The Power of Leaving Things Alone

The new employee has a good reason to postpone the benefits form. She has several good reasons. She is learning the job, getting access to the systems and trying to remember the names of people who already know hers.

There is a retirement plan to consider. She wants to save. There are also bills due before retirement, which makes the decision less abstract than the brochure suggests. She puts the form beside the documents she intends to read when the week settles down.

Now change one detail. In a second version of the workplace, an employee enters the plan automatically unless she decides otherwise. The same busy week produces a different result.

A default can do that. It changes what happens while attention is elsewhere.

The attraction is easy to understand. A person may have a genuine goal and still fail to take the step that begins it. Making a helpful action the starting point can remove that gap. But choosing a default also means exercising power over people who have not made a fresh, deliberate choice. We should want to know how that power is used.

The starting point can feel like advice

Imagine a phone arriving with its text set at a certain size. You can make it larger, but first you must notice that the setting exists. The original size may feel normal because it is already there. If you have difficulty reading, you may decide the problem is your eyes or your skill with the phone before you think to change the device.

A default does more than save a click. It can suggest what the designer considers suitable.

There are several possible reasons people may leave a setting unchanged. They may like it. They may be busy. They may not know it can be changed. They may think the organization knows better. The route to changing it may be unpleasant. These explanations can produce the same visible result: a setting remains where it began.

That means staying with a default is incomplete evidence of approval.

Suppose a company sees that most customers retain an optional service. It would be too quick to conclude that all of them value it. Some might. Others might not have found the cancellation page. A better investigation would examine whether people understood the choice, could change it without trouble and still wanted the service after being asked in a clear way.

This is also why removing every default would not produce a world without design. Someone must choose the order of the options, the wording of the question and the moment it appears. If a choice is left blank, someone must decide what happens when the person does not answer. Even the demand to make a choice uses time and attention.

We cannot avoid making a process. We can make its influence easier to see and harder to abuse.

Enrollment is a beginning

Retirement saving is a useful test because success has more than one stage. A person can enter a plan, contribute for a while, leave the job, take money out or change saving elsewhere. The number of people enrolled is important. It does not by itself tell us how much additional money they will have available later.

At one large US firm, Brigitte Madrian and Dennis Shea found higher participation after automatic enrollment. Many employees kept the chosen contribution rate and investment fund. Their before-and-after study was not randomized. It showed the importance of the starting arrangement, while leaving open exactly how much came from inertia, perceived advice or other influences.¹

The distinction matters even before we look at a study. Suppose two workers each enter a plan and put in the same amount for a year. One leaves the money untouched. The other withdraws some after losing a job and needing cash. Their first-year participation records could look alike while their later balances differ.

It would be unfair to describe the second worker as simply failing to follow through. The withdrawal may meet an urgent need. It would also be misleading to count money as permanently saved merely because it once appeared in the account.

A sound policy evaluation has to follow what happens, rather than protect the most flattering early number.

Defaults after the deductions

Automatic enrollment

Measure	Value
Scope	Four firms; 2003–2008 introductions; modeled annual effect, percentage points of pay
First-year model	2.20
Five-year data	−0.36
Unvested matches	−0.36
Withdrawals	−0.86
Net model	0.62
Axis	0, 1.25, 2.5
Limits	Sequential adjustments, not isolated causal effects.

Source: Choi et al., Table 1

Figure 1. Following the money changes the estimated payoff.

Choi and colleagues compared nearby hiring cohorts. Their four-firm calculation translates plan savings into a modeled annual rate, assuming repeated job changes and withdrawals and a five-percent return. Later behavior reduces the early extrapolation. These are sequential accounting adjustments, not separately identified causal effects or observed lifetime savings.²

Taha Choukhmane’s 2025 study adds a broader warning. Using evidence from more than one hundred US retirement plans and a representative UK panel, he examines how saving changes over time. In the US plan evidence, early gains attenuated by thirty-six months. The paper reports modest average increases alongside reduced saving inequality. A policy can therefore affect both the amount saved and how saving is distributed, with different answers to each question.³

This leaves room for a more serious conversation about good defaults. They can help people begin. Their value should be tested against the full purpose of the policy, with enough time to see later decisions and with attention to who benefits.

An average result can conceal another question. Does the policy help people who were previously least likely to take part? Does it improve their position enough to matter, even if the average gain is modest? We should ask those questions without quietly changing the outcome whenever a favored policy disappoints. State the goals in advance, measure them separately and report the less impressive findings too.

For a household, the desired outcome may include both future security and enough money for present needs. A policy that improves one while making the other harder deserves careful design, not a slogan about personal discipline.

Who chose the choice?

Now move from a savings plan to a subscription. The stakes and evidence are different, but the design question travels.

