NLP Modelling
Put your best performer in front of the team and ask how they do it. You will hear “I build relationships.” “I listen.” “I read the room.” Every word true, none of it teachable. Research on expertise puts a number on what is missing from that answer — around 70%. Modelling is the discipline built to recover it.
In Short
Modelling is the process of finding the structure that produces somebody’s exceptional result, coding it into a form another person can practise, and testing whether the copy produces the same result. It is the original discipline of NLP rather than one technique among many. Richard Bandler and John Grinder modelled three gifted therapists in the early days, and every pattern taught since — the Meta Model, the Milton Model, anchoring, rapport — is a product of that work. NLP is what modelling left behind.
Written And Reviewed By
Founder And Lead Trainer, NLP Limited
Rajiv Sharma is an NLP Master Trainer certified through The Society of NLP (USA), trained and coached directly by Dr. Richard Bandler, co-creator of Neuro-Linguistic Programming. He holds the ICF PCC credential and created the MARK Model® and the LeadFORTH framework. He is the author of Make a MARK in Life and AI-ENABLE for Sales. Over a career beginning in 1990, he has trained more than 850,000 professionals across 57 countries. His work is endorsed by Marshall Goldsmith, Brian Tracy and John Mattone.
The Number Behind The Whole Discipline
Every organisation runs the same experiment. The top salesperson is asked to present at the national meeting. The surgeon with the lowest complication rate is asked to teach the registrars. The trainer whose classes people talk about for years is asked to write the facilitator guide. In each case a genuinely brilliant person stands up, speaks generously for forty minutes, and transfers almost nothing. Everyone leaves impressed and unchanged.
That outcome has been measured. Researchers in cognitive task analysis found that when experts describe how they perform a difficult task, they unintentionally leave out around 70% of the information a beginner needs in order to do it — a finding so consistent it is known in the field as the 70% rule. One study of experienced surgeons teaching a procedure recorded omissions of roughly 71% of the knowledge steps and 73% of the decision steps. Unprompted, they described about 44% of the procedure. Carefully prompted, that rose to about 66%.
The cause is the expertise itself. Skill becomes automated with practice, which is precisely what makes it fast and reliable — and automated processes run below the level where anyone can inspect them. Your top performer omits the crucial 70% because from the inside there is nothing there to report. Ask them “how do you do that?” and they answer honestly: “I read the room.” The reading is real. The account of it is a summary of a summary.
Modelling exists to recover the missing part. It replaces the interview with something sturdier: observation of the exemplar in action, entry into their experience through second position, precise questioning that goes after the sequence rather than the summary, contrast against people who do the same job less well, and — the step that separates modelling from admiration — a test. You take the copy into the real situation and find out whether it produces the same result. Anything that survives that test is a model. Anything that does not was a story about excellence.
New here? Start with What Is NLP? or the wider library of NLP tools and techniques.
Where It Comes From
In the early days a mathematics student and a professor of linguistics at Santa Cruz set out to answer a narrow question: what exactly do the three most effective therapists alive do that their competent colleagues do not? Richard Bandler and John Grinder studied Fritz Perls in Gestalt work, Virginia Satir in family therapy, and Milton Erickson in hypnosis. They watched, recorded, imitated, and tested.
Their crucial choice was to leave the question of why alone. No theory of personality, no explanation of the psyche — only the observable and recoverable structure of what these three were doing, described precisely enough for somebody else to reproduce it. Satir, asked to explain her results, gave the answers experts give. What the two modellers coded from watching her was a set of patterns she had never named.
The consequence people miss. NLP was the by-product, not the goal. The Meta Model is the coded output of modelling Perls and Satir. The Milton Model is the coded output of modelling Erickson. Every technique in the library is a fossil left by a modelling project — which is why modelling is described as the heart of the field rather than one item in it. Learn the techniques and you inherit somebody else’s discoveries. Learn to model and you make your own.
Judith DeLozier and John Grinder later formalised the modelling state itself. Robert Dilts extended the method to historical figures in his study of the thinking strategies of people such as Aristotle, Mozart, Disney and Einstein, working entirely from written records and artefacts. David Gordon and Graham Dawes built the Experiential Array, a structured framework for gathering the beliefs, criteria, strategies and emotions behind an ability. Four routes to one destination: a description sturdy enough that a stranger can follow it and get the result.
The Framework
Watch a novice imitate an expert and you will see the top layer copied faithfully and the result missing entirely. Skill is stacked, and the layers underneath the visible one are where the result actually lives. A model that omits a layer produces a convincing impersonation and no outcome.
