports should carry an interpretation even when they do not carry a recommendation. A report with no interpretation at all delegates the thinking to whoever reads it, which is usually nobody.
What is the single best structural improvement?
Separate sections, labelled finding and recommendation. It is a formatting change and it changes how both are read and challenged.
What should be recorded after the decision?
The recommendation, the decision taken, and the monitoring agreed. That record is what allows a programme to learn, and its absence is why the same arguments recur. Industry context on commercial decision practice is collected by FMI.
### What should a brand actually do with this?
Write the finding and the recommendation as two separate paragraphs, always, even in an email. The separation costs nothing and it preserves the independence of the finding when the recommendation is disputed.
Then state what would change the recommendation. A recommendation with a stated tripwire invites a decision about risk rather than a debate about confidence, which is a considerably more productive conversation.
Working with retail data? See NRS Insights data solutions or contact NRS Insights.