%202.png)
Implication Development
Turn an insight about the world into what follows for this organisation, with every assumption about the organisation named, sourced and open to challenge.
Free. Works on Claude, ChatGPT, Gemini or any assistant that accepts a skill file.
What this skill does
The method, encoded.
Between "here is why customers behave this way" and "here is what you should do" sits a step almost nobody writes down. It is where the research meets the organisation: what this truth means for a business with these products, these processes, these metrics, these constraints and this history. Skipped, it produces the two commonest failures in reporting. Either the report stops at the insight and the client says "interesting, so what", or it jumps to a recommendation whose reasoning is invisible, so nobody can tell whether the action follows from the evidence or from the researcher's assumptions about a company they have known for six weeks.
The step gets skipped because it needs knowledge the research does not contain. The professional answer is not to skip it but to make the borrowed knowledge explicit. This skill generates the consequence through six function lenses, because the same insight lands differently on product, service, commercial, communications, measurement and capability, and the sharpest consequence is rarely for the team that commissioned the work. Then it puts every organisational assumption into a register with its source, so a reader who knows the company better than you do can correct one proposition instead of rejecting the argument.
It also insists on two things people avoid: stating the expensive or inconvenient implication at full strength, and reporting honestly when an insight has no consequence for this organisation at all.
Best used for
- Turning insights into consequences a specific organisation can check
- Reports read by several functions who each need the consequence in their terms
- External suppliers who must state what they are assuming about a client
- Insights that undermine a programme already funded
- Deciding honestly whether an insight has any consequence for this organisation
- Auditing AI-drafted implications for organisational naivety
Typical inputs
What you give it.
At least one insight with confidence, findings and bases, An organisational frame: what it does, who acts, what is open, what it measures, The decision the research informs, and its owner, What is already underway or recently committed (optional), What has been tried and failed, and why (optional), The organisation's own metric and reporting definitions (optional), Client-supplied operational or transactional data (optional), Regulatory, contractual, funding and capability constraints (optional), Stakeholder interviews or the debrief record (optional)
Typical outputs
What you get back.
Implication records with consequence, assumptions, confidence and review points, Implication matrix of insights against six function lenses, Consolidated assumption register with source labels for every assumption, Insights with no implication for this organisation, with reasons and review triggers, Marked uncomfortable implications, tracked from working document to deliverable, Approximate sizing separating measured quantities from labelled assumptions, Marked human judgement points for the strategic step, per K5 section 2.3
Method coverage
What the skill works through.
- Insight, implication, recommendation: three levels people collapse into one
- The knowledge the research does not contain, and what to do about it
- Building an organisational frame, and labelling every element by source
- The six lenses: product, operations, commercial, communications, measurement, capability
- Why the measurement lens usually produces the sharpest implication
- Writing an implication that states a consequence and contains no instruction
- The assumption register, and why it is the deliverable
- The "so what" test, applied honestly, with deletion rather than rewriting
- Sizing approximately without inventing a number
- The expensive implication, the one that contradicts a funded decision, and the obligation not to soften
- When an insight has no implication for this organisation
- Why an implication can never be more confident than the insight above it
Download
Free skill. One file.
Enter your email once. Every skill you download after that takes a single click.
How to install
Add the skill file and the five kernel protocols to a Claude Project, a ChatGPT Project, a Gemini Gem, or paste them at the top of any assistant conversation. Then give it your real research material, not a description of it.
Download skillQuestions
Common questions.
What is the difference between an insight and an implication?
An insight is a claim about the world: why people behave as they do. It would be true whichever organisation commissioned the research. An implication is a claim about one organisation: what is now true of its situation given that insight. The practical test is portability. If the sentence could be pasted into a report for a competitor without changing anything, it is still the insight.
What is the difference between an implication and a recommendation?
An implication is a consequence; a recommendation is an action. The mechanical test is whether the sentence contains a verb the organisation is meant to perform. "The availability metric counts a substitution as a fulfilled order, so the operating scoreboard cannot see the failure" is an implication. "Add a second measure to the weekly report" is a recommendation. Keeping them apart preserves the organisation's right to choose its own response, and keeps the reasoning visible at exactly the point where money starts moving.
How do I write implications when I don't work inside the organisation?
Write the assumptions down. Build a short frame covering what the organisation does, which functions could act, which decisions are open, what is already committed, what it measures itself on and what constrains it. Label every element by source: from a client document, from a stakeholder's account, from your own inference, or unknown. Then put every assumption an implication rests on into a register with what changes if it is wrong. A reader who knows the company can then correct one proposition rather than dismiss the whole argument.
Why do the same findings mean different things to different teams?
Because a mechanism touches an organisation in several places at once. An insight about substituted grocery items lands on the product team as a screen design question, on operations as an optimisation objective, on commercial as an exposure in a different channel, and on measurement as a metric that records the failure as a success. Run all six lenses deliberately. The consequence that matters most is frequently for a function that did not commission the study.
What is the "so what" test and how do I apply it honestly?
Three questions. Would this consequence be stated regardless of how the finding came out? Does the organisation already know this and act on it? Does anything about its situation change if this is true? When one fails, delete the implication rather than rewriting it. Rewriting produces something that reads better and says the same nothing, and that is the single most reliable failure of AI-drafted implications.
Should I put a number on an implication?
Only where the number was measured. Research data can support incidence, the proportion of a sample affected, the share of a described population, and rough magnitude. It cannot on its own support revenue, cost, conversion, elasticity or lifetime value. In particular, never multiply a measured incidence by an unmeasured value and present the result as a size. State what is measured, with its base, and name what would be needed for the rest.
What do I do with an implication that contradicts a decision the client has already made?
State it at full strength, then state the boundary of your work: whether a committed decision should be reversed depends on contract terms, sunk cost and strategic intent that the study does not observe. That sentence is not a hedge, it is accurate, and it makes the difficult consequence easier to state rather than harder. Watch for the four ways softening happens: moving the consequence into a subordinate clause, adding a hedge the evidence does not require, placing it last, or generalising it until nobody has to own it.
Can an insight have no implication for us?
Yes, and reporting that is a real result. Four reasons are legitimate: the mechanism is outside your control, you already address it and the evidence confirms the current approach, it affects a population you do not serve, or the magnitude is below the level at which you act on anything. Record the insight, the reason and a review trigger so it can be picked up when circumstances change. A manufactured consequence is worse than none, because it will be acted on.
Can an implication be more confident than the insight it comes from?
No. It inherits the insight's confidence and then adds assumptions of its own, so it can only lose. Apply two caps: never exceed the insight's level, and drop one level where the implication rests on an unverified organisational assumption whose falsity would change the consequence. A moderate insight with two unverified assumptions is a low-confidence implication and needs the language and validation statement that go with that.
Why does AI produce implications that sound right and are wrong?
Because organisational naivety is invisible on the page. A language model generates consequences that are logically sound and ignore funding cycles, contract terms, capability, a restructure announced last month, or the fact that this was tried in 2023. The fix is structural: build the organisational frame before generating anything, label inferred elements as inferred, and route the strategic step to a person who holds the knowledge the research does not contain.
The skill chain
Works well with.
Research where people already are.
Analyse it where you already work.
Yazi helps researchers conduct surveys, AI interviews and longitudinal research directly through WhatsApp.
