The Ladder of Inference

We act on a story we built in seconds

We like to think we act on facts. We mostly act on a story assembled from a sliver of the facts, built in a couple of seconds, and by the time we notice it we experience it as simply what happened.

The Ladder of Inference, from Chris Argyris and later popularized by Peter Senge, maps that climb. It runs from everything that actually occurred at the bottom to what we do about it at the top. The rungs are worth knowing individually, because catching yourself mid-climb is only possible if you can name the rung you are on.

The rungs

Observable data. Everything that actually happened, in the sense a camera would have captured it.

Selected data. Out of all of that, you notice some of it. What you notice is filtered by what you were already expecting to see, and the filtering does not announce itself.

Added meaning. You interpret what you selected, layering on meaning drawn from your own history and your organization's culture.

Assumptions. You fill the gaps in that interpretation with assumptions, which feel like reasonable inference rather than invention.

Conclusions. You reach a conclusion, which by now feels like a plain reading of events.

Beliefs. The conclusion hardens into a belief about the person or the situation, and stops being something you arrived at.

Action. You act on the belief rather than on the data you started from.

The loop that keeps it in place

The part that makes this more than a description of ordinary thinking is the reflexive loop. Once you hold a belief, it governs what you select next time. The ladder feeds its own bottom rung, so the evidence accumulates in one direction and the belief gets more comfortable rather than more tested.

This is why a fixed view of a colleague tends to strengthen over time regardless of what they do. Their good weeks are not selected, but their bad ones are, and each one confirms what you already concluded.

Where the climb usually goes wrong

Consider a report that arrives late. That is observable. What follows usually is not.

You select that it is the third time this quarter, and do not select that two of those coincided with a system migration. You add the meaning that deadlines are treated as optional here. You assume the person is not particularly bothered. You conclude they are careless, you come to believe they are unreliable, and you start routing important work around them.

The action is now several rungs away from the only thing you actually observed, and the person has no idea any of it happened. From the outside, all they see is that the interesting work stopped arriving.

The most useful place to intervene is the second rung, where the selecting happens. Two honest people can select different facts from the same meeting and climb to opposite conclusions without either of them lying, because the divergence happened one rung above the data and neither of them saw it.

One thing to try this week

Pick a situation where you are currently certain about why someone did something. Certainty is the signal, because the ladder is invisible when the conclusion feels obvious.

Then answer two questions in writing. What did I actually observe, restricted to what a camera would have recorded. And what else could that same data support, generated honestly rather than as an exercise in fairness. If you cannot produce a second reading that a reasonable person could hold, you have probably not gone far enough down. Go and ask the person before you act on the first one.