Goal Gradient Effect

011 min

Takeaways

• What it is: People work faster near the end of a task than at the start of it.
• Origin: Clark Hull, 1932 and 1934, timing rats in a straight runway.
• Human evidence: Kivetz, Urminsky and Zheng, 2006, using a real café coffee card.
• Where it bites: Buying, rushing or overworking to close a reward you did not plan for.
• Guard against it: Price each reward from zero before you start, and review the last items with fresh eyes.

021 min

Where the goal gradient effect shows up

An illustrative case. A platform engineer owns a checklist for moving 40 internal services to a new deployment system. For the first two weeks they move about two services a day. They read each service's config carefully and ask owners questions before they switch anything.

In the final week, with six services left, their behaviour changes. They move four services in one afternoon. They skip lunch. They approve the last two config changes without waiting for the owners to reply. The remaining work is the same kind of work as before. What changed is how much of the list is left.

Now change one input. The same engineer gets a 42-item checklist, and two setup items at the top are already ticked because another team did them last quarter. The real work is still 40 migrations. But the list now opens at about 5% done instead of 0%, and the speed-up starts earlier.

The general pattern: people put more effort into a goal as the remaining share of it gets smaller. That share depends on how the goal is framed, not only on how much real work is left.

032 min

Why the goal gradient effect happens

Hull, a behaviourist, explained the effect through conditioning. In his account, the response of moving toward a reward is reinforced most strongly at the points closest to the reward. Points further back in the sequence receive weaker reinforcement. The result is a gradient: approach behaviour gets stronger with each step closer to the goal.

That account was built for rats. For people, Ran Kivetz, Oleg Urminsky and Yuhuang Zheng proposed a more precise version in 2006, called the goal-distance model. In it, motivation depends on the proportion of the original distance still remaining.

There is also a rational reading of the same behaviour. Near the goal, each extra step removes a larger share of what is left, so its value per unit of effort really is higher. Hyperbolic discounting adds another reason: a reward that is closer in time is valued more. Both explanations predict a speed-up near the end.

The proportional model makes one prediction the rational reading does not. Add the same amount to the total requirement and to the progress already made. The real work left stays the same. But the proportion left gets smaller, so motivation should rise. Kivetz and colleagues called this "illusionary goal progress" and tested it with coffee cards, described in the examples section. The result supported the proportional model.

Why would a mind work this way? One reading is that it puts effort where each unit of effort has the most visible impact. Cryder, Loewenstein and Seltman made this argument for charitable giving in 2013: late contributions feel more impactful. That adaptive account is plausible but not settled. The literature describes the pattern more firmly than it explains it.

041 min

Where the goal gradient effect comes from

Clark Hull stated the goal-gradient hypothesis in Psychological Review in 1932. He predicted "that animals in traversing a maze will move at a progressively more rapid pace as the goal is approached."

Two years later he tested it directly. Hull built a straight runway with electrical contacts placed so that he could precisely measure the time it took rats to cross each of several six-foot sections. The animals ran faster the closer they were to the food reward, and the finding held across several variations of the procedure and the apparatus. Judson Brown extended the work in 1948 by recording how hard rats pulled toward the food when they were stopped close to it or far from it.

For decades almost all of this evidence came from animals. Kivetz, Urminsky and Zheng wrote in 2006 that they knew of no published study giving clear behavioural evidence of the gradient in humans. Their paper was the first large test with real human behaviour.

051 min

Individual effects

For one person, the effect changes a spending or effort decision: whether to act again now, or later. In the Kivetz study, members of a campus café's reward program had to buy ten coffees to earn one free. The study found that participants in a real café reward program purchase coffee more frequently the closer they are to earning a free coffee. The effect held after the researchers controlled for weekly sales and other trends over time.

The error points in one direction: more purchases, sooner. Kivetz and colleagues estimated that, to earn one free coffee, customers bought two more coffees than they would have otherwise. In the authors' reading, the sales lift was larger than the cost of the reward. For the customer, a reward worth one coffee cost two extra purchases.

