Most analysis dies in the presentation. The work was correct, the room nodded, and nothing changed. That outcome is predictable and preventable, and the fixes are mostly about structure rather than statistics.
The pre-work that determines the outcome
Agree the question and the threshold before you analyse. Go to the decision-maker first and ask: "what would you need to see to act?" Get an answer, write it down, and analyse against it. Findings that arrive against a pre-agreed threshold get decided; findings that arrive as a surprise get "interesting, let's discuss further" - which means no.
Find out who actually decides, and what they care about. A finance director and an operations director need the same finding framed differently - one in cash, one in capacity.
The structure that works
Open with the recommendation. Then:
- What we found - one sentence
- Why it matters - in money, time or risk
- How confident we are - a range and a plain-English confidence statement
- What we recommend - specific, owned, with a date
- What we would need to be more certain - if relevant
Detail and method go in an appendix. If you begin with methodology you will lose the room before the recommendation, and the decision will default to whatever was already planned.
Chart discipline
- One message per chart. If you cannot write the message as the chart's title, redesign the chart.
- Title the chart with the finding, not the variables. "Conversion has fallen in the north since April" beats "Conversion by region over time".
- Show the comparison. A single series with no benchmark, target or prior period is decoration.
- Show uncertainty where it matters - error bars, or a shaded range on a forecast.
- Start the axis at zero for bar charts. Truncated axes are how honest analysts accidentally mislead.
- Cut every chart that does not support the recommendation. They dilute attention and invite tangents.
Language that gets action
Replace "there appears to be a possible correlation which may suggest" with "sales in the north are down 11% since April; the most likely cause is the depot change, and we recommend testing that by ___". Hedged language reads as low confidence and gives everyone permission to defer.
Be genuinely uncertain about uncertain things - but express the uncertainty as a range, not as vagueness in the verbs.
Volunteer your limitations early
Counter-intuitively, stating limitations up front increases the chance of action. It removes the objection before someone raises it as a reason to wait, and it builds the credibility that makes your next finding land faster.
After the meeting
Send one paragraph within an hour: the finding, the decision made, the owner and the date. Most "decisions" evaporate because nobody wrote them down. Then track whether the action happened, and follow up with the outcome - that loop is what turns you from a reporting function into an influential one.
The measure of success
Not whether the analysis was correct. Whether anyone did anything differently. If your analyses never change decisions, that is the problem to solve - see also explaining concepts to non-technical stakeholders.
The Statistics for Business course at London School of Business UK includes communicating and defending findings as part of the assessed work. Enquire today.