Richard Thaler and Shlomo Benartzi’s 2004 Save More Tomorrow paper is the rare behavioral economics intervention that produced a measurable, scalable, decades-durable real-world result. Field-tested at Ispat Inland Steel, codified into the 2006 Pension Protection Act, and now standard practice across U.S. corporate retirement plans. Here is why it worked when so many other behavioral interventions did not.
If you have been reading this hub long enough to recognize the pattern, you already know what is coming. Most of the canonical mid-2000s behavioral-economics findings cataloged elsewhere in this collection did not survive scrutiny. Ego depletion did not replicate. Money priming did not replicate. Power posing did not replicate. The willpower-as-glucose hypothesis was dismantled. The original stereotype-threat magnitudes shrank to confidence intervals that straddle zero. Even the more careful findings --- the affect heuristic, the framing effect, the availability heuristic --- have survived only with substantial revision of the boundary conditions and the applied implications.
Save More Tomorrow is different. It is a behavioral-economics intervention that works. It works in the field, not just in the lab. It produces an effect size large enough that the policy implications are genuinely actionable. It has been adopted at scale by employers and codified into federal pension law. And the mechanism is well-understood enough that the boundary conditions are predictable, not surprising. This article is an anti-example in the same sense that the Defaults article in this hub is an anti-example --- a finding that did not collapse, a finding that strategists and policymakers can rely on without the qualifiers and asterisks that most behavioral-economics interventions require.
The reason Save More Tomorrow is worth a separate hub entry rather than just a mention in the Defaults piece is that it is a different mechanism stacked on top of defaults --- it combines four behavioral principles into a single intervention, and the combination is what produces the effect size. Understanding why the four principles compound the way they do is the practical lesson for anyone trying to design a behavioral intervention for retirement planning, benefits design, or any other domain where the goal is to get a population to do something today that pays off in the long run.
The Thaler-Benartzi 2004 Paper
The founding paper is Thaler, R. H., & Benartzi, S. (2004). “Save More Tomorrow: Using Behavioral Economics to Increase Employee Saving.” Journal of Political Economy, 112(S1), S164—S187. DOI: 10.1086/380085.
The starting point of the paper is the well-documented puzzle that U.S. household saving rates are persistently below what a standard life-cycle model would predict given the income trajectory of a typical worker. People say they want to save more for retirement. They report intending to save more for retirement. They acknowledge that their current saving rates are inadequate. And then they do not raise their contribution rates. The intention-action gap on retirement saving is one of the most consistent findings in household finance, and it had defeated the standard policy interventions for decades. Financial education programs produced modest knowledge gains and essentially no behavior change. Matching contributions produced participation increases at the participation margin but very small contribution-rate increases among those already participating. The structural problem was not that workers did not know they should save more; the structural problem was that workers were not willing to absorb the perceived loss of take-home pay required to raise contribution rates today.
Thaler and Benartzi proposed a different design. Instead of asking employees to raise contributions today, the Save More Tomorrow plan asked employees to commit today to raising contributions automatically in the future, with the increases scheduled to coincide with future pay raises. The mechanics were specific: an employee enrolled in the SMarT plan signed up at a planning meeting, accepted a default contribution rate, and committed to having that contribution rate increase by a fixed amount --- typically 3 percentage points --- at each future pay raise, until the contribution rate hit a predetermined cap. The employee could opt out of any individual increase at any time, but the default was that each scheduled increase would happen automatically.
The design exploits four well-documented behavioral principles simultaneously, and the simultaneity is the reason the effect size is so much larger than the effect size produced by any one principle in isolation.
The Four Behavioral Principles Built In
The first principle is hyperbolic discounting. People discount future outcomes much more steeply than the standard exponential-discounting model predicts, and the steepness of the discount in the near term is greater than the steepness of the discount in the far term. A consequence of this is that people are far more willing to make sacrifices in the distant future than to make the same sacrifice today --- a phenomenon sometimes called “preference reversal” or the “want-should gap.” Asking an employee to raise their contribution rate today triggers the immediate-loss heuristic and is refused. Asking the same employee to commit today to raising their contribution rate at some future date triggers the future-loss heuristic, which is sharply discounted, and the commitment is accepted. The SMarT plan converts the rejected immediate-action ask into the accepted future-commitment ask, and exploits the asymmetry between the two evaluation modes.
