[Skip to content](https://academy.innguma.com/en/ai-applied-to-intelligence/analysing-relationships-and-producing-intelligence/#contenido) Lesson 3.4 · Learning path 03 # Analysing relationships and producing intelligence How do you move from a collection of records to strategic hypotheses without turning correlations into facts? **You take away** hypothesis matrix and draft briefing with uncertainties. ≈ 35 min · Intermediate [Do the practice](https://academy.innguma.com/en/ai-applied-to-intelligence/analysing-relationships-and-producing-intelligence/#hazlo-tu-tres-hipotesis-una-recomendacion-condic)[Exercise](https://academy.innguma.com/en/ai-applied-to-intelligence/analysing-relationships-and-producing-intelligence/#compruebalo)[Read the lesson](https://academy.innguma.com/en/ai-applied-to-intelligence/analysing-relationships-and-producing-intelligence/#el-analisis-no-es-un-resumen-mas-largo) In 1 minute an intelligence product is not «the answer»; it is a reviewable representation of the evidence and its implications for a decision. ## Analysis is not a longer summary A summary reduces length. An analysis tries to explain what is changing, why it might matter, what evidence supports it and which alternatives are plausible. A good intelligence product may conclude that **there is not yet enough evidence to recommend an action**. That is not a failure; it is a useful conclusion if it identifies what is missing and when to review. The analyst combines three levels: - **Description:** what we know and from which documents. - **Interpretation:** which compatible explanations the signals offer. - **Implications:** which decisions could change depending on which hypotheses are confirmed. AI can propose groupings, build timelines, identify contradictions and draft alternatives. Its danger is filling gaps with plausible stories. We must require that each hypothesis has supporting evidence, contrary evidence and a refutation criterion. Interactive · Tag the text ### Description, interpretation or implication Label each sentence of this draft analysis. Activity: label sentences as description, interpretation or implication. ## From a signal matrix to testable hypotheses A simple structure has these columns: hypothesis, signals that support it, signals that contradict or qualify it, assumptions, alternative explanation, decisive test, implication if confirmed, next review. There is no obligation to score hypotheses with numbers that look scientific. For Company A: **H1: «Industrial thermal storage is moving into a commercial phase that is relevant for Company A».** S01, S03 and S04 illustrate activity, but they do not prove sales. The alternative explanation H2 is «technical feasibility is being explored without broad commercial adoption». The evidence that would help tell them apart would be a documented purchase, validated industrial operation or identifiable contracts. A third hypothesis, H3, could be «the regulatory change will accelerate demand». S06 and S15, within the simulation, force us to check status and timeline; they are not enough to assert a binding obligation in force or future adoption. Three rival hypotheses with their current support. The least contradicted one is not thereby proven (fictional case). ## Choosing a product according to the decision Do not start with «I want a SWOT» if the decision calls for a technology radar. A **SWOT** organises internal and external factors to compare options; a **competitor profile** explains one actor; a **signal matrix** lets you prioritise indications; an **executive briefing** communicates a decision; a **scenario** explores futures under uncertainty. Each product needs different sources and rules. That is why Intelligence Lab does not produce a single generic prompt. The user first selects the job to be done and then gets a specific protocol: data, analytical steps, outputs, review, limitations and repetition. The final analysis is run in the AI the user chooses, not inside this Lab. Interactive · Match ### Which product does this decision call for? Don't start with “I want a SWOT”. Choose the product according to the need. Activity: match each need to the right intelligence product. Do it yourself ## Do it yourself · Three hypotheses, one conditional recommendation (25 min) Using S01–S16, write three incompatible or partly alternative hypotheses. For each one, ask the AI for evidence for and against, identified by ID. Then build a recommendation for a **reversible action** for Company A: what to investigate over the next 60 days, which signals trigger a review and which decision is NOT yet justified. **Deliverable:** hypothesis matrix and draft briefing with uncertainties. R10 · Analysis ### Matrix of hypotheses and alternatives Assess rival explanations of a signal. [See the template →](https://academy.innguma.com/en/library/matrix-of-hypotheses-and-alternatives/) R14 · Intelligence product ### Intelligence product specification and briefing Define the need and deliver actionable conclusions. [See the template →](https://academy.innguma.com/en/library/intelligence-product-specification-and-briefing/) Check it · self-assessment ## Check your hypothesis matrix This part needs professional judgement: there is no automatic marking. Complete your work in the practice, rate it with the lesson's rubric and compare it with the indicative solution. What your answer should include - Three rival hypotheses with evidence for and against, by ID. - The recommendation is a reversible action. - You point out which triggers would force a review. - You state which decision is not yet justified. Self-assessment rubric (0–2 per criterion) This rubric does not determine the grade of the certified exam. Claim-evidence correspondence Uncertainty management Error detection Decision