A useful model is judged by whether it helps someone see, decide or act better, not by whether its letters spell a word. Start from the decision, test the model before naming it, and say where it does not apply.
The business world has an extraordinary appetite for frameworks. Every complicated process can become five stages. Every management dilemma can become four quadrants. Every sensible observation can be attached to an acronym that looks persuasive on a presentation slide.
There is something appealing in this. A framework promises that complexity can be handled without being denied. It gives a group a common language, turns an intimidating question into a set of parts and makes an idea easy to remember.
But a word arranged vertically with five encouraging verbs is not automatically a model. Sometimes it is a useful mnemonic. Sometimes it is a checklist. Sometimes it is a compelling explanation attached to a shape. These are not necessarily bad things. The failure begins when an attractive label is mistaken for tested knowledge.
A good framework should do more than help someone remember what they already agree with. It should help them make a better distinction or reach a more defensible decision.
Begin with the decision, not the letters
Imagine someone asks for a framework for improving agency delivery. It is easy to propose a sequence of verbs: Plan, Organise, Deliver, Review. Each verb is reasonable. Together they are so general that nearly any organisation could claim to follow them. What is still missing is the important part: where and why delivery fails, what choices managers must make, and what the framework helps them notice that ordinary descriptions conceal.
A more disciplined starting question is: What repeated decision is hard, and what is the current failure mode?
Perhaps leaders repeatedly mistake capacity overload for a hiring problem. Or they approve work without explicit acceptance criteria. Or they speed up production without measuring the rework caused downstream. Each issue calls for a different model because each requires a different distinction.
The model should be designed around the discrimination the user needs to make, not around an available word with a satisfying number of letters.
A framework is a selective representation
Philosophers of science distinguish models from the full systems they represent. Models simplify, idealise and emphasise particular relationships to help people reason about a target.1 A framework for professional practice does not necessarily operate at the same level of scientific rigour as a formal physical model, but the caution carries over: any representation leaves something out.
That means every framework has a boundary. A client-retention model may help diagnose service governance but do little to explain a sudden regulatory prohibition. A decision protocol may reduce circular discussion but cannot guarantee that the evidence available is complete. A capability matrix may identify skill gaps but miss organisational politics.
The obligation of the framework creator is to specify what it is for and what it cannot answer. Without those boundaries, apparent universality becomes a way of avoiding responsibility.
Use an existing model honestly
My unpublished essay The AI Opinion Spiral contains a CLEAR protocol: Clarify the decision, Lock the evidence standard, Engineer disagreement, Assign a human owner, and Record the decision with a revisit trigger. Its useful feature is not that CLEAR is an easy word to remember. It is that the protocol names a specific failure: repeated AI-generated advice being confused with independent evidence.
Its proposed remedy intervenes at the process level. Before seeking opinions, define the decision. Before generating explanations, establish what would count as evidence. Before accepting an answer, deliberately search for reasons it might be wrong. Then ensure a human remains accountable for deciding and recording what would reopen the question.
This makes CLEAR a plausible operational aid. It does not make it a scientifically validated intervention. Whether teams using the protocol produce better decisions than comparable teams without it is an empirical question. Nor can the protocol rescue a decision-maker from bad underlying data simply by producing a more organised record.
The distinction between documented original provenance and proven effectiveness must remain explicit.
The agency SYSTEM model in another of my unpublished essays attempts a different task: diagnosing whether apparent talent problems actually arise from the way work is structured. The name is useful only insofar as its components expose missing ownership, poor standards, weak signals and broken learning loops. It should be judged against comparable diagnoses, not against its visual elegance.
These models deserve to be treated as working propositions that can be inspected and improved, not as universal laws.
The six tests a framework should survive
A useful model should first pass the problem test: can a user describe the specific situation in which it helps? If the answer is “all businesses everywhere,” the claim is too broad.
It should pass the distinction test: does it separate cases that would otherwise be confused? If every possible problem produces the same recommendation, it offers little diagnostic value.
Then the action test: after applying it, can someone do something differently, and do they know who owns that action? A concept may be intellectually interesting without being operational.
Next comes the boundary test: can the author say when not to use it? A framework that has no counterexample is often a slogan protected from criticism.
The evidence test asks whether its assumptions are based on primary research, documented practice, anecdote or speculation, and whether that distinction is visible. Finally, the revision test asks what observation or failed result would require the creator to change the model.
WORKSHEET
Try the tests on a framework you already use
0 of 6 tests passed.
Your answers stay in your browser. Nothing is sent or saved.
A framework that passes these questions may still be imperfect. That is not a defect. A model is most useful when its limitations are legible.
Give it a trial before giving it a trademark
Suppose a team develops a new editorial quality framework. It can begin with a deliberately modest pilot. Two editors independently assess a small set of drafts using the proposed criteria. Record where their judgments agree, where they differ and which definitions cause confusion. Compare the framework’s recommendations with actual errors identified later in fact-checking or author review. Revise criteria that fail to distinguish serious risk from minor preference.
Such a trial does not prove that the model works universally. It can reveal whether the language is clear, whether reviewers apply it consistently, and whether the process notices problems that a normal checklist would miss. More rigorous evaluation would require better controls, clearly defined outcomes and a representative sample.
The order matters. First test whether a model has practical meaning. Then communicate it elegantly. Presentation should serve the underlying intellectual work.
Why the simplicity argument deserves respect
It would be foolish to reject acronyms altogether. A simple reminder can be valuable in a noisy operating environment. People may need to recall a safety step quickly or remember to document an exception. Complexity is not automatically a sign of seriousness. An unusably sophisticated framework fails its own purpose.
But simplicity should arrive through careful selection, not careless omission. The goal is to reduce unnecessary cognitive effort while preserving the distinctions that matter. A surgical checklist is short because its critical steps have been chosen, not because the designer believes all procedures are identical.
Likewise, a management framework may be concise while carrying detailed notes about context and failure conditions. The visible diagram is an interface to the thinking, not the whole of the thinking.
The question behind the attractive name
When someone presents a model, there is a respectful but demanding question worth asking: what would I see differently after using this?
If the answer is a clearer classification of the problem, a more reliable decision, or a testable action, the model may deserve attention. If the answer is only that the idea feels memorable, it may still be a good piece of communication. It has simply not yet earned the stronger title of a decision framework.
Good models are not born from the desire to own an acronym. They emerge from sustained discomfort with a problem that ordinary language has not explained well enough.
A memorable name can help a good idea travel. It cannot make a weak idea true.
ABOUT THIS ESSAY
Published 20 November 2026.
Citations & references
- Stephan Hartmann, ‘Models in Science’, Stanford Encyclopedia of Philosophy (substantive revision 2 April 2025). https://plato.stanford.edu/entries/models-science/
Connections
THOUGHT ATLAS
PhilosophyCATEGORIES
Framework Design & Mental Models