---
name: executive-summary-development
description: >
  Writes the executive summary of a research report: the conclusion first, the
  evidence behind it, confidence stated for every headline, and what could not
  be established. Use for "write the executive summary", "summarise the
  findings for the board", "the exec summary", "one page for the leadership
  team", "top line summary", "management summary", "they will only read the
  first page".
category: 11 Reporting and Storytelling
ref: 11.02
tier: 1
inherits: [K2, K3, K4, K5]
---

# Executive Summary Development

## 1. One-line description
Distils a checked research report into a summary that leads with its conclusions, states confidence for each, names what could not be established, and can be acted on safely by a reader who reads nothing else.

## 2. What this skill is used for

**The research problem it solves.** The executive summary is the most-read and least-checked part of any research deliverable. Most of the people who act on a study read only this. It is also, in most projects, the part written last, fastest, and by whoever is still available, usually by copying the first sentence of each chapter into a list. Two failures follow, and both are expensive because they happen at the point of maximum readership.

The first is **structural**: a summary that summarises the report rather than the argument. It has a paragraph per chapter, in chapter order, and it tells the reader what topics were covered rather than what was concluded. It is a table of contents in prose. The second is **evidential**: confidence inflation in transit. A finding that the analysis called an association becomes a driver, a moderate reading becomes a statement, a caveat present on page 34 is absent on page 2, and the summary asserts more than the report supports. Nobody does this deliberately. It happens because summarising is compression, compression removes qualification first, and qualification is what carries confidence. **This is the single place in a research deliverable where overconfidence does the most damage**, per K3 §5.2, because it is read by the people who spend money.

**Where it sits.** Reporting. Downstream of the narrative, the compiled report and its QA pass, because the summary is written from checked content rather than from memory of the analysis. Upstream of nothing inside the document, and upstream of everything outside it: the summary is what gets forwarded, pasted into a board pack, and quoted back six months later.

**Typical use cases.**
- A full research report needing its opening summary before delivery.
- A study whose decision-makers will not read the body, and whose researchers know it.
- A summary that will be extracted and circulated on its own, without the report attached.
- A report where the evidence is uneven and the summary must convey which conclusions are solid.
- A rewrite of a summary that has been criticised as a list of topics.
- A tracking wave where the summary must say what changed without overstating movement inside noise.

**Who uses it.** Research directors and senior researchers who sign the deliverable; insight leads writing for an executive committee; consultants whose summary is the only part a client principal will read; anyone whose report will be quoted without its qualifications attached.

## 3. When to use it

- A substantial report exists, its claims have been checked against source, and it needs its summary.
- The audience includes people who will decide without reading the body, which is most audiences.
- The summary will travel independently of the report, in an email, a board pack or a slide.
- The evidence supports some conclusions strongly and others weakly, and the reader needs to be able to tell them apart.
- The report contains a finding that is politically difficult and the question of whether it reaches the summary is live.
- A previous summary has been read as promising more than the study delivered.
- The study answered some objectives and not others, and the summary must be honest about which.

