Skip to content
Tutorials Center
My learning

Topics

Lesson 33 of 76 ยท Visualization and Storytelling

Communicating Results to Non-Experts

Turn analysis into decisions: know your audience, write the one-sentence finding, structure a report, choose tables or charts, and state caveats honestly.

An analysis that nobody understands changes nothing. The best model in the world is worth little if the person who must decide cannot tell what it found, how sure you are, and what they should do. Communication is not a decoration added at the end of a project. It is half of the job, and the half that decides whether the work has any impact.

This lesson is about writing and presenting for people who are smart and busy but not statisticians: managers, product owners, clinicians, teachers, customers. You will learn how to start from the audience, state the finding in one sentence, structure a short report, choose between a table and a chart, and write caveats that are honest without burying the message. Two live examples help with the mechanical parts: formatting a summary table, and generating a finding sentence from numbers so the text always matches the data. For the visual side, see Principles of Effective Visualization.

Start with the audience and the decision

Before you write a word, answer three questions:

  1. Who will read this? A technical peer, an executive who has two minutes, a team that will build on your work?
  2. What decision or action should follow? Launch or stop? Spend more here or there? Change a process? If there is no decision, ask why you are producing the report.
  3. What do they already know, and what do they care about? A finance lead cares about money and risk. An engineer cares about what changes in the code. A doctor cares about patients.

Most communication failures are not about unclear writing. They answer a question nobody asked, or go into detail the reader does not need.

flowchart LR
    A["Audience<br/>and decision"] --> B["One-sentence<br/>finding"]
    B --> C["Evidence<br/>(1 to 3 visuals)"]
    C --> D["Caveats<br/>and confidence"]
    D --> E["Recommendation<br/>and next step"]

Work from the reader's decision backwards: finding first, evidence second, caveats and action to close.

The one-sentence finding

If the reader remembers only one thing, what should it be? Write that sentence before anything else. A good finding sentence has a subject, a direction and size, a comparison and, where it matters, a sense of certainty.

  • Weak: "We analyzed churn by plan type and tenure."
  • Better: "Customers on the monthly plan cancel at 14%, almost three times the 5% of annual-plan customers."
  • Best (with certainty and action): "Monthly-plan customers cancel at 14% versus 5% for annual plans (difference 9 points, 95% interval 6 to 12). Moving even a fifth of them to annual plans would cut churn noticeably."

Notice the habits: numbers are rounded to what matters, units are clear (percent, points), and the sentence says what to compare against. If you cannot write the sentence, you do not yet know your finding; go back to the analysis.

When findings are generated from data, build the sentence in code so it cannot go stale. The example below turns a result into text with sensible rounding, and handles the case where the interval includes zero, so the wording changes honestly.

Python Generate the finding sentence from the numbers

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The second sentence refuses to claim a win: the interval includes zero, so the wording says so. That is the kind of honesty that builds trust with your readers over time.

Structure of a short report

For most audiences, use the inverted pyramid: the most important information first, details later, so a reader who stops early still has the main point. A one-to-three page report works well:

  1. Title: the finding as a statement, not a topic.
  2. Executive summary (3 to 5 sentences or bullets): the question, the answer, the confidence, the recommended action. Anyone who reads only this should be able to act.
  3. Key evidence: one to three visuals or tables, each with a title that states its point.
  4. Method in brief: what data (source, period, size), what was done, in a short paragraph. Details go in an appendix.
  5. Caveats and limitations: what could make the answer wrong.
  6. Recommendation and next steps: what to do, who, by when, and how you will know it worked.
  7. Appendix: tables, code location, definitions, extra charts.

Here is a template for the executive summary, which is the part most people will read. It is deliberately short:

TEXT
Question:   Why did subscriptions drop in Q3?
Answer:     About 60% of the drop comes from new customers in two regions;
            existing customers are stable.
Confidence: Medium. Based on 3 months of data; one region has few customers.
Action:     Review onboarding in regions X and Y; re-measure in 6 weeks.

And a skeleton for the whole document:

TEXT
# <Finding as a title>
## Summary         (5 lines max)
## What we found   (1-3 charts, one message each)
## How we know     (data, method, in plain words)
## What could be wrong (caveats)
## What we recommend (actions, owners, dates)
## Appendix

Tables or charts?

