David Hopkins, IIP Director and Head of Learning, is presenting the ideas behind this article at the AIPIO Intelligence Conference 2026 in Melbourne. He'll speak on Thursday 27 August in the Developing Your Tradecraft stream, teaching the three steps that make an intelligence assessment adhesive: identifying the requirement it serves, calibrating it for the person reading it, and committing to a clear recommendation. Drawing on case studies spanning Silicon Valley Bank's collapse and Kodak's decades-long failure to act on its own market research, his session shows what happens when analysts get two of the three right and miss the third. The full program is available on the conference website.
This paper sets out to help intelligence analysts become more effective by making their intelligence assessments more adhesive. It draws on real experience and established tradecraft research to set out practical steps analysts can apply directly to their own work. It is written for the analyst producing an assessment, but it can also help readers recognise missing steps and know what they need to ask for. Let’s start with an observation by a career policymaker who spent decades reading what intelligence analysts sent him.
US Ambassador Robert Blackwill spent decades on the receiving end of intelligence analysis. That included a posting as Special Assistant to the President and Senior Director for European and Soviet Affairs on the National Security Council Staff during 1989 and 1990, a period spanning the collapse of the Soviet Union and the reshaping of Europe. In a series of interviews conducted between 1991 and 1993, Blackwill told CIA officer and tradecraft specialist Jack Davis, that most of the analysis crossing his desk during that period was not useful to him because the analysts who wrote it did not know what he needed. He termed this nonadhesive analysis.[1] The material that did stick came from analysts who had taken the trouble to learn his schedule and his needs.
Blackwill's diagnosis points to its own solution. If analysis fails to stick because analysts do not know what the reader needs, then analysis built from understanding the reader’s needs should stick. Adhesive intelligence applies this reversal directly. Instead of starting from the analyst's expertise or process and working outward, the analyst starts from the reader's situation, schedule, decision, and needs, and works backwards to the assessment.
Building an assessment this way involves three steps. The first step is identifying the requirement an assessment needs to serve. The second step is calibrating that assessment for the person who needs it. The third step is committing to a recommendation with an honest confidence level attached. This paper explores these three steps and brings them together through real cases.
The foundational reference on structured analytic techniques by Randolph Pherson and Richards Heuer[2] provides a technique for preparing assessments before drafting begins. AIMS sets out four things to consider in advance: the audience, the issue or intelligence question, the message, and the storyline. The first two steps in this paper build directly on two of those elements. Identifying the requirement draws on AIMS' issue or intelligence question, and calibrating for the audience draws on AIMS' own audience step. AIMS’ third element relates to developing the message – what needs to be said and how best to say it. This paper goes a step further than AIMS. For an assessment to be adhesive, it requires committing to a recommendation with an honest, separately stated likelihood and confidence level.
AIMS' fourth element, storyline, concerns how the narrative is structured to carry the message, and it sits outside this paper’s scope.
To make an assessment more adhesive, it is important to ask five questions before starting to draft the assessment:
An analyst who knows what the assessment is for, who needs it, and the assessment deadline can write a shorter, sharper assessment than one working from a general brief to "assess the situation." The second kind of assessment tends to be more comprehensive and often unused. The first is narrower and easier to read and digest.
Paul Prouse, an intelligence trainer and founder of Astro Intelligence Systems who advises organisations across sectors on intelligence capability and process design, has observed the same pattern as Blackwill in his own practice. He distinguishes an assessment that is interesting from one that is useful.[3]
The interesting version demonstrates expertise. The useful version is built to drive action, and that difference shows up in every choice the analyst makes, from what to include, to how the assessment opens.
These five questions are easiest to see in action, so let's work through a live example. Take an embassy security team monitoring a wave of anti-government protests in a country's capital. The protests have turned violent in isolated pockets over the past week, and the ambassador has asked for an assessment of the situation.
Traditionally, an analyst would take this request and produce a comprehensive assessment of the unrest itself, covering its causes, the political actors involved, and how the situation compares to past periods of instability, without first considering who the assessment is for or why and when it is needed.
Asking what requirement this assessment needs to serve surfaces the real need behind the request. The embassy holds a phased drawdown plan that moves non-essential staff and dependents out on commercial flights, and activating it early avoids the risk and cost of a later evacuation using military or chartered transport. The real requirement is whether to activate that drawdown plan now, not whether the unrest itself is serious in the abstract. A general assessment of the protest movement and its causes would be interesting. This narrower requirement, framed around a specific contingency decision, is what makes the assessment useful.
Asking who needs this assessment identifies the ambassador, who holds sole authority to activate the drawdown plan, not the interagency taskforce that reviews the country's political situation monthly. That single reader changes the tone and length of the assessment, and how quickly it needs to be delivered. A task force needs enough context to align on a shared political assessment. One decision-maker who can act immediately needs only what moves them towards making a call they are already positioned to enact.