## 4. When NOT to use it

- **The report is not finished or not checked.** A summary written from an unchecked report inherits every unverified claim and then strips its qualifications, which is the worst combination available. Write the summary last, from a report whose major claims have been QA'd, per **12.03** step 11 and **12.06 Research Report QA**. The tell is a summary written to brief the writing of the report, which inverts the process: the report then gets written to satisfy the summary.
- **The summary is the whole deliverable.** A board note, a one-page readout, a five-slide executive session or a decision memo with no report behind it is **12.05 Executive Research Reporting**, which handles an artefact that must stand alone by design. This skill produces the summary section of a report that exists.
- **The narrative has not been settled.** A summary is the argument in miniature, and there is no miniature of an argument nobody has made. Where the through line is still contested, run **11.01 Research Narrative Development** first. Attempting to settle the narrative inside the summary produces a document that argues with itself in 400 words.
- **The reader needs a reference document rather than a conclusion.** Where the audience is a technical team who will use the study as a data source, what they need is a key findings list or a data book, not an executive summary. These are different documents and Section 8 sets out the difference. Producing the wrong one is a common and quiet failure.
- **The evidence supports no conclusion at the level the summary operates at.** Where a study is genuinely inconclusive, a summary asserting a headline is a fabrication of certainty, per K4 §1. The correct summary states what was established, what was not, and what would settle it, and it is shorter. It is not padded to look substantial.
- **The request is to remove the caveats so the summary reads more decisively.** Say plainly that a caveat travels with its claim, per K3 §7, and offer the alternative: a shorter summary that only carries conclusions strong enough to be stated without qualification, with the weaker material clearly held back. That is decisive and honest, per K4 §9.
- **The conclusion is fixed in advance.** Where the summary is expected to say what has already been agreed in a meeting, this is not summarisation. Say so per K4 §4.2 before writing.
- **The question is one of prose craft rather than distillation.** How a sentence carries a number, how a quote is integrated, how uncertainty is worded without becoming unreadable, is **11.03 Research Report Writing**. This skill decides what the summary says and in what order. 11.03 governs how any given sentence in it is built.

## 5. Required inputs

**Required.** Without these the skill cannot run. If absent, ask. If no answer is available and work must proceed, state the assumption at the point where it bites, per K5 §5.

- **The completed report, with its major claims checked against source.** The summary is a distillation of checked content, not a first draft of the conclusions.
- **The evidence map**, per K2 §5. Every claim entering the summary must already have a row in it, because K2 §5 rule 2 requires exactly this: every claim in an executive summary appears in the map. A claim that has no row is not summarised, it is investigated.
- **The narrative statement or the report's spine**, from **11.01**. The summary is the spine plus the load-bearing evidence plus the consequence. Without it, the summary defaults to chapter order.
- **The confidence level of each headline conclusion**, assessed per K3 §3, with the caveat that qualifies it.
- **The decision the report informs and its owner.** The summary is written to a reader who will act. Without knowing the act, the summary cannot be ordered by consequence.
- **The objectives as agreed**, so the summary can state which were answered and which were not.

**Optional, and what each one adds.**

- **The recommendations with their traceability**, from **08.04**: allows the summary to close on action rather than on observation, which is what most executive readers are looking for. Without them, the summary ends at implication, which is legitimate but should be stated.
- **The list of what could not be established**, from the report: makes the summary's honesty section a distillation rather than a fresh construction, which is where things get missed.
- **The previous wave's summary**: shows what the audience already believes and lets this summary state change rather than restate stability.
- **Knowledge of how the summary will be circulated** (email body, board pack, intranet, appended to a strategy paper): determines length, whether it can rely on a chart, and how defensively each sentence must be written for travelling alone.
- **Stakeholder political context**: informs whether a difficult conclusion needs to be stated more plainly rather than less, and identifies where a K5 §2.1 materiality judgement belongs.

## 6. Questions to ask before starting

1. **Who reads only this, and what will they do after reading it?** Determines ordering, length and whether the summary must close on recommendations. *Default if unanswered:* assume the most senior named recipient reads only the summary, and write for them.
2. **How will it travel?** A summary pasted into an email body, quoted in a board paper or read on a phone has different constraints from one that opens a bound report. *Default:* write it to survive being extracted with no report attached.
3. **What is the length ceiling, and is it a real constraint or a habit?** Determines how many conclusions can be carried. *Default:* one page, or roughly 400 to 600 words, and fewer conclusions rather than shorter sentences.
4. **Which conclusions are strong enough to be stated declaratively, and which are not?** Determines the confidence language and prevents uniform hedging, per K3 §1. *Default:* assess each against K3 §3 before writing, not while writing.
5. **Is there a finding the client will not want in the summary?** Determines where the materiality judgement sits. *Default:* include it if it is material to the decision, and mark the judgement per K5 §2.1 rather than deciding it silently.
6. **Which objectives were not answered?** Determines the content of the "what we could not establish" element, which is required per K3 §5.2. *Default:* list them, briefly, in the summary and not only in the report.
7. **Does the client expect an executive summary, an abstract, or a key findings list?** Three different documents, frequently conflated. *Default:* produce the executive summary, and offer the key findings list as a separate short annex where the audience genuinely needs one.