and review Professional reuse **Indicative solution** H1, growing experimentation; H2, possible commercial transition; H3, demand influenced by regulation. H1 has several technical indications; H2 lacks purchases and independent results; H3 requires verifying legal scope and validity. Solution: limited exploration, exploratory contact and monitoring with explicit triggers; not an irreversible investment based on two pilots. **Mistake to avoid:** Declaring the market proven. Illustrative solution for a fictional case: reasoned alternatives are valid. ## Common mistakes Confusing temporal sequence with causality, attributing business intent without proof, turning isolated signals into trends, mixing assumed internal strengths with verified external facts, drafting recommendations without options or costs of being wrong. **Rule to take away:** an intelligence product is not «the answer»; it is a reviewable representation of the evidence and its implications for a decision. **Optional extension Deep dive · A recommendation is worth its conditions for change** A hypothesis is an explanation subjected to testing, not a story that sounds possible. To avoid selective confirmation, the same R10 matrix must include signals that support and signals that refute or limit each interpretation. When sources are dependent, five headlines are not equivalent to five independent proofs. A responsible analysis makes explicit **which new piece of data would change the conclusion and who will look for it**. ### Company A · Three interpretations compatible with the fictional signals | Working hypothesis | Possible support | Missing / alternative | Review trigger | | --- | --- | --- | --- | | H1 · Technical exploration is increasing | S01, S03 and S04 | Technical intensity does not imply a purchase | New independent technical test | | H2 · Commercial adoption is approaching | Some communications about pilots | Operations, clients and volume are missing | Contract or installation with verifiable performance | | H3 · A regulatory change accelerates demand | Regulatory signals S06 and S15 | Status, territory and validity must be clarified | Final official text, timeline and scope | No hypothesis is «true» simply because it appears in the table. The analysis compares which one is compatible with what has been observed and which uncertainty is costliest to ignore. The output for management should not be «we must invest», but a set of differentiated options: explore internally at a bounded cost; look for a validation partner; observe with explicit conditions for deciding again. ### Example of a 100-word briefing > **Decision:** whether or not to devote resources to exploring industrial thermal storage. **Observations (simulation):** S01/S03/S04 show technical activity and communication about a demonstrator. **We do not know:** scale of adoption, comparable performance or purchasing conditions. **Interpretations:** the activity may foreshadow adoption or remain at an experimental stage. **Options:** limited exploration, contact with partners or monitoring without investment. **Next milestone:** check documented installations and applicable regulatory changes. **Review:** 60 days or earlier if relevant contractual evidence emerges. ### Practice · Explain your own uncertainty Ask the AI for a matrix and check: does each hypothesis have evidence, counter-evidence or a gap? Are alternatives distinguished from implications? Is causality attributed without proof? Rewrite a categorical recommendation as three reversible options, each with its cost of being wrong and an update trigger. Another person should be able to point out which part of the decision would change when a new signal is added. **Deliverables:** R10 Matrix of hypotheses and alternatives and R14 Intelligence product specification and briefing; ## References and limits - Richards J. Heuer, Jr. · Center for the Study of Intelligence, CIA. *Psychology of Intelligence Analysis (cap. 8, Analysis of Competing Hypotheses)* (1999). [Document](https://www.cia.gov/resources/csi/static/Pyschology-of-Intelligence-Analysis.pdf) Basis of the competing-hypotheses method: the aim is to refute hypotheses with inconsistent evidence, not to confirm one. Reference verified on 2026-10-18. - Company A is a fictional case: its data, companies and figures do not describe real markets or organisations. - The guidelines on AI are the Academy's own work. They are not a NIST procedure and do not replace a risk assessment of your organisation. Intelligence Lab · L06 ## Hypothesis matrix Ready to apply it? Open the Hypothesis matrix in the Lab and download your rival explanations and the available evidence. Output: hypothesis matrix. Works in the browser, no server, no account and no connected AI. The result may need reviewing and adapting to your case. - [Open the tool](https://academy.innguma.com/en/lab/hypothesis-matrix/) - [Back to the learning path](https://academy.innguma.com/en/ai-applied-to-intelligence/) Intelligence Lab · L09 ## Intelligence prompt factory Ready to apply it? Open the Prompt factory in the Lab and generate a reusable prompt for your next intelligence product. Output: reusable prompt. Works in the browser, no server, no account and no connected AI. The result may need reviewing and adapting to your case. - [Open the tool](https://academy.innguma.com/en/lab/intelligence-prompt-factory/) - [Back to the learning path](https://academy.innguma.com/en/ai-applied-to-intelligence/) **Download** - [R14 · Intelligence product specification and briefing](https://academy.innguma.com/en/library/intelligence-product-specification-and-briefing/) - [R10 · Matrix of hypotheses and alternatives](https://academy.innguma.com/en/library/matrix-of-hypotheses-and-alternatives/) Open reading, no registration.