Layer One — Visible
What a camera records. Posture, gestures, pace of speech, breathing, tonality, where they place themselves in a room, what they do with their hands and eyes, the sequence of actions and their timing.
How you gather it Observation and mimicry. Copy it in your own body until it runs without attention, rather than describing it in a notebook.
Copied alone A good impression of the person, and none of their results.
Layer Two — Internal
The order of internal steps that produces the behaviour. What they picture, say to themselves, and feel — and in what sequence. What tells them to begin, what tells them the step worked, what tells them to stop. This is the layer of cues and decisions, and the one the 70% rule eats.
How you gather it Elicitation while they are doing it or immediately after, aimed at sequence rather than summary: what came first, what happened next, how did you know to switch.
Copied alone A checklist that works in the calm and collapses under pressure.
Layer Three — Deepest
What they hold to be true about the task and themselves, what they treat as important, and what they take themselves to be while doing it. The negotiator who believes the other side wants an agreement behaves differently from one who believes they are being hunted — before either says a word.
How you gather it Second position, plus questions about what matters and what would have to be true. Often visible in what they refuse to do.
Copied alone Inspiration that fades within the week, with no behaviour attached.
A complete model carries all three layers and states how they connect: in this situation, believing this, they attend to that cue, and do this — until this signal tells them to stop. That sentence, filled in accurately, is worth more than a fortnight of shadowing.
The Split Most Articles Skip
Read three articles on this topic and you will find three different procedures presented as the procedure. They belong to two genuinely different schools, and the choice between them is the most consequential decision in a modelling project. Choose by the kind of skill in front of you.
| Implicit-First | Explicit-First | |
|---|---|---|
| Associated With | John Grinder and the New Code approach | Robert Dilts, and David Gordon with Graham Dawes |
| The Core Move | Suspend everything you know, absorb the skill by imitation and repetition until you get the exemplar’s result, and only then work out what you have been doing. | Gather the structure first through observation and structured elicitation, code it into a framework, then practise the coded version. |
| The Starting State | A deliberate know-nothing state — no analysis, no interpretation, no theory running while you watch. | Open curiosity with a framework in hand: criteria, beliefs, strategies, emotions, external behaviour. |
| Strongest For | Physical and craft skills, timing, delivery, presence, anything the body learns faster than the mind describes. | Decision-making, judgement, diagnosis, sales and negotiation, anything whose value sits in criteria and choices. |
| Needs | Live access, repeated exposure, and permission to be a beginner for a while. | Interview access, several examples, and discipline in questioning. |
| Fails When | You analyse too early — your existing map overwrites what is actually in front of you. | You accept the first answer — and the first answer is the 30% the expert has conscious access to. |
The working answer for most projects is a hybrid, in this order. Absorb first: watch a great deal, imitate physically, get somewhere near the result with your body before your notebook. Then code what you have acquired, and only then interview the exemplar and compare notes. Interviewing last is deliberate. Ask first and their explanation becomes the map you look through, and you spend the rest of the project confirming the 30% they already knew about.
Interactive Tool
Four questions about the skill you want and the access you have. The planner answers what reading alone leaves open: which school fits this particular project, what exactly to ask, and where this specific set-up is most likely to come apart. It produces a brief you can act on this week.
Step 1 Of 2
What kind of ability is it?
What access do you have to them?
How many exemplars do you have?
Fill in both boxes and choose one option in each of the three rows.
Your Brief
This is what will be sent:
Your brief is private to you — nothing was stored on this page. Sharing it is entirely your choice.
The Method
This is the hybrid sequence used on most real projects: absorb, then code, then test, then transfer. Steps five and six are the two that separate a working model from an appreciative write-up, and they are the two most often left out.
“Great at sales” is unmodellable. “Keeps a cold prospect talking past the first thirty seconds” is a project. Narrow it until you can name the moment it happens, the signal that it worked, and how you would recognise the result in somebody else. The tighter the definition, the shorter the project.
Pick people who get the result consistently rather than occasionally, and where you can, pick two or three who get it differently. One exemplar gives you a model of a person. Several give you a model of the ability, since whatever they share is closer to the structure and whatever differs is personal style.
Before observing, put down what you already believe about how this is done. Everything you know is a filter that shows you your own methods in their behaviour. The modelling state is deliberate emptiness: watching with curiosity and no running commentary, letting detail arrive that your theory has no slot for.
Watch many examples. Copy the physiology in your own body — posture, pace, breathing, timing — until it runs without effort. Then step into second position and be them doing it, noticing what pulls your attention and what feels important from inside their experience.
