i

Description

Ancre abandonnée posée sur un récif corallien, survolée par un plongeur

Source

NOAA (auteur non indiqué) · original page

Licence

PD Public domain

Original photo, unretouched. The background is a version of it stylised by the site.

Anchoring

The first number dropped on the table becomes everyone's starting point, even if it's based on nothing.

In short

• What: a number heard first pulls the estimates that follow.
• Why: we start from the given value and adjust too little away from it; it also steers memory.
• Detect: a discussion revolves around a number whose origin no one knows.
• Avoid: estimate before hearing it, reveal at the same time, ask where the number comes from, consider the opposite.

On this page

1. Definition
2. The mechanism
3. Where you encounter it
   1. In project management
   2. In management
4. Famous cases
   1. The “$999” at the iPad launch January 27, 2010
   2. J.C. Penney and the end of “crossed-out” prices 2012-2013
   3. Anchoring in software estimation 2005
5. Detection signals
6. Avoidance reflexes
7. With AI: detect, correct, mitigate
   1. Detect
   2. Fix
   3. Mitigate
   4. Prompts to integrate into your workflows
   5. How to integrate them
   6. Precautions
8. Limitations
9. Sources and verification
10. See also
    1. Combines with, or reinforces
    2. Related biases
    3. Counterweights and related reflexes

Definition#

Fictitious illustration

Illustration inspired by the person

Anchoring is the tendency to let initial information, the anchor, influence a later estimate, even when that information is unrelated to the question. It was described by Tversky and Kahneman in 1974.

It appears in several forms: an explicit number (“we're talking about three months”), a past amount (last year's budget), an initial ask in negotiation, a previous rating, a market price mentioned in passing.

The mechanism#

Two likely complementary explanations:

• Insufficient adjustment: we start from the anchor and adjust, but stop as soon as the result seems plausible, so we stay too close to the starting point.
• Selective accessibility: the anchor makes compatible information more available (a high number brings to mind things that are expensive, long, difficult).

The effect also impacts professionals. In an experiment by Englich, Mussweiler, and Strack (2006), experienced judges read a shoplifting case file, rolled rigged dice (landing on 3 or 9), then set a sentence: the average sentence was about five months after 3 and about eight months after 9. (Rounded values, as reported by Kahneman.)

Where you encounter it#

In project management#

• Estimation: the sponsor announces “we planned 100 days”; all team estimates cluster around it.
• Budget: the previous year's budget becomes the discussion baseline, whatever the scope.
• Planning poker: when cards are not revealed at the same time, the first stated value pulls the others toward it.
• Negotiation with the client or a vendor: the first proposed price sets the discussion range.

In management#

• Hiring: current salary or the candidate's first ask anchors the offer.
• Evaluation: the previous year's rating steers this year's.
• Objectives: last year's objective becomes the basis for the next one, rather than the actual need.

Famous cases#

Detection signals#

• “How much are we talking, three months?”
• The discussion revolves around a number whose origin no one knows.
• Estimates from different people are oddly close, and close to a value mentioned earlier.
• “Just to set the stage...”: a number given without justification but that sticks.
• Gaps versus last year are always a few percent, whatever the changes.

Avoidance reflexes#

1. Estimate before hearing. Ask everyone to write down their estimate alone, before any discussion and before the sponsor gives a number.
2. Reveal at the same time. Cards turned over together (planning poker), ballots collected then opened (Delphi method): no one speaks first.
3. Ask where the number comes from. “Where does this number come from? What is it based on?” An arbitrary anchor loses strength when it's shown to have no foundation.
4. Build your own anchor. Before the meeting, build an estimate from data: comparable past projects, observed rates, history (see The Planning Fallacy).
5. Consider the opposite. Ask yourself: “What if the right value were half as much? Twice as much? What arguments would support that?” Mussweiler and colleagues (2000) showed this exercise reduces the effect.
6. Separate estimating and negotiating. Work estimation should not depend on expected price or timeline; these are two sequential discussions.
7. In negotiation, anchor back—with justification. If you speak first, give a well-supported number; if the other side speaks first, do not comment on it, start again from your analysis and state where it comes from.