The same direction of error applies to work. The engineer in the checklist scenario does not lose money. They lose review quality on the last services, which are exactly the ones moved in a hurry.

061 min

Systemic effects

At the level of a whole program, the effect compounds through design. A team that knows effort rises near the goal can move the goal. Kivetz and colleagues pointed out that many reward programs, including American Express Membership Rewards and Hyatt Gold Passport, already give bonus points to new members. They added that managers can raise the point requirement by the same amount as the bonus. That creates the feeling of progress without reducing the real work.

The effect also shapes churn. In the café data, customers slowed down right after earning a reward and were most likely to leave the program at that moment. Kivetz and colleagues concluded that it is particularly important to communicate with and motivate customers immediately after they earn a reward. So a program built on the effect creates a predictable weak point at every reset. Any retention metric that is averaged over the whole cycle hides it.

072 min

Examples

The 12-stamp coffee card

Kivetz, Urminsky and Zheng ran a field experiment with 108 customers of the same café. Research assistants randomly handed out one of two cards. One was a regular 10-stamp card. The other needed 12 stamps but arrived with two "bonus" stamps already on it. Both groups needed ten real purchases.

Customers given a 12-stamp card with two bonus stamps already on it completed the ten required purchases faster than customers given a regular 10-stamp card. On average they finished nearly three days sooner. The researchers also ran questionnaire studies to rule out a sunk cost explanation, in which people might value the bonus stamps as something already invested.

Joseph Nunes and Xavier Drèze published a parallel result in the Journal of Consumer Research the same year. At a car wash, cards that started partly stamped were completed more often than plain cards needing the same number of washes. They named this the endowed progress effect.

An illustrative onboarding checklist

A product manager at a subscription app is deciding how to show the setup steps new users must finish before their first project. The real steps are five. One option shows "0 of 5". The other shows "2 of 7", counting account creation and email confirmation, which every user has already done.

The decision point the definition does not settle: is the second option honest? The two finished steps are real, so it is not a false claim. The cost appears later. If the team also adds two new required steps to "balance" the head start, users get the feeling of progress while doing more work. A fair version counts only steps the user really did.

081 min

A second case: giving to charity

The café and the car wash both paid people a reward. Cryder, Loewenstein and Seltman asked whether the effect appears when nothing is paid back to the person acting.

Their first study used Kiva, a microlending website that shows each borrower's funding progress as a percentage and a progress bar. Using web robot technology in an Internet field study of micro-lending, Study 1 demonstrated that charity contribution rates increase as recipients approach their fundraising goals. The team tracked 209 borrowers every hour for about a week.

Their second study was a mailed appeal to lapsed donors of a real charity. Letters said a fund was 10%, 66% or 85% of the way to its goal, or gave no progress figure. Only the 85% letter changed behaviour: receiving goal progress information only increased contributions in the 85% condition.

The variable that changed between the café and Kiva is who benefits. In the café, the person speeding up gets the coffee. On Kiva, the person contributing gets nothing when the goal is met. The effect appeared in both. So a personal reward is not required. Cryder and colleagues argue the driver is a sense of impact: the last contribution feels like it makes the difference.

092 min

How the goal gradient effect shows up in product, design, and AI

For a designer or product manager, the effect is behind most progress displays: onboarding checklists, profile-completion meters, and stamp-style loyalty screens. Laws of UX, a widely used design reference, states the practical rule this way. "Providing artificial progress towards a goal will help to ensure users are more likely to have the motivation to complete that task." This is deliberate use of the effect. It becomes a dark pattern when the head start is paid for with extra hidden requirements, as in the bonus-point design Kivetz described.

For an analyst measuring a reward or streak feature, the effect changes what a model should include. Kivetz and colleagues modelled purchase timing as a function of proportional distance to the goal, not calendar time. An analysis that only looks at average activity per week will miss both the speed-up before a reward and the drop just after it. A useful check is to plot activity against the share of the goal remaining, and to track churn in the days after each reward.