Write what you have acquired as an explicit sequence with its cues and decision points. Now go to the exemplar and compare notes, using Meta Model questions that hunt for sequence: what happens first, how do you know when to change, what would stop you. Ask about the last specific instance rather than their general approach.
The step almost everyone omits, and the one that produces the real finding. Take an element out of your copy and run it again. Result holds — that element was style. Result collapses — that element is the difference that makes the difference. Repeat until you have the smallest version that still works. That minimum is your model.
Hand the model to a person who was never near the exemplar and see whether they get a comparable result in a comparable time. This is the only honest verdict. A model that works for you alone may be a description of your own talent. A model a stranger can run is transferable knowledge, which is the entire point.
Does it produce the result? Not “is it accurate about the exemplar’s mind” — nobody can verify that. A model is a working map, and maps are judged by whether people arrive.
Does it travel? If it works only for its author it has not left the workshop. Transferability is what turns a personal insight into something an organisation can train.
Where It Earns Its Place
A Result From The Field
At Dangote, growth stopped being a matter of hiring more people and became a matter of finding what the highest performers were doing that the rest of the organisation had never been shown. Coding those patterns and training them across the business accompanied a rise in revenue from $1 billion to $4 billion. The pattern was already inside the company. What changed was that it stopped living in a handful of heads.
One person hits target every quarter and the ride-alongs teach nobody anything. Model the opening, the qualifying question and the moment they stop talking — then train the coded sequence. See NLP sales training.
Thirty years of judgement walks out at the end of the quarter and the handover document contains none of it. Modelling is the most direct route to capturing what the handbook never held — while the person is still in the building.
Some managers walk into a crisis and the temperature drops. That is a modellable sequence of state, stance and first sentence, teachable to a whole layer of managers. Core to leadership development.
The most overlooked exemplar is you at your best. Recall a day when the difficult conversation went well, and model it as if it belonged to a stranger. Most people find a sequence they had assumed was luck.
A client who freezes in board meetings is fluent in three other settings. Model their own fluency and transfer it, rather than importing a stranger’s method. Trained in ICF coach certification.
Course content built from what experts say they do inherits the missing 70%. Content built from what they were observed doing, coded and tested, closes the gap between the classroom and the floor.
Refining The Skill
Cause: It is the obvious move. Fix: Watch first, absorb, code your own version, then interview. Asking first hands you the 30% the expert has access to and makes it the frame you look through for the rest of the project.
Cause: Behaviour is the visible layer. Fix: Add the strategy and the beliefs. Their pauses work because of what they are attending to during the pause. Copy the pause alone and you have silence.
Cause: The brief was “great leadership”. Fix: Narrow to a moment with a visible result. Ability at that size has a structure. Ability at the size of a job title has a biography.
Cause: Skipping the subtraction test. Fix: Remove elements one at a time and watch what happens to the result. Without that, your model contains the exemplar’s accent, their handshake and their coffee order, weighted equally with the part that works.
Cause: One accessible star. Fix: Add a second exemplar who gets the same result differently. What survives across both is structure. What varies is personality, and personality does not transfer.
Cause: The model felt obviously right. Fix: Give it to somebody with no exposure to the exemplar and measure the result. Elegance on paper and transferability in practice are unrelated qualities.
Eighteen Questions Answered
Every question, answered in full. Filter by what brought you here — or read all eighteen in order.
Modelling is the process of finding the structure that produces somebody’s exceptional result, coding it into a form another person can practise, and testing whether the copy delivers the same result. It gathers three layers — observable behaviour, the internal sequence of thinking, and the beliefs and criteria underneath — and states how they connect. It is the original discipline of NLP rather than one technique within it.
Because NLP is what modelling produced. Richard Bandler and John Grinder set out to code what made Fritz Perls, Virginia Satir and Milton Erickson so effective, and the patterns they extracted became the field’s core material. The Meta Model came from modelling Perls and Satir; the Milton Model came from modelling Erickson. Every technique in the library is the coded output of a modelling project.
Research in cognitive task analysis found that experts describing a difficult task leave out around 70% of the information a beginner needs, because practised skill becomes automated and runs below conscious inspection. In one study of surgeons teaching a procedure, roughly 71% of knowledge steps and 73% of decision steps were omitted. This is the exact gap modelling was built to close, and the reason interviews alone are a weak instrument.
Imitation reproduces what can be seen. Modelling reproduces what produces the result, which usually sits below what can be seen. Imitation stops at posture and phrasing; modelling continues into the internal sequence and the beliefs, then subtracts elements to find which ones actually carry the outcome. The test is decisive: imitation looks like the exemplar, and a model performs like them.