With AI: detect, correct, mitigate#

Prompts to integrate into your workflows#

1. Blind estimation. To ask before giving or showing the slightest number.

Estimate [the effort / the cost / the duration] of this work, with a range, using only the description below.
You have not been given any number and I will not provide one: do not assume any starting value.
Break down your reasoning by component and indicate the level of uncertainty for each one.

Description: [scope, constraints]

2. Questioning the origin of a number. To use when someone announces a number in a meeting or in a document.

This number was given to me: [value] for [item]. I don't know where it comes from.
Write eight precise questions I should ask the person who gave it to find out: what it is based on, when and in what context it was established, what it includes and excludes, and what margin of error it has.
Add the two or three questions that would help determine whether this number is mainly being used as a starting point for negotiation.

3. Consider the opposite and counter-anchors. When a number has already been stated and you want to check your judgment.

A number has been proposed: [value] for [item]. Your role is not to validate it.
1. Give the best arguments that the right value could be at least twice as low.
2. Give the best arguments that it could be at least twice as high.
3. Cite the references or data that should be sought to decide (without inventing any).
4. In one sentence, say what value you would keep if you had to estimate it from scratch.

4. Detect anchoring in meeting notes. On anonymized text.

Here is the transcript of an estimation meeting (names removed). Identify:
- the first number mentioned, by whom (role), and when;
- the numbers mentioned afterward and their distance from the first;
- the passages where someone justified their number with data, and those where they did not.
Conclude with one sentence: does anchoring seem to have played a role, and based on what factual elements?

Transcript: [text]

How to integrate them#

• Before the meeting: each estimator runs prompt 1, then records their own number, before reading the result.
• During: when a number appears, prompt 2 immediately provides a list of questions to ask.
• After: prompt 4 is used as a process review, not a blame exercise.

Precautions#

• Never give the number first. If you write “we're thinking 100 days, what do you think?”, you have just anchored the tool.
• Even without a number, wording steers judgment (“simple,” “small,” “complex”). Describe the scope neutrally.
• References cited by AI may be fabricated: verify any data before using it.
• Do not paste transcripts containing personal or confidential data into an unauthorized tool.

Limitations#

• The effect is very robust in experiments, but its magnitude varies: it is stronger when the domain is less familiar and weaker when the anchor is clearly absurd.
• Not every reference is a toxic anchor: starting from relevant historical data is a method. Bias begins when the reference is chosen for availability, not relevance.

Sources and verification#

• Tversky & Kahneman (1974), “Judgment under Uncertainty: Heuristics and Biases,” Science — the wheel experiment.
• Englich, Mussweiler & Strack (2006), “Playing Dice With Criminal Sentences: The Influence of Irrelevant Anchors on Experts' Judicial Decision Making,” Personality and Social Psychology Bulletin.
• Mussweiler, Strack & Pfeiffer (2000), “Overcoming the Inevitable Anchoring Effect: Considering the Opposite Compensates for Selective Accessibility,” Personality and Social Psychology Bulletin.
• Furnham & Boo (2011), “A literature review of the anchoring effect,” The Journal of Socio-Economics — overview.
• Jones & Steinhardt (2022), “Capturing Failures of Large Language Models via Human Cognitive Biases” — anchoring in language models (NeurIPS 2022).
• Echterhoff et al. (2024), “Cognitive Bias in Decision-Making with LLMs” — model cognitive biases and mitigation.
• Fortune, live coverage of the iPad presentation (January 27, 2010)
• VentureBeat, “Apple’s iPad to sell for as little as $499”
• JCK, “J.C. Penney Lost Nearly $1 Billion in 2012” — fiscal year 2012 figures.
• Aranda & Easterbrook (2005), “Anchoring and adjustment in software estimation,” ESEC/FSE

See also#