# External Insights > Gathers, verifies, and synthesises outside-in intelligence on a company, person, industry, or trend, and delivers a sourced insight brief rather than a pile of links. Enforces the standard that every fact carries a named source and a date in the same line, that load-bearing numbers are two-sourced or visibly marked as single-sourced, and that what a company says about itself, what a third party reports, and what has been inferred never blur into one another. Use this skill whenever someone asks to research a company, prepare for a meeting with an external party, understand what is happening in an industry, monitor a competitor or a client, find signals before a pitch, build background for a leader before a call, or says "what is going on with them", "brief me on", "find out what they are doing", "what are people saying about", or "have a look at this company before Thursday". Trigger for any request to bring outside information into an internal decision, and always before outreach, pitches, and partner meetin - Skill: `ingridleiria/external-insights` (Agent Skill) - Install (CLI): `npx skillmds@latest add ingridleiria/external-insights` - Raw SKILL.md: https://api.skillmd.com/api/skills/ingridleiria/external-insights/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: ingridleiria (https://skillmd.com/u/ingridleiria) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/ingridleiria/external-insights --- # External Insights Two documents get produced when someone asks what is going on with a company. The first is a pile of links and a paragraph that could have been written from the company's own homepage: founded in this year, serves these industries, recently announced a partnership. It is accurate, it takes an afternoon, and it changes nothing, because nothing in it is news to the person who asked. The second is worse. It contains a fact that was true eighteen months ago: a leadership appointment that was reversed, a product that has since been retired, a funding round that was actually a debt facility. The executive repeats it in the room. A gap in a brief costs nothing, because the executive simply asks the question. A wrong fact costs the credibility of everything else in the document, and it is repeated with confidence precisely because it came from a brief. Insight is information that changes what someone will do in the next meeting. Everything else is background. This skill collects widely, verifies strictly, and delivers narrowly. ## When to use this, and when not to Use it before a meeting with an external party, before outreach or a pitch, when a competitor or client needs monitoring over time, when an industry shift needs to be understood well enough to act on, and when a leader asks what is happening with a named company. Do not use it to size a market or map a competitive set for a strategy decision, which is `market-research`. Do not use it to write the one page the executive actually carries into the room, which is `executive-briefing`; this skill supplies the external half of that page. Do not use it to produce the seller-facing comparison used in live deals, which is `competitive-battlecard`. Do not use it to assess a counterparty against criteria for a go or no-go, which is `partnership-assessment` or `vendor-evaluation`. Do not use it on your own internal data, which is `pipeline-deep-dive`. Do not use it to verify academic or scientific claims, which needs bibliographic verification rather than business sourcing. ## What you need before starting **Who reads it, and what meeting or decision it feeds, and when.** A brief for a leader walking into a partnership call at three o'clock is a different document from a monthly competitor monitor. Missing: ask in one line. If the answer is that there is no meeting and no decision, the honest response is that this is background reading and should be twenty minutes, not two days. **The entity, named precisely.** The legal entity, the trading brand, the group, and the specific subsidiary are four different things and filings sit under only one of them. Missing: establish it from a register or a filing before collecting anything, because research attached to the wrong entity is worse than none. **What the reader already knows.** The two or three facts the brief should not spend space on. Missing: assume they know the public basics and lead with what changed, which is the safer default in both directions. **The relationship history, where one exists.** Prior meetings, commitments made, open threads. This is internal, not external, and it comes from a customer system, an email archive, or the people involved. A connected relationship or calendar tool speeds this up where one is available. Missing: ask the two people most likely to have met them, and mark anything unconfirmed as unverified. **The deadline.** Missing: assume the meeting time minus two hours, so there is room to correct something. **Access to sources.** Public filings, company sites, archived versions of pages, job boards, trade press, and a search tool. A connected company-data service helps for firmographics, funding, and headcount trends where one is available; where none is, the same facts can usually be assembled from filings, careers pages, and archived snapshots at the cost of an hour. Missing: proceed on public sources and say in the brief which questions could not be answered. ## The method ### 1. Write the frame in one line Reader, meeting or decision, date. Then the single question the brief has to answer for that reader. Everything collected is judged against it, and this is what prevents the company profile that nobody needed. ### 2. Set the question set before collecting Five questions, always the same five, which is what makes briefs comparable over time: 1. What are they trying to achieve right now, and what evidence shows it? 2. What changed in the last ninety days that matters? 