Case Study 1 — One Memo, Three Passes

A composite, fictional-but-realistic scenario built from the kind of internal memo a working data scientist writes constantly. Dana Whitfield and the specifics are invented; the document and the passes are exactly the shape this chapter teaches.

The situation

Dana Whitfield, a data scientist at a mid-size SaaS company, has spent a week investigating why paid-plan churn ticked up in Q2. Her director, Marisol, decides Thursday whether to greenlight a retention project — and she'll read Dana's memo on her phone, between meetings, in about ninety seconds. Dana has a solid first draft. It is not yet a good memo. Here is what she wrote, hot, on Wednesday afternoon.

DRAFT (level 0) — sent to herself, not yet to Marisol:

"Subject: Q2 churn analysis

Over the past week I conducted an analysis of the churn that we
observed in Q2 among our paid-plan customers. The methodology involved
pulling cohort data from the warehouse, segmenting by plan tier and
signup channel, and cross-referencing against the support-ticket logs
in order to identify any correlations that might exist. It is important
to note that the data quality for the EMEA region was somewhat
incomplete for April, which is something we should probably fix at some
point. After conducting the analysis, what I found is that there are
several factors which appear to be associated with the elevated churn.
The most significant of these is that customers who did not complete
the onboarding checklist within the first seven days churned at a rate
that was substantially higher than those who did. I also looked into
seasonality but the effect was negligible. Based on all of this, I think
there is a reasonably strong case to be made that we should invest in
improving the onboarding completion rate, possibly through a guided
setup flow, as a way of addressing the churn problem going forward."

It's all there — the finding is real and the recommendation is sound. But Marisol, scanning on her phone, would hit 150 words before learning what Dana wants. Dana lets it go cold overnight and revises it Thursday morning, top-down.

Pass 1 — Content (is the substance right, complete, relevant?)

Dana reads only for what's here and what's missing, touching no sentences.

  • Cut the EMEA aside. "Data quality for the EMEA region was somewhat incomplete… we should probably fix at some point" is a real issue, but it belongs in a data-eng ticket, not a churn recommendation to a director deciding on a project. It dilutes the message. → Cut.
  • Cut the seasonality dead-end. "I also looked into seasonality but the effect was negligible." A null result Marisol doesn't need to act on. → Cut (or demote to a one-line footnote at most).
  • Flag a missing number. The memo says onboarding non-completers "churned at a rate that was substantially higher" — but never says how much higher, and never says what the churn costs. For a funding decision, the "so what?" is underbuilt (Ch 3). → Flag: add the actual rate gap and a revenue figure.

Pass-1 output: Cut EMEA and seasonality. Add the churn-rate gap and the dollar impact.

Pass 2 — Structure (is it in the right order for the reader?)

Now the order, still no sentence work.

The recommendation — invest in onboarding completion — is the last sentence. That's the buried-conclusion failure (Ch 4): a scanning reader works through method and findings before learning the ask. Invert it. Lead with the recommendation and the headline number; let method and detail follow for the reader who wants them.

Pass-2 output: Recommendation + headline number first. Then the one finding that supports it. Method last, brief.

After Passes 1 and 2 — the global passes, revision proper — here is the reordered, de-tangented content. Note that it's deliberately still clunky; Dana has not touched the sentence level yet.

AFTER GLOBAL PASSES (rough on purpose):

"Recommendation: invest in improving onboarding completion, probably
through a guided setup flow. The most significant thing that I found is
that customers who did not complete the onboarding checklist within the
first seven days churned at a rate that was substantially higher. Over
the past week I conducted an analysis of the churn that we observed in
Q2 among our paid-plan customers, pulling cohort data and segmenting by
plan tier and signup channel."

It reads a little awkwardly and the method is jammed against the finding. That's fine — global revision often makes a draft temporarily uglier before the local passes clean it up. Dana resists polishing here; she's still deciding what and in what order, not how it reads.

Pass 3 — Sentences and words (now make it read well)

With content and structure committed, Dana descends to the local rungs and fixes the breakage she'd been ignoring.

  • "The most significant thing that I found is that customers who…" — an expletive-style throat-clear hiding the real subject. Make the customers the subject: "Customers who skipped the onboarding checklist…"
  • "churned at a rate that was substantially higher" — vague and now backed by the number she flagged in Pass 1. Replace with the figure: "churned at nearly triple the rate of those who did."
  • Bloat sweep (Ch 3). "Over the past week I conducted an analysis of the churn that we observed" → "Last week I analyzed Q2 paid-plan churn." "In order to" and "it is important to note that" — gone wherever they survived.
  • Proofread (level 6), cold and slow. Is it "Q2" consistently? Is the number formatted the same way throughout? Any doubled words after all the moving? She reads it aloud once; her mouth doesn't stumble.

Here is the memo Marisol actually receives:

FINAL:

"Subject: Q2 churn — recommend investing in onboarding completion

Recommendation: build a guided onboarding flow to lift checklist
completion. Customers who skip the onboarding checklist in their first
seven days churn at nearly triple the rate of those who finish it — and
non-completion alone accounts for an estimated $180K in Q2 paid-plan
churn.

Last week I analyzed Q2 paid-plan churn: cohort data from the warehouse,
segmented by plan tier and signup channel, cross-referenced with support
tickets. Onboarding completion was by far the strongest signal;
seasonality was negligible. Happy to walk through the cohorts."

What changed, and why the order made it possible

Set the bookends side by side. The draft was ~155 words that made Marisol hunt for the point and never told her what the problem cost. The final leads with the decision, quantifies the stakes ($180K, triple the rate), and reads in one scan.

Crucially, it got there because the structural move — lead with the recommendation — happened before the word-level polish. Picture the alternative: had Dana started at the bottom on Wednesday, de-bloating the EMEA and seasonality sentences, she'd have polished two sentences the content pass deletes — straight into the recycling bin. Top-down didn't just produce a better memo; it spent her ninety minutes only on words that survived.

The takeaways

  1. Cut before you polish. The EMEA and seasonality sentences were the first things to go in Pass 1 — so Dana never wasted a minute making them pretty. Content decides what survives; only survivors earn sentence work.
  2. The buried conclusion is the default failure. Dana's recommendation started in the last sentence, like almost everyone's first draft. Moving it to the front (Pass 2) was the single highest-leverage change in the whole revision.
  3. Let the draft be ugly between passes. The "rough on purpose" middle version looked worse than the draft for a moment. That temporary ugliness is the price of not polishing text you'll move or cut.
  4. The cold gap did the real work. Hot on Wednesday, Dana agreed with every sentence. Cold on Thursday, she saw the buried ask and the missing number in the first thirty seconds. Nothing about her skill changed overnight — her relationship to the draft did.

One sentence to remember: Fix what you say and in what order before you touch how it reads — or you'll spend your best attention polishing sentences you're about to delete.