For anyone reading late-stage surges, the risk is crediting the effect for a pattern with other causes. Kickstarter shows the problem. Venkat Kuppuswamy and Barry Bayus studied 25,058 Kickstarter projects from 2010 and 2011. They found that backer support over time is typically bathtub shaped: busy at the start and the end, quiet in the middle. It is tempting to read the end-of-campaign rise as the goal gradient effect. The authors explained it mainly through deadlines and creator updates, which is covered below.

An AI agent that lists its completed steps is also a progress display. The evidence on this effect comes from rewards, lending and giving, not from agent interfaces. How much the effect changes user patience with an agent is an open question, and a team should test it rather than assume it.

101 min

How to guard against the goal gradient effect

Two techniques change the decision itself rather than relying on awareness.

1. Price the reward from zero before you start. Write down what the full requirement costs in money or hours, counting from nothing. Ignore any head start the program gives you. When you are close to the goal and tempted to speed up, compare the remaining cost with the value of the reward, not with the progress bar. This works because it replaces the shrinking proportion with a fixed number, and the proportion is what drives the speed-up.

2. Set a fixed review rule for the last items. In the checklist scenario, the engineer can decide in advance that every migration gets owner sign-off. The position of a service in the list does not change that. A rule set at the start is not affected by how close the end feels. Teams that ship in batches can do the same with a pre-agreed checklist for the final items of a release.

Knowing about the effect is not enough by itself. Carey Morewedge and colleagues wrote in 2015 that early attempts to reduce decision biases with training met with little success. Researchers turned instead to incentives and to changing how decisions are presented. Both techniques above are changes of that kind. They change how the decision is presented to you, not what you know.

111 min

Common misunderstandings

"Motivation just builds up over time." Many people think the speed-up is habit or momentum. The café data rules this out. Customers who sped up toward their first free coffee slowed down once they started a new card, then sped up again as the second reward came close. A habit would not reset when the card changed.

"Any progress bar speeds people up." The popular version on design blogs suggests showing progress always helps. The evidence is narrower. In the Cryder charity mailing, progress figures of 10% and 66% did nothing measurable. Only the figure near the goal helped. Showing people that they are far away, or halfway, may not help and can hurt, as the next section shows.

Where the effect is useful. Guarding against the effect is not always right. A visible, shrinking amount of remaining work is one of the few reliable ways to get a long task finished. If a team removes progress displays only because they feel manipulative, the likely result is more unfinished work, not better decisions.

121 min

The goal gradient effect vs. nearby concepts

The closest neighbour is the endowed progress effect. It is one specific way to trigger the goal gradient effect: give a head start so the remaining share looks smaller while the real work stays the same.

Sunk cost looks at what is already spent. The goal gradient effect looks at what is left. The 12-stamp experiment separated them.

The Zeigarnik effect is about memory: unfinished tasks are recalled better than finished ones. It says nothing about speed.

Gamification is a design practice, not a bias. It often uses the goal gradient effect through points, levels and progress bars.

132 min

Where the goal gradient evidence is contested

The basic effect in rewards programs is well supported. The contested part is its shape.

Bonezzi, Brendl and De Angelis challenged the idea that motivation rises steadily. In 2011 they showed that motivation can be higher when people are far from or close to the end state, and lower about halfway. In one experiment, students found fewer solutions when a target word was presented fifth in a series of nine than when it came second or eighth. Their explanation is that people measure progress from the start early on, and from the end later. Each frame makes the middle steps feel small. So in their account the goal gradient is only the second half of a U-shaped curve.

A second challenge is about cause. In the Kickstarter data, Kuppuswamy and Bayus explained the late rise in support mainly through reduced diffusion of responsibility near the deadline, the process behind the bystander effect. They also found that potential backers are much less interested in supporting a project after it reaches its funding goal. That drop fits the goal gradient effect, but deadlines and goal closeness arrive together, so one dataset cannot separate them.

What to do while this is unsettled: expect a slump in the middle of long goals, not only a rise at the end. Test a to-go frame ("3 left") against a to-date frame ("7 done") instead of assuming one is always better.

?7 questions

Questions people ask

§9 sources

Sources