3. Where is the pressure: what problem are they under, visible in hiring, financials, product moves, or leadership statements? 4. What does this mean for our decision or our meeting? 5. What could the sources not tell us? ### 3. Collect into a working sheet, not into prose Four columns: fact, source and date, confidence, relevance to the decision. Confidence is verified, single-source, or inferred. Writing prose while collecting produces a document organised by the order you found things, which is never the order the reader needs. Cut ruthlessly at this stage. A fact with no relevance entry does not go into the brief, however interesting it was to find. ### 4. Look where other people do not The signals that produce actual insight are usually unglamorous and public. Job postings show what a company is building three to six months before it announces, and the team name in the posting is often more informative than the role. Archived versions of pricing and product pages show what was removed and when, which is the clearest available evidence of a strategy change. Statutory filings give real numbers where a jurisdiction requires them. Careers page headcount over time gives a growth trend without a paid dataset. Conference talks and leadership posts give priorities in the leader's own words, which is what you want if the meeting is with that leader. ### 5. Verify before anything enters the brief Every fact carries its source and date in the same line. Any number that carries the recommendation is confirmed by two independent sources, or is marked single-sourced in the body rather than in a footnote. Recency is checked explicitly. Leadership, pricing, headcount, and funding facts older than twelve months are flagged as possibly stale, because those four change most often and are the ones an executive is most likely to repeat. Keep three categories visibly apart: what the company says about itself, what a third party reports, and what you inferred. "Stated on their results call in March" and "likely, based on eleven engineering postings in one team since June" are different claims and should read differently. Nothing is fabricated. A gap is reported as a gap with the best available proxy named. ### 6. Synthesise to one implication Answer the five questions in prose, then compress to the single implication that changes what the reader will do. If the research produced no such implication, say that plainly. "Nothing in the last quarter changes our approach; here is what we checked" is a legitimate and useful finding, and pretending otherwise is how briefs lose their credibility. ### 7. Write it narrow Headline insight first, in one or two sentences. Then the facts that support it, five to eight, each with a source and a date. Then the questions for the meeting, which are the ones the research could not answer. Then, where relevant, the sensitive topic to avoid. Reference material and the full source list go at the end, where they belong. ### 8. Set the monitor, if this recurs Where the subject will be tracked over time, record the date, the sources checked, and the facts as at that date, so the next update can report only what changed. A monitor that re-researches from scratch each time produces the same brief repeatedly and nobody reads the third one. ## Source stack, in order of preference 1. **Company primary material**: filings, annual reports, results transcripts, regulatory submissions, press releases, product and pricing pages, job postings, documentation, and leadership posts and interviews. 2. **Connected data services** where available: firmographics, leadership changes, funding, technology signals. Never a requirement; every fact they provide has a public route that costs more time. 3. **Quality secondary**: major business and trade press, analyst notes, regulator statements, national statistics. 4. **Social and community**: posts by named leaders, conference talks, forums. Useful for tone and priorities, weak for facts, and labelled as such. 5. **Aggregators and search-optimised content**: last resort, never the sole source for anything that carries the brief. Where a claim rests only on tier four or five, the brief says so in the line where it appears. ## What not to collect Public professional information only. No personal contact details, no home addresses, no family or private-life information, no speculation about health, finances, or motives. Where research on an individual is needed, it covers their stated professional positions, their public record in the role, and what they have said in public, and it stops there. This is not only an ethical boundary, it is a practical one. A brief containing something the subject would be alarmed to see is a brief that cannot be forwarded, and briefs get forwarded. ## Worked example **Situation.** A 140-person managed technology services firm had a first meeting booked with the newly appointed group operations director at Ardmore Foods, a food manufacturer running nine sites and the firm's second-largest account at about 1.3 million a year, with the contract due for renewal in seven months. The previous sponsor, who had signed every renewal for six years, had left in March. The account director wanted a brief before the call. **Task.** One page, delivered two days before the meeting, answering what the new director would be measured on and whether the renewal was at risk. Good meant the account director learned something she did not already know and did not repeat anything that turned out to be wrong. **Action.