## 7. Step-by-step methodology

**What an executive summary is.** The argument of the report, compressed, in the order a decision-maker needs it, with the confidence of each conclusion visible and the unanswered questions named. It is not shorter than the report because it says less about each thing. **It is shorter because it says fewer things.** That distinction governs every step below.

**Step 1. Fix the reader and the act.** Name the person or role who will read only this, and write down what they do next: approve a budget, choose between two options, brief a team, decide whether to commission more work. The summary is ordered by what changes that act, which is frequently not the same as what the study found most robustly. Write the act down, because it is the tiebreaker at step 4 and it is the reason a summary is not a proportional shrinking of the report. *Correct result:* one sentence naming the reader and the decision, agreed with whoever owns the client relationship.

**Step 2. Take the spine and state it in one sentence.** Retrieve the narrative spine from **11.01** or, where none exists, from the report's own argument, and write it as the summary's opening claim. It leads. Not the method, not the background, not what the study set out to do. A reader who stops after the first sentence should have the answer, and a reader who stops after the first paragraph should have the answer, its basis and its confidence. Where the study's answer is genuinely conditional, the opening sentence states the condition rather than deferring it: "the case for a fee change holds for new customers and not for the existing base" is a lead, while "the results were mixed" is a delay. *Correct result:* an opening sentence that is the study's answer, at the confidence level the evidence supports, with no run-up.

**Step 3. Select the headline conclusions, and cap the number.** Work from the report's prioritised findings, not from its chapters. Select on decision relevance first, then evidence strength, then novelty, applying the same criteria the report used so that the summary does not silently re-rank the study. **Then cap the list at three to five.** The cap is the method, not a guideline: a summary that covers everything summarises nothing, and the reader's ability to retain and act on a conclusion falls off a cliff past roughly five. What does not make the cap is not lost, it is in the report. The correct failure mode when the study is rich is a summary that leaves things out, and the incorrect one is a summary that mentions everything in a clause.

Order the survivors by consequence for the reader, which is frequently not the order of evidence strength and is almost never chapter order. Where a strong conclusion and a decision-critical one compete for the lead, the decision-critical one leads and the strong one supports it, provided the decision-critical one is itself adequately evidenced: relevance decides position, evidence decides admission, and confusing the two produces either a summary that leads on the study's most robust irrelevance or one that leads on its weakest interesting claim. Record which conclusions were dropped, because a stakeholder will ask why their question is not in the summary and "it is on page 22" is only a satisfactory answer if the reasoning exists. *Correct result:* three to five conclusions, ordered by consequence, each of which would change the reader's act, each traceable to a report section, with the dropped set recorded.

**Step 4. Verify every claim against the evidence map before it is written, not after.** This is the step that distinguishes a professional summary from a competent one. For each conclusion, locate its row in the evidence map per K2 §5, open the named source, and confirm the number, the base, the question reference, the direction and the wording. Then check the level: is what you are about to write a finding, an interpretation, an insight or an implication, per K2 §2? Summaries drift upward, stating as insight what the report established as finding, because the higher level is more interesting and takes fewer words. **A claim in the summary that has no row in the evidence map does not get written**, and its absence is a defect in the report, not in the summary, per K2 §5. Where the summary needs a claim that unifies two report claims, the unification is itself a new claim and needs its own row. *Correct result:* every prospective summary sentence carries a map reference before drafting begins.

**Step 5. Build each conclusion answer-first.** The structure of each block is: the conclusion, then the evidence, then the qualification, then the consequence. "Messaging is the dominant communication preference, chosen by nearly two thirds of respondents (Q4, total sample, n=1,204). The pattern holds in every age band and market tested. For the service model this means the phone channel is now a minority route rather than the default." The upgrade from data to meaning is the point of answer-first writing, and its limit is exact: **the framing may do interpretive work for the reader, never evidential work.** "62% prefer messaging" and "messaging is the dominant preference" are the same claim differently served. "Customers have abandoned phone for messaging" is a different claim, because abandonment is a change over time and preference is not behaviour, per K4 §3.3 and K3 §3.4. The test is to strip the sentence of its evidence and ask whether what remains is a fair statement of what was measured, on the base it was measured on. Note that a verbal expression such as "nearly two thirds" is permitted only with its exact figure and base on the same page, per K2 §7. *Correct result:* each block leads with meaning, and no block's framing claims anything the number does not.