A service offers a useful trial. At the end of the trial, the account will move to a paid plan unless the customer acts. The provider knows the date. It has systems able to send messages and take payment. The customer has to keep track of that date among many others.

There may be a fair version of this arrangement: a clear price, a clear end date, a timely reminder and a simple way to stop. There may also be a version that depends on the customer overlooking one step. We should judge the actual path.

Here is a practical moral test I would use. Would the provider still expect the arrangement to succeed if every customer fully understood it and could leave easily?

If the answer is yes, the default may be helping people continue something they value. If the business case becomes weaker mainly because people would stop paying for something they do not want, the problem is much harder to defend.

This is a proposed test, not a legal rule or a proven measure of intent. It helps us inspect incentives without claiming to read minds. A company can have a legitimate interest in predictable revenue. A customer has a legitimate interest in knowing what they are buying and ending an optional purchase without a struggle.

We should not require the customer to discover the provider’s motives before asking for a better cancellation route.

The same test has a public version. Would an agency defend its default if people understood the consequences, had a workable alternative and could challenge a mistake? If the policy depends on people being confused or exhausted, that is a reason to question it. If it protects people while allowing informed revision, that is a stronger defense.

An exit must work when life is busy

An opt-out can exist on paper and still be difficult to use.

Suppose a person can change a setting only by calling during hours when they are working. Or the online switch is easy to find, but the system asks them to confirm through an old address. Or the request succeeds and no confirmation appears, leaving them unsure whether another payment will occur.

Each step may have an explanation. Verification can protect an account. Staff may need working hours. Confirmation messages can fail. The point is to examine the whole route from the person’s intention to the completed change.

A default is more defensible when its exit has been tested by people who did not design it.

The test should start with a realistic task. Ask someone to change the setting, using the instructions a normal user would see. Watch where they get stuck. Do not count a successful outcome only after a staff member tells them the secret. A process that works with expert coaching may need coaching built into the service, or it may need redesign.

Reversibility also has a time dimension. Some choices can be changed tomorrow with little loss. Others create costs that cannot easily be recovered. A default that affects a small display preference can tolerate a different level of friction than one that creates a long commitment or exposes private information.

The more difficult it is to undo the consequence, the stronger the case for making the choice explicit before it takes effect.

This principle will not answer every case. Too many confirmation screens can train people to dismiss all of them. A warning needs to be reserved and written for a decision that deserves attention. Otherwise the system turns a serious safeguard into another routine obstacle.

A default should have a reason you can repeat

Imagine a manager explaining a proposed default to a colleague who will have to answer customers’ questions.

If the explanation takes a page of vague language, the design may not be ready. The manager should be able to say what happens without action, why that is a reasonable starting point and how someone can choose differently. The explanation should include any important cost.

This is not a demand that every policy fit into a sentence. Some arrangements are complex. It is a demand that the core choice remain understandable even when the detail is longer.

A useful notice might tell you that a feature will remain on, what information it uses and where to turn it off. A poor notice might celebrate personalization while leaving the actual setting obscure. Clear language cannot make a bad policy good, but unclear language can make it harder to judge the policy at all.

The duty should also continue after launch. A default chosen for an earlier version of a service may become unsuitable when the service changes. A safe setting for one group may burden another. An organization should decide when it will review the choice, what complaints will trigger review and who has authority to change it.

Otherwise a decision made once can govern thousands of later encounters without anyone having to defend it again.

I would prefer a small, honest review to a grand statement about putting users first. Show what happens if people do nothing. Show how often they change it after understanding it. Show whether the change completes. Then ask whether the outcome is still one the institution can stand behind.

Better defaults require better goals

There is a final trap. We can become so interested in moving a number that we forget to ask whether it is the right number.

A school might want more parents to register for updates. An office might want fewer incomplete forms. A service might want more people to use a security feature. These are plausible goals. Each also needs a reason tied to the person’s experience.

More registered parents may mean little if messages never reach them or are impossible to understand. Fewer incomplete forms may look good because people who cannot finish stop trying. More activated security features may help, but a design that locks people out without recovery creates another problem to solve.

The starting measure is useful when it helps explain the final purpose. It becomes dangerous when it replaces that purpose.

This is where the people affected need a voice. The designer may define success as completing the process. The person may define it as resolving the problem that brought them there. Both should be visible. A completed address change that still sends a crucial letter to the old home is not a successful service from the customer’s point of view.

There is no reason to make every ordinary choice harder in the name of freedom. Thoughtful defaults can spare people repetitive work and help them carry out goals they already hold. The responsibility is to choose those defaults openly, make consequential alternatives usable and follow the effects long enough to learn from them.

Return to the new employee. A good system does not assume that her first busy week reveals her settled priorities. It gives her a sensible starting point, makes its costs and choices plain, and lets her revisit the decision when she understands more.

The institution can help her begin without pretending that beginning is the whole result.

Sources and limits