** The first pass was the wrong document and was thrown away. It was a profile of Ardmore assembled from their website and a company data service: turnover, sites, brands, board, recent announcements. Ten facts, of which the account director could have recited eight, and the two she could not made no difference to the meeting. It had cost most of a day. The reframe was to the first and third questions in the standard set: what are they trying to achieve right now, and where is the pressure. That moved the search off the company and onto the person and the programme. Three things came back. First, the director's own public record. In a recorded conference talk the previous November she had described reducing a supplier list from around sixty to about twenty in eighteen months in her previous role, in her own words, dated, and available to anyone who looked. Second, Ardmore's half-year results statement named a two-year cost programme with third-party spend as one of three stated sources, filed in September. Third, four job postings in eight weeks under a Group Procurement team that had not appeared in postings a year earlier, two of them specifying technology category experience, which is the difference between an intention and a programme with staffing behind it. An archived version of Ardmore's supplier page from four months earlier also showed a preferred supplier tier that had not existed before, which supplied the mechanism. Two facts were left out for failing verification. A trade press piece said Ardmore was exploring a sale of its frozen division, which appeared nowhere in the company's own material and could not be confirmed. And a headcount figure from an aggregator sat about 30 percent above what the filings and the careers page implied, so the brief used the filed figure and marked the discrepancy. The synthesis was one implication: this was not an introductory meeting. It was the opening round of a supplier consolidation in which the firm was one of roughly forty technology suppliers, and the risk was not price but being categorised as a point supplier rather than a platform one before anyone had argued the case. **Result.** The brief was one page: headline, six facts with sources and dates, three questions, and one thing not to raise, which was a redundancy announcement at one of their sites two weeks earlier. The account director did not present the conclusion. She asked the second question on the list, which was what the preferred supplier tier would require of an incumbent. The answer confirmed the reading and the conversation moved to the criteria rather than to the renewal. The firm was invited into the consolidation review four months early and signed a three-year agreement at a lower annual value across a wider scope, which is a smaller number and the one that was still in place two years later. The two facts that did the work came from a recorded conference talk and four job postings. Neither is difficult to find and neither was in the first day's work, because the first day was spent on the company rather than on the question. ### A second scenario, where it goes differently The same company monitors a direct competitor quarterly. In the third quarter, nothing had changed: no pricing movement, no leadership change, hiring flat, no new filings, and no product announcements. The temptation is to fill the page. The correct output is four lines: nothing material changed this quarter, here are the eight sources checked and the dates, here is the one thing that would matter if it appeared, which is any posting for an enterprise sales role in our largest market, and the monitor moves from quarterly to twice yearly with a standing alert on that one signal. What changed is that the finding is negative, and a negative finding delivered honestly is what earns the reader's trust the next time the brief says something has moved. It also saves the two days that would otherwise be spent re-describing a company nobody's view has changed about. ## Output **Pre-meeting brief**, the default when a named meeting exists. One page. ``` BRIEF: [entity] FOR: [reader] MEETING: [what, when, format] PREPARED: [date] HEADLINE [One or two sentences. The insight that changes the approach.] WHAT WE KNOW [Five to eight facts. Each one line, each with source and date.] WHAT THIS MEANS FOR THE MEETING [One short paragraph. The implication, and how confident it is.] QUESTIONS TO ASK [Three, each one the research could not answer.] DO NOT RAISE [The sensitive topic, if there is one. Otherwise say none found.] WHAT WE COULD NOT ESTABLISH [The gaps, named. Sources at the end.] ``` **Working sheet**, kept during collection and attached where the reader may want to check: | Fact | Source and date | Confidence | Relevance to the decision | |---|---|---|---| Confidence is verified, single-source, or inferred. **Company snapshot**, where no meeting is attached: firmographics, leadership, strategy signals, recent moves with dates, and the implication paragraph. **Industry or trend read**: what is happening, who is moving, what the numbers show, where the disagreement is with a named source on each side, and the implication. **Monitor update**: only what changed since the last update, each item dated, plus the list of sources checked and the date checked, so a quarter with no change is a legitimate output. Every format leads with the headline. Where charts are used, keep them monochrome with series separated by marker and dash pattern rather than colour, and the legend outside the plot area, so the page survives being printed or forwarded as an image. ## Failure modes **The link pile.