Both are tools; choose by what the reader will do with the information.

Use a table when Use a chart when
The reader needs exact values The reader needs a pattern, trend or comparison
There are few rows and several different measures There are many points or categories
Units differ across columns The same quantity is compared across groups or time
The reader will look things up The reader will "get it" at a glance

A small, well-formatted table is often better than a chart for an executive summary ("here are the five numbers"). A table with 40 rows and 12 columns is neither: pick the three columns that matter. Good table habits: right-align numbers, use consistent decimals, show units in the header, order rows by importance or value rather than alphabetically, and highlight the row the text talks about.

Python A summary table builder with readable formatting

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The function takes raw numbers and returns a table meant for people: sorted by the thing that matters, rounded, with units in the headers. Putting formatting in one function means every report in the project looks the same.

Show less, explain more

Non-experts benefit from restraint:

  • One message per visual. If a chart needs three sentences of explanation, it carries three messages. Split it.
  • Avoid jargon, or define it once. Say "the chance this result is just luck" with the p-value in a footnote, not the other way around. Say "typical" and explain "median" in brackets the first time.
  • Use comparisons to make numbers meaningful. "A 2-point increase" means little; "about 400 extra customers per month, roughly the size of our smallest region" lands.
  • Round deliberately. Report "about 14%", not 13.8724%. False precision suggests accuracy you do not have.
  • Do not hide the bad news. If the model failed, or the data were too thin, say so up front. A reader who finds it later will distrust everything else.

Caveats without drowning the message

Every analysis has limits, and listing them is part of being truthful. The skill is to be specific and proportionate:

  • Say what could change the answer, not generic disclaimers: "Only customers who answered the survey are included, and unhappy customers answer less, so satisfaction is probably overestimated."
  • Separate correlation from causation in the wording: "customers who use feature X renew more often" is not "feature X makes customers renew". See Causal Thinking and Confounding.
  • State the uncertainty with an interval or a range, and the sample size when it is small.
  • Say what you did not check (a segment, a time period, an assumption).
  • Put caveats near the claim they affect, and keep the main caveats in the summary, not only in an appendix.

Three confidence levels are often enough for a summary: "high" (large, clean data, consistent results), "medium" (reasonable data, some open questions) and "low" (small sample or messy source, treat as a lead, not a conclusion). Explain briefly what pushes the rating up or down.

Python Does the chart carry its own message?

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

This chart follows the advice: the title is the finding, the color separates before from after, values are printed, the axis starts at zero, and a footnote names a confounder (the pricing page) instead of overstating the campaign's effect. The printed change matches the title's claim; whenever a headline contains a number, compute it in code rather than from memory.

Presenting live

For a talk or meeting, the same ideas apply with a few additions: put one idea on each slide, state the conclusion in the slide title, rehearse the first and last minute, and prepare backup slides for the questions you expect. Anticipate the question "how sure are you?" and the question "what should we do?" and have a clear answer to both. Silence after your recommendation is fine; do not fill it with more charts.

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. What should come first in a report for a busy decision maker?

    The inverted pyramid puts the main point first. Details go later and in an appendix, so a reader who stops early still has what they need.
  2. 2. Which finding sentence is the strongest?

    It names the groups, gives sizes, and states the comparison. The others only say that something was examined or related.
  3. 3. When is a table better than a chart?

    Tables suit lookup and exact values for a handful of rows. Charts suit patterns, trends and many points.
  4. 4. A confidence interval for a lift runs from -0.4 to +1.4 points. How should you word the result?

    An interval that includes zero cannot rule out no effect. Honest wording reports the estimate and its uncertainty rather than claiming a win.
  5. 5. Which is the best way to handle limitations?

    Specific, proportionate caveats help readers judge how far to trust the result. Generic or hidden caveats do not.

Summary

  • Start from the audience and the decision; the report exists to support an action.
  • Write the one-sentence finding first: subject, direction and size, comparison, certainty.
  • Use an inverted-pyramid structure: summary, evidence, method, caveats, recommendation, appendix.
  • Choose tables for exact values and charts for patterns; format both for people (rounding, units, sorting).
  • Give each visual one message and put the finding in its title; compute numbers in text from the data.
  • State specific caveats and a confidence level, and be open about bad news and uncertainty.

Next: Descriptive Statistics in Practice, the numbers behind the story.