**Step 6. State confidence for each headline conclusion, in words.** Required by K3 §5.2, and it is the requirement most often skipped, because confidence language reads as hedging to a writer trying to sound authoritative. Use the registers in K3 §4 and use them differentially: a high-confidence conclusion is stated declaratively with no hedge at all, and hedging it is a miscommunication rather than a modesty. A moderate conclusion carries its alternative explanation or its specific limitation in the same passage. A low-confidence reading appears in the summary only where it is decision-relevant, is labelled as a hypothesis, and names what would validate it. **Uniform hedging is the failure to avoid here**: if every conclusion carries the same qualifier, the qualifiers carry no information and the reader discounts all of them equally, per K3 §1. Check the wording against the analysis output the claim came from, never against the chapter that paraphrased it, because confidence inflates by repetition, per K3 §7. *Correct result:* differentiated language, in which a reader can rank the conclusions by firmness without being told to.

**Step 7. Include what could not be established.** Two or three lines, in the summary itself and not only in the body. It covers objectives the study did not answer, questions the evidence cannot settle, and any material contradiction left unresolved. This is not a confession, it is what stops the summary being read as answering questions it did not answer, and it is where the next brief comes from, per K3 §5.2. It is also the element most often cut for length, and cutting it is how a summary starts misleading without containing a false sentence. Where a whole objective went unanswered, name it: a reader who commissioned four questions and received three answers is entitled to know which one is missing without reading 60 pages to find out. *Correct result:* a short, specific paragraph naming the gaps rather than gesturing at limitations in general.

**Step 8. Close on the consequence.** End with what follows: the recommendations, at their confidence level, with a sign-off marker per K5 §3.1, or the implication where recommendations are out of scope. Every recommendation in the summary carries the findings it rests on, by reference, per K2 §5. Where the report contains ten recommendations, the summary carries the two or three that matter for the decision at step 1, and says the rest are in the report. A summary that ends on an observation leaves the reader to construct the consequence, which is the work they commissioned. *Correct result:* a close that names an action or an implication, at its confidence, owned by someone.

**Step 9. Cut to length, by removing conclusions rather than words.** Compression by word-trimming removes qualifications, bases and caveats first, because they are the least fluent parts of a sentence. That is the mechanism of confidence inflation and it operates without anyone intending it. Cut instead by dropping whole conclusions, from the bottom of the step 3 ranking, and keeping the survivors fully qualified. **A base size, a confidence marker and a caveat are not compressible.** Where the length ceiling cannot accommodate three fully qualified conclusions, the ceiling is wrong and should be challenged with that reasoning. *Correct result:* a summary at length, where nothing was shortened by removing its qualification.

**Step 10. Run the standalone test.** Read the summary alone, with the report closed, as someone who will act on it and read nothing else. Ask four things. Could this mislead a reader who reads nothing else, by what it says or by what it omits? Does any sentence claim more than the report supports, particularly causation, behaviour from stated preference, change over time, or reach beyond the sampled population? Would a reader know which conclusions are firm and which are provisional? If this paragraph were quoted in a board paper with no context, would it still be true? Then run the harder version: hand it to a colleague who was not on the project and ask them what they would do next. If their answer is not the decision at step 1, the summary is not doing its job. *Correct result:* a summary that a hostile reader could not use to accuse the study of promising more than it delivered.

## 8. Analytical framework

The summary structure, which is a compression of the K2 chain rather than a selection from it:

    Spine (the answer)
      → Conclusion 1..n  [each: claim → evidence → qualification → consequence]
          → Confidence, stated in words per conclusion
              → What could not be established
                  → Consequence: recommendations or implications, with sign-off

**Applying it.** Each conclusion block is written in full before the next is started, because a half-written block is where qualifications get postponed and then forgotten. The confidence statement is part of the block, not a footnote after it.