** Recognise it when the deliverable's structure is the order in which things were found. Fix by collecting into the working sheet and writing only from the relevance column. **The profile the reader already knows.** Recognise it when the facts are all obtainable from the subject's own front page. Fix by answering the ninety-day question first and building outward only as needed. **The stale fact repeated with confidence.** Recognise it by the absence of a date on any line about leadership, pricing, headcount, or funding. Fix by dating every line, which makes staleness visible without requiring anyone to remember. **Inference wearing the clothes of fact.** Recognise it when a sentence about strategy has no source. Fix by labelling the three categories differently and keeping the evidence attached: eleven postings in one team is evidence, "they are pivoting" is a conclusion, and the brief should show both. **Adjectives standing in for evidence.** Fast-growing, leading, innovative. Recognise them by deletion: if the sentence still says something, the adjective was decoration. Replace with a number or remove. **Over-collection on an individual.** Recognise it when the brief contains anything the subject would be surprised to see written down. Cut it, and cut it before anyone else reads the draft. **The implication buried on page two.** Recognise it when the reader has to reach the end to find out what it means. The headline is the deliverable; the facts exist to support it. ## Edge cases **The subject is small, private, and almost invisible.** Say so early rather than padding. Two or three verified facts and an honest statement that public information adds little is a usable brief, and it redirects the reader to asking the questions directly in the meeting. **Sources conflict.** Present both with their sources and dates, say which you trust and why, and never average them. A discrepancy between an aggregator and a filing is not a range, it is a signal that one source is wrong. **The subject is in a live legal or regulatory matter.** Report only what is on the public record, cite the document, and do not characterise it. Speculation about an unresolved matter is the single most damaging thing a brief can contain. **The research turns up something that changes the deal.** Take it to the person who owns the decision before it goes into a circulated document, and put it in writing the same day. A finding like this loses its value if it arrives after a commitment has been made, and it creates a problem if it arrives in a document with a wide distribution list. **The brief is about a person rather than a company.** Restrict to their public professional record: roles, tenure, stated positions, published work, public talks. Describe behaviour and stated positions, never character. Anything you would not say to them directly does not belong in the document. **No time.** Thirty minutes buys the ninety-day question and nothing else. Deliver three dated facts and the implication, and say explicitly what was not checked, so the reader can calibrate. ## Quality bar - The reader can act on the first paragraph alone. - Every fact carries a named source and a date in the same line. - Numbers that carry the recommendation are two-sourced, or marked single-sourced in the body. - Company claims, third-party reports, and inferences are visibly different from one another. - The brief contains at least one thing the reader did not know, or says plainly that public information adds nothing. - The gaps are listed, and the questions for the meeting come from them. - Nothing in the document is personal rather than professional, and no adjective is doing the work of evidence. ## Adapting this to your context The method was built for outside-in research on companies, using filings, job postings and archived pages in jurisdictions with public registers. Both assumptions vary widely. - **Statutory filings.** The source stack puts them first because in the United Kingdom, much of Europe and for listed companies they carry real numbers. In markets with minimal disclosure, and for partnerships and private trusts, drop them a tier and lean on hiring, procurement notices and public tenders. - **The ninety-day window.** Suits a company that ships and announces frequently. For a university, a hospital or a public body the equivalent is the last funding settlement, inspection or budget cycle. - **The two-source rule.** Sound for business facts. For a scientific, medical or regulatory claim it is not enough: verify against the publication, the register entry or the regulator's own record, and cite that rather than the press release. - **The privacy boundary.** Public professional information only. Some jurisdictions and sectors go further, restricting what may be held about an individual at all, and where the subject is a private person rather than an executive it tightens by default. - **What not to change.** Every fact carries a named source and a date in the same line, and company claims, third-party reports and your own inferences never blur into one another. ## Related skills `executive-briefing` takes this research and turns it into the page the executive carries into the room, adding the internal relationship history and the positions to take. `market-research` handles the market-level question, including sizing and the full competitive set. `competitive-battlecard` converts a competitor read into the seller-facing page. `partnership-assessment` and `vendor-evaluation` take this evidence and score a counterparty against criteria. `outreach-email` uses the signals found here to make contact specific. `principal-simulator` predicts how the reader will react to what the brief says. `decision-memo` carries any resulting choice to a named decider.