Unlearning

Thinking Tools6.4 Conducting Follow-Up and Taking FeedbackAlso known as Belief Dis-Updating, Organizational Unlearning

From the organizational learning work of Chris Argyris and Donald Schön (Organizational Learning, 1978), Bo Hedberg (How Organizations Learn and Unlearn, 1981), and Paul Nystrom and William Starbuck (To Avoid Organizational Crises, Unlearn, 1984).

Audits a firmly held belief against its actual evidence and lowers confidence to match, then adjusts the beliefs and actions that rested on it.

Reweigh each piece of evidence, and the confidence falls to what it supports

A software team held at 90 percent that its customers will never pay for support. Each slider is one piece of the evidence, set as odds for or against the belief, and the first has already been reweighed from the team's old 3 to 1 down to Equal. Reweigh the other three on what each actually shows: the point falls toward the confidence the evidence supports, and every action that needs more confidence than that comes up for review. Illustrative.

Equal
3 to 1, For
Equal
Equal
Confidence the evidence supports, starting from even odds
75%
Gap from the 90 percent the team held
15 points
Actions resting on the belief that now need review
1 of 3
Sell no support plans: needs 85%No support team in the budget: 70%Free support to win deals: 50%0%25%50%75%100%Held: 90%SupportedSell no support plans: needs 85%No support team in the budget: 70%Free support to win deals: 50%0%25%50%75%100%Held: 90%Supported

An example

A company that sells maintenance-planning software to mid-sized manufacturers had run for years on one belief: our customers will never pay for support. It shaped the price list, the budget, and the sales pitch. When the head of customer success asked the leadership team how sure they were, the answer was ninety percent. She asked for the evidence.

There were four pieces. Two customers had refused a paid plan in 2019, before the enterprise tier existed. Sales said customers expected free support, though no one had asked a customer. A rival that bundled free support had won three deals. And fourteen customers had paid for emergency site visits last quarter, a fact no one had connected to the belief. Weighed on what each one actually showed, the evidence supported about one chance in three. Every action resting on the belief came up for review, and the team priced a paid support tier for twenty customers as a trial.

Why it works

Confidence is meant to summarize evidence, and over time the two come apart. A belief that worked gets repeated, becomes policy, and keeps its old confidence long after the evidence that earned it has aged or turned. Unlearning makes the summary answer to its parts again. Each piece is weighed on its own as odds for or against, and the odds multiply, which is how Bayes's rule combines independent evidence.

Two things follow. A stale piece set back to even stops propping up the total, and a piece the team never connected to the belief can pull it down hard. The last step is the one that changes behavior: each action resting on the belief needs a certain confidence to stay justified, and the audit shows which ones have lost it.

The thinking behind it

Bo Hedberg Bo Hedberg, a Swedish management scholar at Stockholm University, brought unlearning into organization studies in How Organizations Learn and Unlearn, a chapter in the Handbook of Organizational Design (1981). He described unlearning as the process by which an organization discards knowledge that has become obsolete or misleading, so that new knowledge has room to take its place. Only a small part of the chapter deals with it, and scholars still argue whether unlearning is a process of its own or a part of learning.

William H. Starbuck William H. Starbuck, an organizational scientist who taught at Wisconsin and New York University, wrote To Avoid Organizational Crises, Unlearn (1984) with Paul C. Nystrom, its first author. Their argument is that past successes harden into beliefs, that leaders hold on to those beliefs as conditions change, and that organizations drift into crises as a result. The old beliefs have to be unlearned before new ones can be learned. The audit on this page is a small, routine version of that discipline.

Chris Argyris Chris Argyris, a Harvard scholar of organizational learning, wrote Organizational Learning: A Theory of Action Perspective (1978) with Donald Schön. They distinguished single-loop learning, which corrects actions within the existing assumptions, from double-loop learning, which questions the governing assumptions themselves. Lowering the confidence in a long-held belief, and then reviewing what was built on it, is double-loop work: the assumption is the thing being corrected.

How to run it

  1. Name the belief and the confidence held. Write it as one plain sentence and ask the people who act on it how sure they are, as a number.
  2. List the evidence actually behind it. Only what someone could show; mark anything that is hearsay or older than the current product.
  3. Weigh each piece on its own. Ask how much more likely this piece would be if the belief were true than if it were false, and write the odds.
  4. Look for evidence the belief has ignored. Recent facts that point the other way often sit in another team's reports.
  5. Recompute the confidence. Start from even odds or a base rate, multiply the odds, and set the result beside the number the team held.
  6. Review the actions that rest on the belief. For each, ask what confidence it needs to stay justified, and reopen any the new number no longer supports.
  7. Log the change. Record the old and new confidence and the date, so repeated overconfidence shows up across audits.

Where it goes wrong

The first failure is auditing only the evidence already on the list. The belief's defenders assembled that list, and the piece that matters most is often one nobody connected to it. Ask a team that does not share the belief to bring evidence too.

The second is double counting: three reports that repeat the same sales anecdote are one piece of evidence. Merge pieces that share a source before weighing them. The third is stopping at the number. A lower confidence that changes no budget, price, or plan has been noted and nothing more; the audit is finished only when each action resting on the belief has been kept or reopened by name.

When to use something else

  • Bayesian Updating. When the belief is new and evidence is still arriving, so the work is updating forward one observation at a time.
  • Double-Loop Learning. When the trouble is a governing assumption behind a whole system of decisions, more than one belief's confidence.
  • Assumption Audit. When a plan rests on many assumptions and the first job is listing them and ranking which would hurt most if wrong.

Where it sits in SOLVE eX

Go to the source

  • Organizational Learning: A Theory of Action Perspective, 1978
  • How Organizations Learn and Unlearn, Handbook of Organizational Design, vol. 1, Oxford University Press, 1981
  • To Avoid Organizational Crises, Unlearn, Organizational Dynamics, vol. 12, 1984

Back to the catalog