Asking ‘by when’ reveals a decision point in seven days, the point at which commercial carriers are expected to suspend flights if the airport becomes a target of protest activity. That timeframe rules out waiting for a fuller picture of how the political situation resolves, which may not be clear for weeks, and it sets a hard limit on how much information and analysis the assessment can afford to wait for before it stops being useful.
Asking what happens if they get nothing from you reveals that the default is to do nothing, keeping staff in place and continuing normal embassy operations. If commercial options close before the situation is reassessed, any later evacuation would depend on military or chartered assets, which take longer to arrange and carry greater risk to the people involved. It means the assessment cannot simply describe the unrest and leave the reader to draw their own conclusion. It needs to carry the case for or against activating the drawdown plan itself, since inaction is what happens without it.
Asking what would influence the reader's thinking identifies the two or three data points that matter to this specific decision, rather than everything the analyst knows about the country's politics or the protest movement's origins. In this case, that means whether unrest is spreading toward the diplomatic quarter specifically rather than remaining contained elsewhere in the city, how quickly comparable periods of unrest in the country's recent history escalated to violence targeting foreign nationals, and whether commercial carriers are still confirmed to be operating on schedule. The assessment can be built around those three points, with general disruption commentary left out or moved to an appendix.
The scenario is a national security one, but the shape of the decision is not unique to it. A contingency plan, a single accountable decision-maker, an external deadline, and a default outcome to weigh against all appear in the same form in corporate risk, financial services, and public health work. The specifics might change from sector to sector, but the five questions do not.
Once the audience for an assessment is identified using the second of the five questions from the previous step, the assessment needs to be calibrated for that reader. This means adjusting the wording, trimming or expanding the supporting detail, and being more or less precise about confidence, depending on who's going to read it.
Four checks help with this:
The same underlying intelligence, written for two different readers, shows how much audience calibration matters.
A regional security team identifies a credible, moderately likely threat to a client's facility over the coming month. The same intelligence is written up for two different readers: the security team's own analysts and the client's general manager, and each of the four checks changes what that assessment looks like.
The security team can act on granular detail, adjusting patrol patterns, coordinating with local law enforcement contacts, and tracking specific indicators as the situation develops. The general manager can act on one thing: whether to approve extra security spending for the facility this month. The assessment for the security team needs to give them everything they can use. The assessment for the general manager needs to answer that one question and stop.
The security team will read the full assessment carefully as part of their ongoing monitoring work, so it can run to four pages, opening with the judgment and following with collection sourcing and methodology. The general manager has roughly ninety seconds between meetings. A four-page assessment with the recommendation on page three is effectively invisible to that reader, whatever the quality of the analysis inside it.
The security team shares the analyst's vocabulary and reads a phrase like "the threat stream is corroborated by reporting from two independent sources and is consistent with pattern-of-life indicators collected over the past fortnight" the way it was intended. Unless the general manager has a security background, they wouldn’t understand this sentence. For that reader, the same information needs the collection jargon removed and plain language used: "we're confident in this because two separate sources have told us the same thing, and it matches what we've been seeing on the ground for the past two weeks."
An analyst reading a term like moderate confidence understands it as a specific point on a defined scale, tied to source reliability and corroboration. A general manager without that background might read the same phrase as weak hedging or as strong certainty, depending entirely on their own assumptions. For that reader, the confidence level needs to be spelled out in terms of what it means for the decision at hand, not just labelled: “we're confident enough in this that we'd be surprised if it didn't happen, though not certain enough to rule out other outcomes.”
The four checks above are what turn a technically correct assessment into a useful one. Getting the analysis right is necessary, but it isn't enough on its own. The assessment only does its job once it has been shaped for the person who actually has to use it.
Calibrating an assessment for the audience maintains the same underlying intelligence and the same level of rigour. The only difference is how deliberately it is built for each reader, a design decision an analyst can control just as directly as they control the analysis itself.
Identifying the requirement and calibrating for the audience only get you so far. An assessment can reach exactly the right person, written exactly the way they need it, and still leave them without a clear enough sense of what to do next. The third step is committing to a recommendation.
A recommendation has three parts. The recommendation itself, stated as a clear position. The justification, measured by how likely the underlying event or situation actually is. And the confidence level, measured separately, reflecting how much trust the analyst places in the evidence behind that judgment.
