Report writing for stakeholders

The structure that wins

Stakeholders read reports the way people read newspapers: the front page in thirty seconds, the section they care about in three minutes, the full piece almost never. A report that does not respect that reading pattern is read once and ignored.

The structure that survives this reading pattern, refined across enough Nepali boards and investor decks to be confident about it, has three parts.

One — the one-page top. What changed this month, why, and what it means for the business. Three short paragraphs, maximum. The first paragraph names the two or three movements that matter; the second explains what drove them; the third says what the team is doing as a result. If the director reads only this page, the director has the story. The temptation to make this section comprehensive — to list every campaign, every channel, every metric — is the temptation that produces reports nobody finishes.

Two — the evidence layer. Below the one-page top, the actual numbers and the actual campaigns. Channel-by-channel performance, the named campaigns and what each cost and produced, the cohort or segment movement that is worth flagging. This section is for the analyst on the team and the curious member of the board. It is not the headline; it is the support.

Three — the forward look. What is happening next month, what assumption it depends on, what would change the plan. This section is short and is the one the director re-reads in the next meeting when they want to remember what the team said they would do.

AI’s role is the structured prose in all three sections — turning the marketer’s notes into clean paragraphs, smoothing the transitions, holding the format consistent across months. The marketer’s role is the substance: which movements matter, what drove them, which campaigns to name, which forward bets to commit to.

The specific-evidence requirement

The single failure mode that ruins AI-drafted stakeholder reports is the absence of specifics. A model handed a vague brief — “write the monthly marketing report for our fintech” — produces a page of fluent generalities. Engagement is strong. Brand awareness is growing. The team is focused on optimisation. The director reads the first paragraph, recognises that it could have been written about any company in any quarter, and stops reading. The report has failed.

The fix is structural and unromantic. Before any prompt to the model, the marketer assembles a short evidence pack. Three to five specific numbers (with their source clearly noted). Two to four named campaigns and what each produced. One or two specific decisions the team made and what they cost or saved. One or two segments or cohorts that moved meaningfully. The model is then handed this pack and asked to arrange it into the three-part structure above.

The output is unrecognisable from the output of the vague brief. The first paragraph names actual numbers. The campaign evidence names actual campaigns. The forward look commits to actual actions. The director keeps reading because each sentence is doing work the previous report did not.

The stakeholder reading habit

Internalise how the reader actually reads, because the report has to be designed around it. A senior stakeholder in a Nepali firm — the director of a bank, the chairperson of an NGO board, the founder of a fintech, the head of a holding company — opens the report at the start of a meeting or on a phone between meetings. They have ninety seconds. They want to know two things: did anything bad happen, and is the team on top of whatever is happening.

The one-page top answers both questions if it is written for that reader. The bad-news disclosure is up front and unhedged. The “we are doing X about it” sentence is concrete. The forward look is a commitment, not an aspiration. This is the report the stakeholder remembers in the next conversation; everything below it is for the people who already trust the report and want depth.

The no-surprises rule

The oldest rule in stakeholder communication is that anything material in the monthly report should already have been raised with the stakeholder informally, in advance. AI does not change this rule. If anything, AI makes it more important, because the temptation to “just send the report” — to let the well-written document substitute for the awkward conversation — is now larger.

A real example, common enough to be worth naming. A campaign underperforms badly in week two of the month. The team spots it, knows it is a problem, and decides to “address it in the monthly report.” The model drafts a clean paragraph explaining what happened and what the team is doing about it. The report goes to the board on the thirty-first. The director reads it on the first, picks up the phone, and asks why they are hearing about this three weeks after it happened. The well-drafted paragraph has not protected the team; it has supplied the director with a clean record of how long the silence lasted.

The fix is unchanged by AI. Material developments — campaigns that miss badly, channel changes that shift the plan, regulatory or platform changes that affect the strategy — get raised informally as they happen. The monthly report confirms and contextualises; it does not break news.

The bilingual question

In several Nepali firms — particularly cooperatives, family-held businesses, and older institutions — the senior stakeholder reads Nepali first and English second, while the analyst layer below reads English first. A monthly report that is written only in English does some of its work and not all of it. A monthly report that is translated by a human after the fact takes long enough to defeat the purpose.

The pragmatic pattern, made affordable by AI: write the one-page top in both Nepali and English in the same pass, and keep the evidence layer in whichever language the analyst team works in. The model handles parallel generation cleanly — and, importantly, holds the same numbers in both versions because it is producing them from the same evidence pack. The marketer reads both versions before sending, with the same Nepali-checking discipline as before, because honorifics and financial terminology are where the Nepali draft is most likely to drift.

The win here is not a marginal improvement. For a firm whose chairperson reads Nepali first, the Nepali one-page top is the difference between a report that is engaged with and one that is politely acknowledged. AI is the thing that makes the bilingual version affordable to do every month rather than once a quarter.

Check your understanding

Quick check

—

True or false: an AI-drafted stakeholder report with no specific campaigns, no specific numbers, and no named decisions is at least a useful starting point that can be improved by editing.

1. ATrue — any draft is better than no draft.
2. BFalse — a draft without specific evidence is corporate-blue prose, and editing it does not add the evidence the report has to carry; the discipline is to assemble the evidence pack (named campaigns, real numbers, actual decisions) before the prompt, so the model arranges real substance rather than producing fluent generalities to be edited.
3. CTrue — the model will fill in the specifics in a second pass.
4. DFalse — but only because the formatting is usually wrong.

Check answer→

What comes next

Chapter Six steps back from the operational work — copy, campaigns, the Nepali market, SEO, analytics, stakeholder reporting — and into the questions that sit underneath all of it. When do you disclose that AI helped produce a piece of work? How do you keep your brand safe when the model occasionally gets things wrong in your name? What does the marketing function look like in three years, when the tools that today require a marketer’s prompt are doing more of the work directly? The chapter is shorter and more opinionated; it is the part of the course that ages fastest, and the part that matters most to the careers of the marketers reading it.