The second principle is loss aversion. Kahneman and Tversky’s prospect theory predicts that losses are weighted approximately twice as heavily as equivalent gains in subjective evaluation. A worker whose take-home pay drops by 3 percent because their 401(k) contribution went up by 3 percent experiences this drop as a loss, and the loss is weighted heavily in their subjective accounting. The SMarT design times the contribution increase to coincide with a pay raise, so the worker’s take-home pay still goes up at the moment of the contribution increase --- just by less than it would have if the contribution rate had not increased. There is no felt loss because the take-home pay is rising, not falling. The contribution increase is invisible against the backdrop of the raise. The loss-aversion barrier that defeats the standard “raise your contribution rate today” ask is structurally avoided.
The third principle is status quo bias. Once an employee has accepted a default arrangement --- an opt-out arrangement rather than an opt-in arrangement --- the default tends to persist, and the cumulative friction of opting out, scheduling an HR meeting, filling out the forms, and revising the contribution rate is enough to keep most participants in the default. The SMarT plan defaults each scheduled increase to “happen automatically unless you opt out.” The default direction is in favor of the increase, and the cumulative friction of opting out is small at each individual decision point but large over the four-to-six-year arc of a typical SMarT enrollment. Status quo bias preserves the trajectory of automatic increases that the employee originally committed to.
The fourth principle is the cap-based termination of the escalation. The SMarT plan does not escalate indefinitely; it escalates until the contribution rate hits a predetermined cap, typically in the range of 13—15 percent of pay. This is important both for the worker’s planning --- they know the increase has a known endpoint --- and for the behavioral commitment --- the worker is not committing to an indefinite future of contribution increases that could in principle eat their entire paycheck. The cap converts the open-ended commitment into a bounded commitment, which makes it psychologically tractable to accept.
The reason the SMarT plan works where individual behavioral interventions fail is the stacking. Hyperbolic discounting alone gets the worker to accept the future commitment. Loss aversion alone, applied at the moment of each increase, would defeat the increases if the increases were felt as losses. The timing-to-coincide-with-raises construction prevents the increases from being felt as losses, so the loss-aversion barrier does not activate. Status quo bias alone would not move workers from a default of low contributions to a default of higher contributions; the active enrollment in the SMarT plan at a planning meeting establishes the new default. The cap alone is just a planning convenience, but it makes the future commitment bounded and therefore acceptable. Each principle handles a different part of the choice architecture, and the absence of any one of them would degrade the effect.
The Ispat Inland Steel Field Trial
The empirical core of the 2004 paper is a field trial conducted at an Ispat Inland Steel manufacturing facility, where the SMarT plan was offered to approximately 315 employees whose contribution behavior had been studied across the preceding several years. The setting matters: this was a real workplace, not a laboratory, and the population was real workers making real retirement-saving decisions about their actual pay. The intervention was rolled out as an opt-in offering at a financial-planning meeting; workers who attended the meeting and accepted the SMarT offer were enrolled in the auto-escalation program.
The contribution-rate trajectory of the SMarT enrollees is the headline finding of the paper. At the start of the program, the SMarT participants were contributing at an average rate of approximately 3.5 percent of pay. Over the following 40 months, across four scheduled pay raises and four scheduled contribution-rate increases, the average contribution rate of the SMarT participants rose to approximately 13.6 percent of pay. The increase was concentrated in the automatic-escalation events; the participants did not opt out of the increases at significant rates, despite the standing offer to do so. The status-quo-bias mechanism held, and the cumulative effect was an approximately 10-percentage-point rise in contribution rate over the three-and-a-half-year window.
The control comparisons in the paper are important for interpreting the magnitude of the effect. Workers who attended the financial-planning meeting but chose not to enroll in SMarT --- a group that had self-selected as being interested enough in retirement planning to attend the meeting but not committed enough to enroll in auto-escalation --- showed essentially flat contribution rates over the same window. Workers who did not attend the meeting at all also showed essentially flat contribution rates. The 10-percentage-point rise in the SMarT group was not driven by some general upward trend in saving behavior at the facility; it was specifically caused by the auto-escalation mechanism, with the meeting-attendee-non-enrollees serving as the cleanest comparison.
