Does Making Something Easier Actually Make People Do It?
What happened when governments and companies digitized required tasks, reminded people to act, or removed the need to act at all.
Making a task easier usually changes how people complete it. It changes completion when it solves the thing that was actually stopping the person.
Four things get called reducing friction, and they behave differently
Making a task easier usually changes how people complete it. It changes completion when it solves the thing that was actually stopping the person, and that is often not the difficulty of the task.
Enormous channel migration, lower cost, fewer errors, sometimes faster completion. Total completion often barely moves.
This can move completion even when the path has been sitting there for years.
Some of the largest effects in the historical record sit here.
The largest effects in the historical record sit here, along with the clearest risk that higher completion can come with a worse underlying choice.
These are not rungs on a ladder. The populations and interventions differ too much for that. They are a recurring pattern in what actually moved and what did not.
Everything about filing taxes changed except the filing
In 2001, about 31 percent of American individual tax returns were filed electronically. By 2010 it was almost 70 percent. By the end of 2023 the IRS reported 92.6 percent.
Two different measures over overlapping periods. Shown side by side to give the scale of the channel change, not as a statistical model of cause and effect.
That is one of the largest interface changes ever imposed on a population. Tens of millions of people stopped assembling paper, stopped buying stamps, stopped driving to the post office in the second week of April. The refund arrived in days instead of weeks. The error rate on a paper return runs around 20 percent. On an electronically prepared and filed return it is under 1 percent.
The IRS also tracks something called the voluntary filing rate, which is the share of required returns actually filed on time. Built from its own administrative records, that series reads 88.8, 90.5, 89.7, 88.6, 88.1, 88.7, 88.3, 88.3, 88.3, 88.5 across tax years 2006 through 2015. Over that decade electronic filing went from roughly half of returns to roughly six in seven.
The series moves once. Tax year 2007, up about 1.7 points, because the Economic Stimulus Act required people to file a return in order to receive a rebate check.
Money and a deadline moved it. The channel did not.
The dollar-weighted version tells the same story from another angle. The voluntary compliance rate was 83.7 percent for tax year 2001 and 85.0 percent for tax year 2022, with everything in between inside a narrow band. A quarter century of digitization sits inside that band.
If making something dramatically easier does not reliably make more people do it, what does?
The case where friction really was the problem
In 2008, researchers ran an experiment inside 156 H&R Block tax offices in Ohio and North Carolina. Low and moderate income families who had just finished their taxes were sorted into three groups.
One group got a personalized estimate of how much college financial aid they qualified for, along with tuition prices at four nearby colleges. Real numbers, about their own family, handed to them at the exact moment their tax return was on the screen in front of them.
A second group got all of that plus help. The software pulled their tax data into the FAFSA, a preparer walked through the remaining questions in under ten minutes, and H&R Block offered to transmit the completed form to the Department of Education.
The third group got a brochure.
Among independent adults with no prior college, 13.8 percent of the control group filed a FAFSA. In the assisted group, 39.5 percent did. Among dependent high school seniors, filing went from 40.2 percent to 55.9 percent, and college enrollment that fall went from 26.8 percent to 34.5 percent. People filed months earlier. Independent adults in the treatment group submitted their applications an average of 113 days sooner.
The information-only group did not move at all.
So the FAFSA case is the counterweight to the tax case. Sometimes the form really is the barrier. The 2008 FAFSA ran eight pages and over 100 questions, with worksheets carrying nearly forty more. It was longer than the 1040, and the IRS estimated the 1040 at thirteen hours. That is not a click. That is a project, and removing it changed what people did.
But the most useful number in the study is buried inside the treatment group.
Some of those families finished the form in the office and then chose to take a paper copy home rather than have H&R Block transmit it. Everything had been done for them. The questions were answered, the math was done, the form was complete. All that remained was putting it in an envelope.
This subgroup chose its own path. It is not a randomized comparison and should be read as a pattern rather than a causal estimate.
They filed at 13.5 percent. The control group filed at 13.8 percent.
That subgroup chose its own path, so it is not a randomized comparison and the authors say so. Read it as a pattern rather than a causal estimate. It is still the sharpest thing in the study, and it is corroborated by the arithmetic: of the 23.5 point increase in filing among independent adults, about 90 percent came from returns H&R Block sent electronically.
Almost done can behave like not started when the person still owns the final step.
The word is covering four different problems
Every organization that says it wants to reduce friction is describing at least four different failures, and they do not respond to the same treatment.
Not aware
The person does not know the task exists, or does not know that now is the moment.
Heavy execution
The person knows perfectly well and is facing an execution burden that is genuinely heavy.
Last step
The person gets almost to the end and still has to perform a final action they own.
