The first number a person sees tends to stick, even when it has nothing to do with the question. That is anchoring bias. It shapes what a customer thinks a reward is worth, what a partner offers for your points and what an analyst assumes for next year’s breakage. This entry covers what anchoring is, how strong the evidence is and where a loyalty team runs into it.

What is anchoring bias?

Anchoring bias is the pull of a starting number on a later estimate. People start from an initial value and adjust from it, and the adjustment is usually too small. Amos Tversky and Daniel Kahneman named it in 1974. A wheel of fortune stopped on a number from 0 to 100, and people then estimated the share of African countries in the United Nations. Groups that saw 10 gave a median of 25. Groups that saw 65 gave a median of 45. Payoffs for accuracy did not shrink the gap (Tversky and Kahneman, 1974).

The anchor does not have to be handed over. High school students had 5 seconds to estimate a product written on a board. One group saw 8 x 7 x 6 and on down to 1. Another saw 1 x 2 x 3 and on up to 8. The median guess was 2,250 for the first group and 512 for the second. The right answer is 40,320 (same paper). Both groups started from their first few steps and stopped too soon.

Why does anchoring happen?

People stop adjusting once a plausible value appears. Nicholas Epley and Thomas Gilovich found that adjustment from a self-made anchor ends when the person reaches a reasonable number. It goes on only when people are able and willing to search for a better one (Epley and Gilovich, 2006). The first number is the starting point. Plausibility sets the stopping point.

How strong is the anchoring effect?

For estimates, it is strong and it replicates. The Many Labs project reran 13 classic effects, four of them anchoring tasks, in 36 samples with 6,344 participants. The anchor moved estimates in every sample, and the effects were larger than in the original study (Klein et al., 2014). For prices, the picture is weaker. Two later studies tried to reproduce the best-known price results and found much smaller effects.

StudyAnchor and taskWhat it found
Tversky and Kahneman (1974)Wheel of fortune, then a UN estimateMedian 25 after an anchor of 10; 45 after 65
Ariely, Loewenstein and Prelec (2003)Last two digits of a social security number, then a bid for a productTop-fifth numbers bid $56 for a keyboard; bottom-fifth numbers bid $16
Many Labs (2014)Four estimate tasks, 36 samplesAnchor effect in every sample; effect sizes of 1.27 to 2.60 against 0.93 originally
Fudenberg, Levine and Maniadis (2012)Redo of the Ariely manipulationVery weak effects on goods; none on lotteries
Li and colleagues (2025)Replication with power raised from 46% to 96%3.4% shift in bids against 31% originally
Galinsky and Mussweiler (2001)Who makes the first offer in a negotiationThe first-offer maker got the better outcome

The honest summary is a split. Anchors move numerical estimates by a lot. Anchors on what people will pay are real in some tests and small in the best-powered ones. Li, Weigel, Ferraro and Messer estimate that most designs in this field reach 30% power at best. So the effects that clear the bar are overstated (Li et al., 2025).

Can an arbitrary number change what people will pay?

In the original test it did. Dan Ariely, George Loewenstein and Drazen Prelec showed 55 MBA students six products. Would they buy each one at a price equal to the last two digits of their social security number? Then each student gave a maximum bid. Students with above-median numbers bid 57% to 107% more than students with below-median numbers. For a cordless keyboard, the top fifth bid $56 on average and the bottom fifth $16 (Ariely et al., 2003).

The result later proved fragile. Fudenberg, Levine and Maniadis reran the test. They found very weak effects on goods and none on lotteries (Fudenberg et al., 2012). The Li team found 3.4% against 31% in the original bids. Expect a smaller price effect than the early papers claim.

Negotiation is firmer ground. Adam Galinsky and Thomas Mussweiler ran three tests. Whoever made the first offer got the better deal. First offers also predicted the final price. The edge vanished when the other side thought about the rival’s options, the rival’s lowest price or their own goal (Galinsky and Mussweiler, 2001).

Where does anchoring show up in a loyalty program?

Three places matter to an operator. The first is the dollar figure beside a points price. Starbucks lists each Star tier with a dollar value. It offers 25 Stars for $1 off, 60 for $2 off, 100 for up to $6, 200 for up to $10, 300 for up to $16 and 400 for up to $20 (Starbucks Rewards Terms). That puts a dollar reference next to every price. By simple division, the implied value per Star runs from about 3.3 cents at the 60-Star rung to 6 cents at the 100-Star rung. Whether the dollar figure steers choices needs a test.

The second is the first number in a partner negotiation. A team selling points to a partner or agreeing a reimbursement rate faces the Galinsky result. The side that names a number first tends to pull the final price toward it. The fix is the one the study found: go in knowing the partner’s alternatives and your own target.

The third is your own forecast. An analyst who starts from last year’s breakage or redemption rate and adjusts is doing what the 1974 students did. Build the estimate from its drivers first, then compare it with last year’s number.

How can a loyalty team test an anchor?

Run a randomized test, and size it for a small effect. For illustration, 20,000 members reach 100 Stars in a month. Half see the app’s maximum dollar value beside the reward, and half do not. If 38% of the first group redeem within 14 days against 34% of the second, the figure added 4 points, about 400 redemptions among 10,000 members.

The sample must match the effect. A standard two-group test at 80% power and a 5% significance level needs about 2,300 members per group to detect a lift from 34% to 38%. Detecting a lift from 34% to 35% needs about 35,500 per group. That is the Li lesson in practice: small tests miss small effects, and the ones that do clear the bar overstate them.

Log each member’s exposure and outcome, and keep the holdout fixed until the window ends. Then repeat the test on a second segment before changing the program, since a small effect can differ by group.

How GRAVTY handles anchoring bias

An anchor is a display and a rule, so a team needs to change both quickly. The GRAVTY technology page says business teams configure loyalty logic without code. On the retail solutions page, Loyalty Juggernaut describes AgenticTest as watching lift in flight and offering live A/B simulation before launch. That fits a test like this one.

Frequently asked questions

What is an example of anchoring bias?

The classic example is the wheel of fortune study. People who saw a wheel stop on 10 estimated that 25% of United Nations members were African countries, and people who saw 65 estimated 45% (Tversky and Kahneman, 1974). The number on the wheel was random, and it still moved the answers.

Who discovered anchoring bias?

Amos Tversky and Daniel Kahneman named it in 1974. Their paper in Science listed adjustment from an anchor as one of three mental shortcuts that cause errors (the 1974 paper).

How do you avoid anchoring bias?

Set your own estimate before you see the anchor, and look for evidence against it. In negotiation, the first-offer edge vanished when the other side weighed the rival’s options, lowest price or own goal (2001 study). Searching for a better number helps too, since adjustment stops at the first plausible one (2006 study).

Is anchoring bias real?

Yes, for numerical estimates. Many Labs found anchor effects in all 36 samples (Klein et al., 2014). Effects on what people will pay are weaker. A well-powered replication found 3.4% against 31% (Li et al., 2025).

How does anchoring bias affect pricing and loyalty programs?

It gives the first visible number extra weight in judgments of value. Starbucks places a dollar value next to each Star price, which gives the member a reference (Starbucks Rewards Terms). The size of the effect on choices needs a test, because price anchors proved smaller in well-powered studies.