Most intelligence assessments are structured around Bottom Line Up Front, known as BLUF. This writing convention provides the judgment or recommendation first, before the supporting detail, so time-poor readers who stop after the opening line still walk away with the recommendation. The convention started in US military correspondence and has been formal doctrine since 1988.[4]
It has since spread well beyond its origins and is standard practice in Australian military writing, as well as in government and corporate intelligence functions more broadly. It is useful anywhere an assessment needs to be understood and acted on quickly.[5]
An analyst can only put the recommendation first if they have actually settled on one. Writing the bottom line before the supporting detail exposes immediately whether the analysis has resolved into a position or is still hedging.
The justification for a recommendation rests on the likelihood of the underlying event or situation. The US Intelligence Community's ICD 203 standard sets out a seven-point scale for stating that likelihood, from almost no chance through to almost certain, each tied to an approximate probability range.[6]
The UK's equivalent is the PHIA Probability Yardstick,[7] maintained by the Professional Head of Intelligence Assessment, using a comparable scale. Australian intelligence agencies operate within the same Five Eyes intelligence-sharing partnership as the US and UK, though the specific probability and confidence terminology used by any individual Australian agency isn't publicly published.
Confidence is rated separately from likelihood, as low, moderate, or high. This rating is based on the quality and corroboration of the sourcing behind the judgment, how reliable the sources are and how many of them independently agree. ICD 203 specifically discourages combining the two in a single sentence, saying something is unlikely with high confidence, because it leaves the reader unsure whether the uncertainty being flagged belongs to the event or to the evidence.[8]
The same PHIA framework pairs the Probability Yardstick with an Analytical Confidence Rating, applying the same separation across UK intelligence assessments.[9] Regardless of the system being used, the discipline behind them is the same. Likelihood and confidence are written as two distinct statements, so the reader can always tell the difference between the likelihood of an event and the level of confidence in its underlying evidence.
Applied to a real assessment, the three parts appear as separate statements that complete the recommendation.
Poorly written: "We assess with high confidence that the group will likely conduct further attacks." This sentence has no recommendation at all. It also combines likelihood and confidence in the same clause, so the reader can't tell whether "high confidence" refers to the attack happening or to the evidence behind that judgment.
Written properly: "We recommend increasing security posture at the facility. We assess it is likely the group will conduct further attacks. This judgment is held with high confidence, based on corroborated reporting from multiple independent sources." Each sentence does one job. The first is the recommendation. The second states the likelihood. The third states the confidence, tied directly to the sourcing behind it.
Committing to a recommendation carries real risk, since a stated position can be wrong in a way a hedged one technically cannot. A well-known example comes from the Butler Review and the Chilcot Inquiry[10] examining UK intelligence on Iraqi weapons of mass destruction. Both found that assessments had been presented with more certainty than the underlying intelligence supported, leaving decision-makers to inaccurately judge the evidence's reliability.
While there is a risk of making a wrong recommendation, the risk of not making any recommendation has much bigger consequences. An assessment that never commits to a position can't be wrong, but it also can't be checked against what happened, and it gives the reader nothing solid to act on. A decision-maker who can't tell how confident the analyst was can't weigh the advice properly. Naming the confidence level is what turns a recommendation into something a reader can use and gives the analyst a track record that builds credibility over time. In fact, without a recommendation, the intelligence product is just analysis, not assessment.
Identifying the requirement and calibrating for the audience can both be completed, and an assessment can still fail without a commitment to a recommendation. The two cases below come from the finance and business sectors and show what can happen without a recommendation. These examples apply wherever someone is required to identify a risk and act on it.
Silicon Valley Bank's exposure to rising interest rates was identified early. Federal Reserve examiners flagged interest rate risk problems in the bank's own examinations in 2020, 2021, and 2022, but none of those findings included recommendations. Supervisors eventually issued a formal finding in November 2022, tied to a planned ratings downgrade, but the bank collapsed in March 2023 before that downgrade was finalised. By then the risk had been understood for years, but no recommendations were made to force the bank to act on those risks.[11]
Kodak's own market intelligence work in 1981, led by Vince Barabba, the company's head of market intelligence, identified that digital photography would displace film. He estimated roughly a ten-year window before the shift would matter commercially.[12] The study's projections proved remarkably accurate.
The study didn't force a recommendation with a decision attached, and Kodak spent that ten-year window doing very little to prepare. The company filed for Chapter 11 bankruptcy protection in January 2012, three decades after the study's ten-year window had closed, having shrunk from around 145,000 employees at its 1990s peak to a fraction of that by the time of the filing.[13]
A confident, well-supported recommendation still needs the other two steps to become adhesive. Without identifying the requirement, a recommendation can answer a question nobody asked, making it technically sound but disconnected from the decision at hand. Without calibrating for the audience, an assessment with the right recommendation loses adhesion when the reader can’t engage with the language or level of detail, or when it doesn’t reach the person who needs it.