Incentive
The person knows exactly what to do, has an easy way to do it, and chooses not to because completing the task costs them something.
The person may not know the task needs to happen, or may not know that now is the moment. The person may know perfectly well and be facing an execution burden that is genuinely heavy. The person may get almost to the end and still have to perform a final action they own. Or the person may know exactly what to do, have an easy way to do it, and be choosing not to because completing the task costs them something.
A single word covers all four. A single intervention almost never does.
The rest of this is what happened historically when organizations solved each one.
Operational improvement is not the same as behavioral change
Vehicle registration renewal is the purest example. California has offered online renewal since around 2000. Two decades later, in fiscal 2019-20, 37 percent of renewals happened online, 20 percent by mail and 13 percent in a field office. The offline channels combined still outweighed the online one on a task that takes about four minutes and involves no documents for most people.
Telehealth is the most recent and most dramatic version. It removed travel, parking, time off work, childcare and the waiting room from a large share of outpatient care, and it did so in a matter of weeks in 2020. A University of Michigan analysis covering 539 million appointments found total outpatient office visit volume flat or falling through mid-2024. A claims study of 660,546 patients found that telehealth users had fewer visits than non-users everywhere the researchers looked, including 0.44 fewer primary care visits and 0.11 fewer emergency department visits. A three-system study of 4.1 million visits found average visits per patient stable across commercial, Medicare and Medicaid populations.
The largest reduction in the cost of seeing a doctor in modern history, and people mostly saw the same doctors the same number of times through a different window.
What digitization reliably improved
- Cost per transaction
- Error rates
- Processing and refund speed
- Convenience and access
- Staffing and physical infrastructure
What it often did not move
- Total completion
Both kinds of win are worth having. Only one of them shows up in a compliance or take-up rate.
None of that makes the digitization a failure. The IRS eliminated an entire transcription workforce and cut the error rate by a factor of twenty. California DMV moved a third of forty million transactions out of buildings it has to heat and staff. Telehealth lowered the cost per visit and gave rural and homebound patients real access. Those are large, durable, compounding wins.
Both kinds of win are worth having. Confusing them is how a company ends up surprised that a beautiful new portal did not move the number the board asked about.
Two footnotes worth carrying forward. The first is what happens to the people who do not complete. In IRS data, late filers fell as a share of required returns between 2006 and 2014 while permanent non-filers rose. The middle of the distribution got faster. The tail got harder.
The second is the IRS Free File program. Since 2002, roughly 70 percent of American taxpayers have been eligible for free, guided, online tax preparation. About 3 percent of them use it. Part of that is documented suppression by the software companies in the program. Part of it is that nothing tells anyone the option exists at the moment they are deciding how to file. Free, online and easy has been sitting there for twenty-two years.
Utah cut the postcards and kept the website
In September 2020, Utah stopped mailing postcards reminding people that their vehicle registration was about to expire. The state was cutting costs during the pandemic and the postcards ran about $900,000 a year. Online renewal was untouched and worked exactly as before.
COVID and reduced driving are real confounds. The exact effect attributable to the reminder is unresolved.
In the first month without postcards, 30.9 percent of vehicles were renewed before their expiration date. When the postcards were going out, the figure was 58.54 percent.
Non-driving during the pandemic is a real confound, and the state tax commissioner named it alongside forgetting without splitting the two. The precise size attributable to the reminder is unresolved and should stay that way.
The direction is not in doubt, and the reaction settles it. Emissions shops reported customers arriving upset. The Utah House voted 72 to 1 to bring the postcards back. The governor signed the bill on March 11, 2021, and the first restored cards went out in April. The whole episode lasted seven months.
Hold the two Utah numbers next to each other. A mandatory annual task, with a hard deadline, where the penalty is losing the legal right to drive, with a working online renewal system available to everyone, ran at 58.54 percent on time. Without the postcard, the observed on-time rate was roughly half as high.
California, notably, wrote the trigger into law. The vehicle code requires the DMV to notify owners at least sixty days before the renewal date, and if it fails to, the delinquency clock changes. The state that processes 68 million transactions a year does not treat the reminder as an optional courtesy.
The reminder evidence outside the DMV is consistent. The IRS mailed a second notice to 35,050 people who had already received one notice about unclaimed Earned Income Tax Credit money and had not claimed it. Twenty-two percent of them responded to the second letter. Covered California mailed deadline reminders to 87,000 households and raised enrollment by 1.3 points, a 16 percent relative increase in a population with famously low take-up. In healthcare, a text message reminder cut non-attendance from 18.1 percent to 11.2 percent in one hospital department, and a test across 161,587 members of an Israeli health plan found that changing only the wording of the reminder moved no-shows from 21.1 percent to 14.2 percent.