A few additional features of the field-trial results are worth noting because they bear on the generalizability of the finding. First, the SMarT participants were on average lower-income workers than the broader employee base at the facility --- the very population for whom retirement-saving inadequacy is most consequential and for whom the traditional financial-education interventions had been least effective. The fact that the SMarT mechanism produced large effects in this subgroup is not a story about high-income knowledge workers optimizing their tax-advantaged accounts; it is a story about a real intervention working on a population that had been resistant to all previous interventions. Second, the cumulative four-cycle effect was disproportionately concentrated in the first three cycles --- by the fourth cycle, the contribution rate was approaching the program’s cap, and the marginal effect of each additional cycle was diminishing. This is what the design predicts: the cap is the endpoint, and the trajectory tapers as it approaches the endpoint. Third, the opt-out rate at each individual escalation event was modest --- the workers who had originally committed to the auto-escalation generally honored the commitment, validating the prediction that status quo bias would preserve the trajectory once it had been established.
The Ispat Inland Steel trial is small in absolute terms --- approximately 315 enrollees over 40 months is not a population-level study --- but the effect size is large enough and the comparison-group structure is clean enough that the finding has held up under subsequent re-examination. The trial was one of the first behavioral-economics interventions to be implemented as a working policy at a real employer and studied longitudinally with proper controls, and the empirical record has not seriously been challenged.
Generalization to Subsequent Implementations
The 2004 paper closed with a discussion of how the SMarT mechanism might be generalized to other employers, and the subsequent decade of empirical work has substantially answered that question. The most important follow-up paper is Benartzi, S., & Thaler, R. H. (2007). “Heuristics and Biases in Retirement Savings Behavior.” Journal of Economic Perspectives, 21(3), 81—104. DOI: 10.1257/jep.21.3.81.
The 2007 JEP review paper aggregates the evidence from the SMarT plan and from a range of related auto-enrollment and auto-escalation implementations across other employers. The pattern that emerges across the broader sample is consistent with the original field-trial result. Auto-escalation plans modeled on SMarT produce large increases in employee contribution rates relative to opt-in plans, with the effect concentrated in the first several years of escalation and tapering as the cap is approached. The opt-out rates at scheduled escalation events are consistently modest, validating the status-quo-bias prediction. The pattern of larger effects among lower-income workers and among workers who had not previously responded to traditional financial-education interventions also generalizes across the broader implementation sample.
The other important strand of generalization evidence is the work coming out of the Beshears-Choi-Laibson-Madrian retirement-savings research program, which has produced a substantial body of empirical work on retirement-saving choice architecture over the 2010s and 2020s. The relevant paper for this hub entry is Beshears, J., Choi, J. J., Laibson, D., & Madrian, B. C. (2017). “The Effect of Providing Peer Information on Retirement Savings Decisions.” Journal of Finance, 70(3), 1161—1201. DOI: 10.1111/jofi.12258. This paper is part of a larger research program that has documented the persistent effectiveness of auto-enrollment and auto-escalation across a wide range of employer settings, with the magnitudes of the effects roughly consistent with what the original SMarT trial predicted. The Beshears-Choi-Laibson-Madrian program has also documented some refinements --- such as the conditions under which peer-information disclosure helps or hurts retirement-saving behavior --- but the core finding that auto-escalation produces large and durable effects on contribution rates has not been seriously challenged in this literature.
The longer-arc consolidation of the SMarT mechanism into the broader literature is captured in Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving Decisions About Health, Wealth, and Happiness. Yale University Press. The Nudge framework codifies the SMarT design as one of the canonical examples of choice architecture --- an intervention that does not restrict the worker’s freedom to choose but reshapes the default in a way that improves outcomes for the workers who do not actively opt out. The book’s broader popularization of the SMarT mechanism was an important part of why the policy adoption arc moved as quickly as it did.
