The Ladder of Inference

A colleague leaves your project meeting early for the second week running and says nothing about the plan on the way out. By the time you are back at your desk you have decided they have lost interest, and you are already thinking about who could take over their part of the work.

Figure 1. The Ladder of Inference

What it is

The Ladder of Inference describes the steps your mind takes between something happening and you acting on it. Chris Argyris developed it, and the seven-rung version used here comes from The Fifth Discipline Fieldbook. You start at the bottom with what actually happened, select part of it, add meaning, make assumptions, draw a conclusion, adopt a belief and then act, usually in a few seconds and without noticing.

The model's claim is that a conclusion can feel like plain fact even when it rests on a small slice of the data. The ladder gives you a way to climb back down one rung at a time and check what your conclusion is standing on before you act on it.

How it works

Observable Data

What would a camera have recorded?

Start with what actually happened, the words that were said, the email that arrived or the date that passed. Leave out any word that describes a motive or a character, such as careless or uninterested, because those words come from higher up the ladder.

Select Data

Which details did I notice, and which did I skip?

Nobody takes in everything, and what you notice is filtered by what you already expect to see. List the details that support your view and then look for the ones that don't. Two honest people can start from the same meeting and select different facts.

Add Meaning

What did I take this to mean?

You interpret the details you selected through your own experience and culture. Write down the meaning you added and ask what the same detail could mean to someone whose experience is different from yours.

Make Assumptions

What am I treating as true that I haven't checked?

Assumptions fill the gaps between what you know and the story you are telling. Mark each one, and decide which of them you could check with a single question.

Draw Conclusions

Would the other person recognize my conclusion?

By this rung the conclusion feels like simple fact. Say it in one sentence and ask whether the data at the bottom would also support a different conclusion.

Adopt Beliefs

What evidence would change my mind?

A conclusion you repeat hardens into a belief about the person or the situation. The model's reflexive loop says that this belief then decides what data you select next time, so you keep seeing what you already believe. Name one thing that would change your view if you saw it.

Take Action

What do I need to ask before I act?

Whatever you do next rests on the belief at the top of the ladder, which may sit a long way from the data at the bottom. Before you act, take the observation you started with to the person, say what you concluded from it, and ask what you are missing.

When to reach for it

You are certain a colleague did something for a bad reason, and the certainty arrived within a few minutes
Two people on your team came out of the same meeting with opposite views of what happened
You are about to act on a strong reaction to an email, a report or a comment in a meeting

“Their whole perspective on Vancouver was built from mainstream media.”

A worked example

Larchmont Print is a fictional commercial printer. Its head of sales told the operations director that the night supervisor had turned down two rush jobs that month and didn't care about customers. The director had seen the two emails and started to agree. They planned to move rush work to the day shift and tell the night supervisor to be more flexible.

Before acting, the director went down the ladder. The observable data was two one-line emails, on the 4th and the 18th, each declining a rush job. The director had selected those two emails and skipped the eleven other rush jobs the night shift had finished that month. They had read "declined" as unwilling, and they had assumed the night shift had the press time to say yes. The conclusion that the supervisor didn't care about customers rested on that assumption.

The director asked the supervisor about the two jobs. Both had needed the large-format press, and it had been booked for scheduled maintenance on both nights. Nobody had told the sales team about the maintenance.

The director kept rush work on the night shift, because the problem was missing information about the press. The supervisor agreed to post the maintenance calendar to the sales team by Friday, and the director agreed to check with the head of sales at the end of the month whether any rush job had been declined without warning.

Try it this week

Pick one judgment you have made about a colleague in the last week or two. By the end you'll have checked it with them and agreed what happens next.

Write down the conclusion you have reached about the person in one sentence.
Under it, write only what a camera would have recorded, with dates and exact words where you have them.
Add one detail you skipped that doesn't fit your conclusion, and one other meaning the same data could have.
Mark each assumption and write the question that would check it.
Take the observation and your question to the person, and let them answer before you share your conclusion.
Agree what happens next, who does it and by when, and note whether your original conclusion held up.

Try it with AI

Copy this prompt into ChatGPT, Claude or any AI assistant and fill in the part in square brackets. It will walk you through the model one step at a time, using your own situation.

Prompt

I've reached a conclusion about someone at work and I want to test it before I act: [describe what happened, who was involved and what you've concluded].

If you can open links, read https://www.colincox.com/library/ladder-of-inference first. It explains the Ladder of Inference, which climbs from observable data through select data, add meaning, make assumptions, draw conclusions and adopt beliefs to take action.

Start at the bottom and take one rung at a time. Ask me one or two questions about each rung and wait for my answers before you move up. When I describe a motive or someone's character as if it were something I saw, point it out and ask what a camera would have recorded. Ask me which details I've left out, and suggest at least one other conclusion the same data would support.

Finish by asking what I'll do next, who will do it, and by when.

Leave out names and anything confidential before you paste in your situation.

From Colin's work

In the late 90s I was living in Vancouver, and I met a couple from a small town in BC who had never been to Vancouver. When I visited their town, their perspective on Vancouver was that it was full of murderers and all kinds of debauchery. They asked me, "How can you even live there? It's such a cesspit."

Their whole perspective on Vancouver was built from mainstream media.

I was at a work event, and I got into the elevator with a woman I knew really well. I thought I was being a gentleman, so I invited her to press her floor button first. She did, and she was fine.

When I saw her later, she said, "Hey Colin, that's not the context to be a gentleman, just so you know." She said, "Well, it's nice when men are gentlemen, but women can perceive risk differently."

Never in my 48 years had I considered that being a risk, because of my experience as a man.

Go deeper

Read the longer article on The Ladder of Inference

Related tools

Source: Peter Senge, Art Kleiner, Charlotte Roberts, Richard Ross, and Bryan Smith, The Fifth Discipline Fieldbook, 1994

Want help putting this to work with your team? Book a free thirty minutes with Colin.