**The three documents that get confused.** This is worth stating explicitly because producing the wrong one is common and the error is invisible until someone complains that the summary "does not say anything".

| Document | What it is for | What it contains | Reader |
|---|---|---|---|
| **Abstract** | Deciding whether to read the full document | What was studied, how, and what was found, descriptively. No argument, no recommendations | Someone triaging documents, usually in an academic or technical context |
| **Key findings** | Reference and extraction | A list of findings, each with its base, ordered by topic or importance. No through line | An analyst or team who will use the study as a data source |
| **Executive summary** | Acting without reading further | The argument in miniature: answer, evidence, confidence, gaps, consequence | A decision-maker |

A key findings list presented as an executive summary is the commonest of the three errors, and its tell is that every item is true, no item is connected to any other, and the reader finishes without knowing what to do.

**The compression rule.** A summary is shorter because it carries fewer conclusions, not because each conclusion is stated more loosely. Everything else follows from this: the cap at step 3, the incompressible elements at step 9, and the reason a summary is written last.

## 9. Output format

**Length.** One page as the working default, roughly 400 to 600 words. Longer only where the audience genuinely reads longer summaries, and never longer as a substitute for selection.

**A. The answer**
One to three sentences. The spine, at its confidence level, with no run-up. No methodology, no background, no restatement of the brief.

**B. Conclusions, three to five**
Each as a short block: the conclusion stated answer-first, the evidence with its base and question reference, the qualification or alternative explanation where confidence is moderate, and the consequence for the reader. Ordered by relevance to the decision, not by chapter.

**C. Confidence**
Carried in the language of each block per K3 §4. For substantial deliverables, additionally as a table, per K3 §5.3:

| Conclusion | Confidence | Rests on | What would change it |
|---|---|---|---|

**D. What we could not establish**
Two to four lines. Objectives unanswered, questions the evidence cannot settle, contradictions unresolved.

**E. What follows**
Recommendations at their confidence with `RESEARCHER SIGN-OFF REQUIRED` per K5 §3.1, or the implications where recommendations sit outside scope. Each names the findings beneath it by reference.

**F. Method, in one line**
Sample, method, fieldwork dates, and any disclosure required by K4 §7 that materially affects how the conclusions should be read: a non-probability sample, a small base carrying a headline, an unusual fieldwork period. One line, at the foot, never at the head. It is there so that an extracted summary carries its provenance.

**When the evidence is thin.** The summary gets shorter and its shape changes. Section A states what the study can and cannot answer. Section B carries only the conclusions that survive on their own bases, which may be one. Section D grows and becomes the most substantial part, naming what was not established and what would establish it. Section E may legitimately contain a recommendation to investigate rather than to act. A three-sentence summary that is true outperforms a full page assembled from fragments, and the reader is better served by being told the study did not settle the question than by a page that implies it did. Per K4 §1, do not fill section B because the format has one.

## 10. Quality checks

Run before the summary is attached to anything. Sits on top of K4 §8.