This paper introduced the concept of adhesive intelligence and set out a process for making assessments adhesive. An adhesive assessment is useful, relevant to the decision at hand, timely enough to matter, and delivered in a form the reader can act on. Three steps turn an assessment from interesting into useful. The first is identifying the requirement it needs to serve. The second is calibrating it for the person reading it. The third is committing to a recommendation with an honest, clearly stated confidence level. Any analyst can learn these steps, and analysts who practise them deliberately write assessments that are read, trusted, and acted on.
Barabba, V. (2011). The Decision Loom: A Design for Interactive Decision-Making in Organizations. Triarchy Press.
Blaxland, J. (2023). Australian regional engagement [Speech]. Australian Army Journal, 20(1). https://researchcentre.army.gov.au/library/australian-army-journal-aaj/volume-20-number-1/speech-australian-regional-engagement
Board of Governors of the Federal Reserve System. (2023). Review of the Federal Reserve's Supervision and Regulation of Silicon Valley Bank. https://www.federalreserve.gov/publications/files/svb-review-20230428.pdf
Butler, R. (2004). Review of Intelligence on Weapons of Mass Destruction. UK House of Commons. Full report sourced from the BBC. http://news.bbc.co.uk/nol/shared/bsp/hi/pdfs/14_07_04_butler.pdf
Chilcot, J. (2016). The Report of the Iraq Inquiry: Executive Summary. UK Cabinet Office. https://assets.publishing.service.gov.uk/media/5a80f42ced915d74e6231626/The_Report_of_the_Iraq_Inquiry_-_Executive_Summary.pdf
Davis, J. (1995). A policymaker's perspective on intelligence analysis. Studies in Intelligence, 38(5), 7–15. https://www.cia.gov/resources/csi/studies-in-intelligence/archives/a-policymakers-perspective-on-intelligence-analysis/
Heuer, R. J., Jr., & Pherson, R. H. (2019). Structured Analytic Techniques for Intelligence Analysis (3rd ed.). CQ Press.
Office of the Director of National Intelligence. (2015). Intelligence Community Directive 203: Analytic standards. https://www.dni.gov/files/documents/ICD/ICD-203.pdf
Professional Head of Intelligence Assessment. (2019). Professional Development Framework for All-Source Intelligence Assessment. UK Cabinet Office. https://www.gov.uk/government/publications/intelligence-analysis-professional-development-framework/the-professional-development-framework-for-all-source-intelligence-assessment
Prouse, P. (2026, June). The one and one third assessment [LinkedIn post]. LinkedIn. https://www.linkedin.com/posts/paul-prouse-741283245_intelligenceanalysis-intelligenceleadership-activity-7470327636368703488-ZhqD
Savitz, E. (2012, January 19). Kodak files Chapter 11. Forbes. https://www.forbes.com/sites/ericsavitz/2012/01/19/kodak-files-chapter-11/
US Department of the Army. (2023). Army Regulation 25-50: Preparing and managing correspondence (Administrative revision). Headquarters, Department of the Army. https://armypubs.army.mil/epubs/DR_pubs/DR_a/ARN42124-AR_25-50-007-WEB-13.pdf
This paper was developed with AI assistance for research, structuring, and editing for clarity. All ideas, arguments, and conclusions are those of the author. Final responsibility rests with the Institute for Intelligence Professionalisation in accordance with IIP's AI Use Policy.
David Hopkins is Director and Head of Learning at the Institute for Intelligence Professionalisation (IIP), where he is leading the design of an accredited intelligence training continuum defining what professional intelligence support looks like at a national qualification level in Australia. He is also the Director and Co-Founder of Fire Hawk Services, which builds decision-making capability for organisations in Defence, emergency services, and public safety aviation. A retired RAAF Intelligence Officer with over twenty-five years of service, his career included deploying unmanned aerial systems with Australian Special Forces, strategic all-source analysis at the Defence Intelligence Organisation, and leading the team that developed the contemporary training continuum for Defence's intelligence workforce. His current work sits at the intersection of intelligence professionalisation, operational decision-making, and learning design for complex, high-stakes environments.
IIP builds accredited training pathways and professional standards for the intelligence workforce, and works with the Australian Institute of Professional Intelligence Officers (AIPIO) to professionalise the sector across defence, government, law enforcement, and the many other fields that rely on intelligence support.
[1] Davis (1995)
[2] Heuer & Pherson (2019)
[3] Prouse (2026)
[4] US Department of the Army (2023)
[5] Blaxland (2023)
[6] ODNI (2015)
[7] PHIA (2019)
[8] ODNI (2015)
[9] PHIA (2019)
[10] Butler (2004); Chilcot (2016)
[11] Board of Governors of the Federal Reserve System (2023)
[12] Barabba (2011)
[13] Savitz (2012)