None of those interventions touched the execution path. The appointment was already booked. The credit was already claimable. The form was already online.
An execution path with no trigger is a website.
What remains in the way is not always the whole task
Sometimes the remaining barrier is not the whole task. It is whatever is left at the end: a heavy form, a synchronous appointment, or one final action nobody counts as friction. The cases that moved completion are the ones where someone identified the specific thing still in the way and took ownership of it.
Heavy task
The 2008 FAFSA. Eight pages, over 100 questions, longer than the 1040.
Synchronous step
The SNAP intake interview, scheduled by the county, that the applicant had to be available for.
Final action
Submitting. Transmitting. The envelope that never gets posted.
Online voter registration gives the cleanest separation of the layers, and it points the opposite way from FAFSA.
In 2010, the year eight states first launched registration portals, researchers ran a field experiment across twenty-five colleges and more than 130,000 unregistered students. One group got an email linking to the downloadable paper form. One got an email linking to the new online portal. The control group got nothing, and had the portal available the entire time.
Effects within a single experiment on one population. Not comparable to the FAFSA figures above, which come from a different study, a different task and a different sample.
The paper form email raised registration by 0.6 points. The portal email raised it by 1.2 points and turnout by 0.5. The control group produced the baseline.
The portal sitting there, free and available and taking four minutes, did nothing on its own. Pointed at, it doubled what a paper form achieved.
Set that beside FAFSA, where personalized information delivered at exactly the right moment did nothing and the execution help moved everything. The two studies disagree, and the disagreement is the finding.
The two studies disagree, and the disagreement is the finding.
Registering to vote takes a few minutes. Filing a 2008 FAFSA took hours and required documents most families had not gathered. When the execution burden is light, the trigger is usually the scarce thing and execution improvements act as a multiplier on it. When the execution burden is genuinely heavy, execution can become the binding constraint and no amount of reminding fixes it.
A trigger with no execution path is a nag. An execution path with no trigger is a website. Which one you are short of depends on how heavy the task actually is, and most organizations have never asked.
A trigger with no execution path is a nag.
An execution path with no trigger is a website.
There is a third version, and it is the one that behaves differently from both.
Los Angeles County ran an experiment with 65,000 SNAP applicants. The barrier was not a form. It was an interview, scheduled by the county, that the applicant had to be available for. Letting applicants initiate the interview themselves raised approvals by 6 points, doubled early approvals, and raised long-term participation by more than 2 points.
That is not a lighter form. It is the removal of a step that required the person to be somewhere at a time they did not choose. The FAFSA transmission is the same category. So is a lot of what looks superficially like convenience.
Why defaults behave differently
The largest completion effects in the record often came from systems that stopped asking the person to complete the task at all.
Self-service
- Makes the task easier for the person.
Automatic
- Removes the need for the person to act.
Automatic enrollment took 401(k) participation from 49 percent to 86 percent at the company Madrian and Shea studied, with no change to any economic feature of the plan. Automatic voter registration produced registration increases reported between 9 and 94 percent across adopting states, and electronic transmission of DMV records roughly doubled registration rates at motor vehicle offices in Kansas, Rhode Island and Washington. Credit card autopay users show delinquency around 6 percent against 17 percent for people paying manually, and the one plausibly causal design in that literature finds large chargeoff reductions on one identification strategy and weaker results on a second, so treat the magnitude loosely and the direction as solid.
One company, studied by Madrian and Shea. Not every automatic system produces this tradeoff. It is a warning about what to measure, not a law.
The temptation is to file these as very good self-service. That is exactly the mistake. Grouping defaults with digitization is what makes the friction thesis look better supported than it is. These systems did not make the task easier for the person. They took the person out of the loop.
And they came with a bill.
At the company Madrian and Shea studied, 76 percent of automatically enrolled participants stayed at the 3 percent default contribution rate, and the average contribution rate among participants fell from 6.4 percent to 4.4 percent. Far more people were saving. They were saving less, and they were saving at a number an administrator picked.
Completion and quality are different outcomes and they can move in opposite directions. A system that completes things on someone's behalf can raise completion while lowering the quality of the underlying choice, and the second effect is much harder to see on a dashboard.
When the person is not confused, just unwilling
There is a fourth constraint that none of this touches, and the tax data shows it plainly.
Sometimes the barrier is not awareness and not execution. It is that completing the task has an actual cost.
American taxpayers who are owed a refund file on time at a rate roughly twenty points higher than taxpayers who owe a balance. Same system, same forms, same deadline, same reminders, same software. People with only wage income file about thirty points above people with only self-employment income, whose earnings are largely invisible to third-party reporting.