1. Does the first sentence contain the answer, rather than the topic, the method or the background?
2. Does every claim in the summary appear in the evidence map, per K2 §5?
3. Has any claim gained confidence between the analysis output it came from and this summary, checked against the analysis rather than against the chapter?
4. Does every conclusion carry a confidence level in words, and are the levels differentiated rather than uniform?
5. Does every moderate-confidence conclusion carry its caveat or alternative explanation in the same passage, per K3 §4.2?
6. Is every strong conclusion stated without a hedge, so its strength is visible?
7. Does the summary carry between three and five conclusions, or state why it carries more?
8. Is every base size present where a proportion is reported, and is any small base flagged per K4 §7?
9. Does any answer-first framing claim causation, behaviour from stated preference, change over time, or population reach the study did not measure?
10. Does the summary state what could not be established, including any objective not answered?
11. Does every recommendation in the summary name the findings beneath it and carry a sign-off marker?
12. Is any material contradicting evidence present in the summary rather than only in the body, where it bears on a headline conclusion?
13. Was anything shortened by removing its qualification rather than by dropping the whole conclusion?
14. Read alone, could this mislead someone who reads nothing else, by assertion or by omission?
15. Would a colleague reading only this identify the decision it was written for?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **The table of contents in prose** | One paragraph per chapter, in chapter order, describing what each covers | Build from the prioritised conclusions at step 3, never from the chapter list |
| **Confidence inflation in transit** (the signature AI failure) | The analysis said "appears to", the chapter said "shows", the summary says "drives" | Step 4 and step 6. Check wording against the analysis output, not the chapter |
| **Compression by trimming** | Bases, caveats and confidence markers absent from the summary but present in the body | Step 9. Cut whole conclusions, never qualifications |
| **Uniform hedging** | Every conclusion carries "appears to" or "may"; the reader cannot rank them | Differentiate per K3 §4. A hedged strong finding is a miscommunication |
| **The covered-everything summary** | Eight to twelve conclusions, each a sentence, none retained | Cap at three to five. What did not make the cap is in the report |
| **The missing gap section** | No statement of what was not established; the summary reads as complete | Step 7. It is required per K3 §5.2 and it is the first thing cut for length |
| **The buried headline** | Two paragraphs of background and method before the finding | The answer is the first sentence. Method goes in one line at the foot |
| **The over-upgraded lead** | Answer-first phrasing that has quietly added causation, behaviour or trend | Strip the evidence from the sentence and check what it still asserts |
| **The orphan recommendation** | A recommendation in the summary with no finding named beneath it | Every recommendation names its findings, per K2 §5. No chain, no recommendation |
| **The key findings list mislabelled** | Every item true, nothing connected, reader finishes not knowing what to do | Section 8. Decide which of the three documents is wanted, and say which was produced |
| **Written first** | A summary drafted early that the report is then written to satisfy | Write last, from the checked report, per step 4 |
| **The politically edited summary** | The report contains a difficult finding; the summary does not | Materiality is a K5 §2.1 judgement, made by a named human and recorded, not made silently by omission |
| **Summary and body disagree** | A number, a base or a direction differs between the summary and its chapter | Verify against source rather than against the chapter, and re-check after any late edit to the body |

## 12. AI guardrails

Skill-specific only. Universal prohibitions are inherited from K4. Confidence registers follow K3 §4; traceability follows K2 §5.

1. **Never write a summary claim that has no row in the evidence map.** Per K2 §5, every claim in an executive summary appears in the map. A claim without one is not summarised; the absence is investigated and reported.
2. **Never let a claim change confidence level between the analysis and the summary.** Check the summary's wording against the analysis output the claim came from, never against the chapter that paraphrased it, per K3 §7.
3. **Never drop a caveat, a base size or a limitation in order to meet a length constraint.** These are incompressible. Drop a conclusion instead.
4. **Never hedge a high-confidence conclusion.** Undifferentiated hedging destroys the reader's ability to tell strong from weak, per K3 §1, and it is as much a calibration failure as overstatement.
5. **Never upgrade a finding into a summary conclusion that claims causation, behaviour from stated preference, change over time, or reach beyond the population sampled.**
6. **Never omit a material contradicting finding from the summary because it complicates the headline**, per K4 §4.1. Where inclusion is genuinely a materiality judgement, mark it per K5 §2.1 and let a named human decide.
7. **Never state an objective as answered where the study did not answer it.** Objectives without answers appear in the "what could not be established" element.
8. **Never generate a summary from the report's headings.** Headings are labels. Summarising labels produces a document that describes the report rather than the findings.
9. **Never introduce a conclusion in the summary that does not appear in the report.** A summary is a distillation. A new claim at this stage has bypassed every check the report went through.
10. **Never pad a thin summary to look substantial.** Where the evidence supports one conclusion, the summary carries one, per K4 §1.