Behaviour and physiology — posture, gesture, tonality, timing, the visible sequence. Strategy — the internal order of pictures, sounds, self-talk and feelings, with the cues that start each step and the signals that end it. Beliefs, criteria and identity — what the person holds true, treats as important, and takes themselves to be while performing. A model missing any layer produces an impersonation rather than an outcome.
Define the ability narrowly enough to name the moment it happens. Choose two or three exemplars who get the result consistently. Enter a know-nothing state so your existing methods do not overwrite what you see. Absorb through observation, physical mimicry and second position. Code what you have acquired as a sequence, then interview the exemplar and compare. Subtract elements until the result breaks, keeping the minimum that works. Then test the model on somebody who has never met the exemplar.
It is a deliberate suspension of what you already believe about how the skill is performed, held while you observe. Everything you know operates as a filter, and an experienced observer tends to see their own methods in somebody else’s behaviour. Watching with curiosity and no running commentary lets detail arrive that your current map has no slot for, which is precisely the detail worth having.
Both, in that order: observe first, interview last. Ask first and their explanation becomes the frame you look through, and the 70% rule means that explanation is largely the accessible remainder rather than the working method. Absorb and code your own version first, then bring it to them and compare notes. Their corrections at that point are far more precise than their answers would have been at the start.
Questions aimed at a specific instance and a sequence. Walk me through the last time it happened. What did you notice first? What told you to change what you were doing? What would have made you stop? What are you saying to yourself at that moment, and whose voice is it? What would a competent colleague do here that you never do? That final question is unusually productive, since experts recall their differences more easily than their steps.
It is the smallest set of elements whose removal collapses the result — the working core, once style has been stripped away. You find it by subtraction: take one element out of your copy, run it again, and observe. If the result holds, that element was decoration. If the result breaks, it belongs in the model. This is what turns a long description into something short enough to teach.
One gives you a model of a person. Two or three who achieve the same result by visibly different routes give you a model of the ability, since whatever they share is structural and whatever differs is personal. Adding a contrast case — a competent performer who gets an ordinary result — sharpens it further, because the gap between them and the exemplar is where the useful material sits.
A tightly defined ability with good access takes days rather than months: several hours of observation, an hour of elicitation, and a week of practice with subtraction tests. Broad abilities take far longer, mainly because the definition keeps loosening. Time in a modelling project is governed almost entirely by how narrowly the ability was defined at step one.
Yes, with reduced material. Robert Dilts modelled historical figures such as Aristotle, Mozart, Disney and Einstein entirely from writings, letters, notebooks and accounts by people who worked with them. You lose live observation and elicitation, which is a real loss, and you gain an unusually complete written record. Treat the output as a well-supported hypothesis and test it hard.
It has to work. NLP treats a model as an instrumental map rather than a claim about what is happening inside somebody’s head, which nobody can verify. The question is whether a person following the description arrives at the result. A model that produces the outcome is a good model even where the exemplar describes their own experience differently.
It transfers structure, and it transfers nothing that rests on physical capacity, accumulated knowledge or years of exposure. Modelling a champion sprinter’s mental strategy leaves their physiology behind. Modelling a surgeon’s decision-making leaves their anatomical knowledge behind. It shortens the path to a skill for somebody who has the raw material, rather than replacing the raw material.
Modelling with consent, credit and a share in the outcome is straightforwardly good practice, and most exemplars find the process genuinely interesting since they learn about their own ability. The care needed is around ownership and attribution: coding a colleague’s method and presenting it as your own damages trust permanently. Ask, name them, and let them see the model before anyone else does.
To turn an individual’s advantage into an organisation’s standard. Sales teams model their top performer’s opening and qualifying sequence. Operations model the shift that never has an incident. Companies model retiring experts before the knowledge leaves. At Dangote, coding what the highest performers were doing and training it across the business accompanied revenue growth from $1 billion to $4 billion.
Practitioner level teaches the techniques that earlier modelling produced. Master Practitioner is where modelling itself is taught and where students run a supervised project on an ability they choose. That sequence is deliberate: the elicitation skills, the language patterns and second position are the instruments modelling runs on, so they come first. See certified NLP training.
Where To Go Next
Techniques give you what earlier modellers found. Modelling gives you the ability to find what nobody has coded yet — in your best performer, your competitor, or yourself on the day it went well. It is taught at Master Practitioner level, with a supervised project on an ability you choose. Train it with an NLP Master Trainer on a certified NLP programme, or bring it to your organisation through corporate training.