And the single strongest predictor in the whole dataset has nothing to do with the interface. People who filed on time last year file on time at 96 to 97 percent. People who did not run somewhere between 35 and 50 percent. That gap is larger than any channel effect anywhere in this study, and it held roughly steady across the decade in which the channel changed completely.
The corresponding evidence from the EITC experiment is worth sitting with, because it looks like an execution result and is not. What raised claiming was a shorter, clearer notice and displaying the dollar amount at stake, which added 8 points. What did nothing at all was the arm designed to reduce the perceived cost of applying, along with the arms addressing stigma and audit fear. The simplification that mattered was about whether people understood what was being offered.
Sometimes people know exactly what to do, have an easy way to do it, and are choosing not to because completing the task means writing a check, revealing income, or accepting an outcome they would rather defer. Convenience is not a lever on that. Neither is a reminder. Treating an incentive problem as a friction problem is one of the most expensive category errors an operating team can make, because the intervention is cheap, the theory is flattering, and the number does not move.
Treating an incentive problem as a friction problem is one of the most expensive category errors an operating team can make.
The industry is building better hands
Almost every task currently described as an obvious opportunity for AI agents was digitized between fifteen and twenty-five years ago. Filing a return, renewing a registration, booking an appointment, paying a bill, enrolling in benefits, registering to vote. All of them became doable from a couch, in minutes, without speaking to anyone. The completion rates mostly did not move.
What many agent products are improving
- Execution
- Form filling
- Clicks
- Navigation
What history suggests may be scarcer
- Detection
- Timing
- Context
- Authority
- Completion through confirmation
That should at least raise the question of whether a great deal of current investment is aimed at the layer with the weakest historical track record. The industry is building better hands. History suggests the scarce thing has more often been knowing when to use them.
The industry is building better hands. History suggests the scarce thing has more often been knowing when to use them.
The layers that did move behavior look like this. Knowing an obligation exists at all. Detecting that action is now required. Surfacing it at the moment it can be acted on. Knowing what information the task needs and having it already. Doing the work. Then carrying it through the final confirmable step rather than handing back something ready to submit.
That last one deserves emphasis because it is the part most current products get wrong. A prefilled form awaiting approval is the FAFSA that went home in a paper envelope.
Now the caveat, and it is a serious one.
Every large completion effect came from a counterparty already inside the transaction
- An employer
- A bank
- A government agency
- A health plan
- A tax preparer
Every large completion effect in this record came from a counterparty already inside the transaction. An employer running the payroll. A bank holding the account. A state agency that issues the registration. A health plan that pays the claim. A tax preparer the family was already paying and sitting in front of.
There is no historical analogue for an independent agent operating across institutions on a person's behalf. Whether the FAFSA result survives when the party pressing send is not H&R Block is unknown, and nothing in this evidence answers it. Trust and authority were not variables in these experiments because they were already settled before the experiments began. For a third-party agent they are the entire question, and it is open.
There is no historical analogue here proving that an independent third-party agent operating across institutions will get the same result.
Five questions worth asking
These are not a validated framework. They are the questions this history keeps raising.
Do people already know the task needs to happen, and that now is the time?
If a mailed notice, a deadline, a statement or a symptom is already doing that job, an easier interface has very little room to work. If nothing is doing that job, the interface is not your problem.
Is execution actually what is stopping them?
Compare the real weight of the task to the FAFSA at one end and voter registration at the other. Four minutes of typing is rarely the binding constraint. Six hours of document gathering often is.
Does your process end in completed, or in ready for the person to complete?
Find the last step the customer owns and measure the drop-off across it specifically. That number is usually worse than anyone expects and it is usually not on a dashboard.
Can the task happen automatically, or by default?
If it can, expect a larger completion effect than self-service alone, and measure whether the quality of the underlying choice changes with it.
Is there an economic or trust barrier that convenience cannot touch?
If the non-completers are systematically the people who owe money, or who would rather not be visible, or who do not trust the party asking, then the redesign will produce a cleaner experience for people who were already going to complete.
The web changed how obligations get discharged more than who discharges them
The web made an enormous number of obligations possible to discharge from anywhere, at any hour, in minutes. That was worth doing. It removed cost, removed error, removed buildings, removed waiting.
What it mostly did not do is change who discharges them.
The people who file, renew, enroll, claim and attend are, to a striking degree, the same people who did so before, doing it faster and more accurately through a different window. The people who do not are held back by something else. They do not know it is time. They are facing something genuinely heavy. They are one action short of done and that action never happens. Or they have looked at the cost of completing and quietly decided against it.
History suggests the next bottleneck is often no longer how to do the thing. It is knowing something needs to happen, knowing when, identifying what is actually stopping it, and getting it all the way to done.