## 13. Best-practice principles

1. **Assume this is the only thing that gets read, because for the people who act it usually is.** Every decision about the summary follows from taking that seriously rather than repeating it.
2. **A summary is shorter because it says fewer things, not because it says things more loosely.** This is the whole craft in one line.
3. **Write it last, from the checked report.** Written first, it becomes a brief the report is then written to satisfy, and the report loses its ability to surprise its own authors.
4. **Lead with the answer.** The reader's first question is what the study concluded. Background, method and objectives answer questions they have not asked yet.
5. **Confidence is content.** "We are confident about this, less so about that" is information a decision-maker needs and can act on. Removing it does not make the summary stronger, it makes it unusable for calibrating risk.
6. **The caveat travels with the claim.** A qualification that lives only in the body has been dropped, because the summary is where the claim will be read from.
7. **What you leave out is the summary.** Selection is the work. A summary that mentions everything has made no selection and delegated the job back to the reader.
8. **The gaps section is not a weakness, it is a defence and a sales document.** It prevents the study being read as answering what it did not, and it is where the next commission comes from.
9. **Test it standing alone, quoted and stripped of context**, because that is how it will be used. A sentence that is only true next to its chart will eventually appear without its chart.
10. **Answer-first is a service to the reader, not permission to conclude harder.** The upgrade is in the framing, never in the claim.
11. **A summary that ends on an observation has stopped one step short.** The reader wants to know what follows, and constructing it for them is what they paid for.
12. **If the summary and the body disagree, the summary is wrong.** It was written later, faster, and under compression, and that is where the error will be.

## 14. Worked example

*Fictional scenario, used for illustration only. The organisation, findings, figures and quotes below are invented for the purpose of demonstrating method.*

**INPUT.** A retail bank has completed a study into why a newly launched savings product is under-selling. The report is compiled and QA'd: six chapters, 31 findings, an evidence map, four recommendations. Inputs to the summary: the narrative statement (spine: the product is not failing on its features, it is failing at the point where an existing customer has to prove their identity again), a customer survey (n=2,140), 24 depth interviews, client-supplied application funnel data, and a list of two objectives the study did not answer. The decision: the retail director decides in six weeks whether to change the product's rate, change the application journey, or increase marketing spend.

**PROCESS.**

*Steps 1 to 2.* The reader is the retail director; the act is a choice between three investments. The spine becomes the opening: "The product is competitive and the application is not: three in five customers who begin an application abandon it, and abandonment concentrates at the identity verification step (client-supplied funnel data, June to August)." The client-supplied label is carried into the summary, per K2 §6, because the strongest evidence for the headline is not evidence the team produced.

*Step 3, the cap.* Eleven candidate conclusions are ranked. Four survive: the abandonment concentration; the finding that rate sensitivity is low in this segment (Q22, n=2,140), which removes one of the three investment options; the interview evidence that re-verification is read as distrust rather than as friction, which explains why simplification alone may not fix it; and a segment finding that customers who hold a current account with the bank abandon at a materially lower rate. Seven conclusions are dropped to the report, including two that the team liked.

*Step 4, and the correction.* Drafting begins on a fifth sentence, "marketing spend is not the constraint". Checked against the evidence map, no row supports it: the study measured awareness among applicants, not among non-applicants, so it cannot speak to whether marketing reaches people who never arrive. The sentence is removed, and the point moves to the gaps section, where it becomes materially more useful: the study cannot say whether the funnel is under-filled at the top.

*Step 6, and the judgement call.* The interview conclusion about distrust rests on 16 of 24 participants and is an interpretation rather than a finding, per K2 §2. The pressure is to state it declaratively because it is the most interesting thing in the study and it is what makes the recommendation coherent. It is written at moderate confidence with its alternative in the same passage: "The most likely reading is that re-verification reads as distrust rather than as inconvenience (16 of 24 interviews). The alternative, that the step is simply long, is not ruled out by this design, and the two imply different fixes."

*Step 7.* The gaps section names the two unanswered objectives, the marketing question uncovered at step 4, and the fact that the funnel data covers three months and cannot separate the launch period from steady state.

