Communicating a result to someone who will not read your notebook

You have the answer. Now it has to survive the journey into someone else's head — someone with four minutes, three other priorities, and no context on your methodology.

This is a skill, it is learnable, and it is what separates analysts whose work gets used from analysts whose work gets thanked.

Lead with the answer#

The instinct is to walk through the journey: here was the question, here is the data, here is what I tried, here is what I found. That is the order you discovered it in and the worst possible order to communicate it in.

Invert it:

Churn is concentrated in accounts under 20 seats, which are 62% of accounts and 11% of revenue. We should stop trying to save them and focus retention on the 50+ seat band.

Supporting detail follows.

Decision first. The reader who stops after one sentence has still got the point. The reader who wants more keeps going.

One chart, not six#

An analysis usually produces many charts. A communication should contain one, maybe two. Choose the one that carries the argument and link the rest.

Six charts do not make an argument six times stronger. They make the reader work out which one matters, and readers who have to work that out stop reading.

Title the finding, not the axes#

• ✗ "Churn rate by seat band, 2024–2026"
• ✓ "Churn is three times higher below 20 seats"

The axes already say what is plotted. The title is your one chance at a sentence that survives being screenshotted into a Slack thread with no context — which is what will happen to it.

Annotate directly#

Put the explanation on the chart, next to the thing it explains. A short label pointing at the spike — "pricing change, 14 March" — is worth three paragraphs below, because it is read at the moment the reader is looking at the spike.

Charts get separated from their captions. They rarely get separated from their annotations.

Say what would change your mind#

Volunteering the limitation buys enormous credibility, and it protects the finding from being over-applied:

This is observational — the 50+ seat accounts differ in ways beyond seat count, so some of this gap is selection rather than a seat-count effect. A held-out retention test on the small band would settle it.

Two sentences. They will not be read as weakness; they will be read as competence. Analysts who never state limits get discovered eventually and lose the trust all at once.

Match the register to the audience#

Executive. One sentence, one number, one recommendation, one chart. The methodology goes in an appendix nobody will open, and that is fine — it exists so that the one person who asks can be answered.

Product or operational owner. The finding, the segment breakdown, and the specific action. They want to know what to do on Monday.

Another analyst. Everything: the code, the assumptions, the splits, the things you tried that did not work. This audience is doing peer review and needs to be able to break your work.

Writing one document for all three serves none of them.

Numbers a human can hold#

• Round. "About 4.2 million", not "4,183,927.44". Precision you do not have is a lie about certainty, and precision the reader cannot use is noise.
• Two or three significant figures, matching how much could change the decision.
• Compare. A number alone is not information. "Up 8% on last quarter" or "twice the industry benchmark" makes it usable.
• Choose the frame deliberately. "Churn rose from 2% to 3%" and "churn rose 50%" and "one extra account in a hundred left" are the same fact framed three ways with three different emotional weights. Pick the one that represents the magnitude honestly, and be aware you are picking.
• Absolute alongside relative. A 200% increase from 1 to 3 needs both numbers to be understood.

The write-up template#

FINDING     One sentence. The answer and the recommendation.
SO WHAT     What should change, and who owns it.
EVIDENCE    One chart, titled with the finding. Two or three supporting numbers.
CAVEAT      The load-bearing limitation, in plain language.
DETAIL      Everything else, linked rather than pasted.

Five blocks, most of them one line. It fits in a message, it fits on a slide, and it survives being forwarded.

Presenting it live#

Say the finding in the first fifteen seconds. Do not build to it — you are not telling a story, you are giving someone information they need.

Then expect the two questions that always come: "is this real?" (have your sanity checks and interval ready) and "what do we do?" (have a recommendation, even a tentative one; "I do not know" from the person closest to the data is a wasted meeting).

And when you are wrong — you will be — correct it fast, in the same channel, at the same volume as the original. A visible correction costs a little credibility once. A quiet one costs all of it later.