*Step 9.* The draft runs to 780 words against a 500-word ceiling. The segment conclusion about current account holders is cut whole rather than compressed, and moves to the report. No base, caveat or confidence marker is removed.

*Step 10.* The standalone test catches one sentence: "customers do not want to re-verify" reads as a behavioural claim and rests on stated attitudes. It becomes "customers describe re-verification as the point at which the process stops feeling like it is for them". A colleague reading the summary alone names the application journey as the investment, which is the decision at step 1.

**OUTPUT.** A 520-word summary: the answer in two sentences, three conclusions each with base, confidence and consequence, a confidence table, a four-line gaps section naming two unanswered objectives and one newly identified limit, two recommendations at moderate confidence with sign-off markers and their findings named, and a single method line at the foot carrying the sample, the fieldwork window and the client-supplied status of the funnel data.

## 15. Advanced usage

**Summaries that will be quoted in public.** Where the summary may be extracted into a press release, a public report or a regulatory submission, every sentence must survive without its neighbours. Write each conclusion so that it carries its own base and qualification inside the sentence rather than in the block around it, accept the loss of fluency, and route the whole thing through **13.02 Source and Citation Verification** and a sign-off per K5 §2.8. External publication raises the standard from "true in context" to "true alone".

**Multi-audience summaries.** Where a study serves an executive committee and an operational team, resist the single summary that averages both. Write one summary to the decision, and a separate short key findings annex for the operational audience, labelled as such per Section 8. Two documents with clear jobs outperform one that tries to be both and is the wrong length for each.

**Tracking waves.** The summary's job changes: it states what moved, what did not, and what that means, rather than restating the market. Stability is a legitimate headline where change was expected, and the discipline is to avoid narrating noise. Every wave-on-wave movement in the summary confirms that question wording, base definition and method match, per **05.05**, and untested movements are described as observed rather than as change, per K4 §3.1.

**When the summary must carry bad news.** Do not soften the claim. Soften nothing, and instead give the difficult conclusion the same structure as the others: claim, evidence, confidence, consequence. A difficult finding stated plainly with strong evidence is defensible in a room; the same finding hedged reads as uncertain and gets challenged on its evidence rather than debated on its implications. Where the materiality of including it is genuinely in question, that is K5 §2.1 and belongs to a named human, recorded.

**When the standard approach does not fit.** A study with two incompatible readings, per **11.01** step 9, gets a summary that says so in its opening sentence and presents both, with the conditions under which each would be right and what would adjudicate. This is harder to write and more useful than a summary that picks one, and it protects the reader from acting on a synthesis the evidence does not support.

## 16. Skill chain

**Recommended previous skills:**
- **12.03 Research Report Compilation.** Hands over the compiled, QA'd report with its evidence map, which is the input this skill distils. The summary is written after that QA pass, not before it.
- **11.01 Research Narrative Development.** Hands over the spine and the load-bearing findings, which become the summary's opening sentence and its conclusion set.
- **08.04 Recommendation Development.** Hands over recommendations with their traceability and confidence, so the summary can close on action.
- **12.06 Research Report QA.** Where an independent check has run, hands over a report whose claims have been verified by someone other than the author.

**Recommended next skills:**
- **12.02 Research Report Design.** Takes the finished summary and lays it out without altering claims, bases or confidence language.
- **11.05 Research Presentation Development.** Takes the summary's conclusion set as the opening of a live session, where the same content behaves differently.
- **12.05 Executive Research Reporting.** Takes the summary where a standalone executive artefact is also required, and builds it as its own deliverable rather than an extract.
- **14.01 Insight Activation and Socialisation.** Takes the summary as the thing that circulates, and works on whether it produces action.

**Runs well alongside:**
- **11.03 Research Report Writing**, for the sentence-level craft of integrating numbers and stating uncertainty readably.
- **13.03 AI Output Verification**, run against the summary specifically, because it is the highest-risk paragraph in any deliverable.
- **K2 §5**, which requires every summary claim to appear in the evidence map, and **K3 §5.2**, which requires confidence per headline conclusion and a statement of what could not be established.

---
A Yazi Supplied Skill and resource.