The Pension Protection Act of 2006
The most consequential downstream effect of the Thaler-Benartzi 2004 paper is the codification of SMarT-style auto-enrollment and auto-escalation into U.S. federal pension law via the Pension Protection Act of 2006. The PPA created a statutory safe harbor for employers offering 401(k) plans that included an auto-enrollment default and an auto-escalation schedule conforming to certain minimums. Employers who adopted plans meeting the safe-harbor criteria were granted relief from certain non-discrimination testing requirements and certain fiduciary exposures associated with the auto-enrollment design. The effect of the safe-harbor structure was to make it substantially easier and lower-risk for employers to adopt SMarT-style designs in their 401(k) plans, and the subsequent adoption curve was steep.
The post-PPA adoption pattern is what would be expected from an intervention with the SMarT effect size. According to industry surveys conducted by the Profit Sharing Council of America (now the Plan Sponsor Council of America) and other industry research organizations across the late 2000s and the 2010s, the share of U.S. 401(k) plans incorporating an auto-enrollment default rose from roughly a quarter of plans pre-PPA to a majority of plans by the mid-2010s, and the share incorporating an auto-escalation schedule rose similarly. By the late 2010s, auto-enrollment and auto-escalation had become the de facto default architecture for new 401(k) plans, with the SMarT design as the most common template for the escalation schedule. The actual contribution rates of U.S. private-sector retirement savers rose meaningfully over the same window, and the consensus in the retirement-research literature is that a substantial share of the rise is attributable to the SMarT mechanism and its descendants.
This is the policy-adoption arc that almost no other behavioral-economics finding has produced. The intervention was identified in a laboratory-adjacent field trial, generalized in subsequent empirical work, codified into federal statute four years after the founding paper, and adopted as standard practice across a multi-trillion-dollar retirement-savings system in the decade after the statutory change. The intervention has been working in the field continuously since the original Ispat Inland Steel trial in the late 1990s, and the cumulative effect on national household saving is large enough to be visible in aggregate retirement-asset statistics.
Why This Works When Most Behavioral Interventions Do Not
The question worth answering, given the broader replication-crisis context of this hub, is why SMarT has worked at this scale when so many other behavioral-economics interventions have not. The answer has several components, and the components together constitute the closest thing to a diagnostic checklist for distinguishing a behavioral intervention that will scale from a behavioral intervention that will collapse.
First, the SMarT design rests on a stack of well-replicated underlying findings, not a single contextually-thin laboratory effect. Hyperbolic discounting, loss aversion, and status quo bias are each among the most robust findings in behavioral decision research, with effect sizes that have replicated across decades of follow-up work. The intervention is engineered to compose these findings rather than to depend on a single fragile effect. When a behavioral intervention rests on a single laboratory finding whose effect size collapses under replication --- ego depletion, power posing, money priming --- the intervention collapses with it. When the intervention rests on a stack of robust findings, the intervention inherits the robustness of the stack.
Second, the SMarT design exploits the structural features of a specific real-world choice environment --- the U.S. corporate 401(k) plan with periodic pay raises --- rather than trying to produce a behavior change in an abstract decision task. The intervention works because pay raises happen periodically, contribution rates can be set as percentages of pay, automatic payroll deductions are technically straightforward to implement, and the choice architecture of the 401(k) enrollment process allows for the explicit commitment-with-default-escalation design. A behavioral intervention that requires real-world choice architecture to instantiate cleanly is more likely to scale than one that requires careful experimenter framing to produce its effect.
Third, the magnitude of the SMarT effect at the individual level is large enough that small-magnitude implementation noise does not destroy the policy case. A 10-percentage-point increase in contribution rate over 40 months is not a 5-percent shift detectable only in a 10,000-person study; it is a transformation of the individual retirement trajectory. Interventions whose effect sizes are small enough that they require enormous samples to detect are also interventions whose effect sizes are small enough that they get lost in the noise when implemented at scale. Interventions whose effect sizes are large at the individual level survive implementation noise.
Fourth, the SMarT design produces an effect that is directly observable in a hard, easily-measured outcome --- the contribution rate on the worker’s pay stub --- rather than an effect that is observable only in a downstream attitudinal or self-reported variable. The intervention does not work because it changes how workers feel about retirement saving; it works because it changes the percentage on the deduction line. When the outcome variable is hard and measurable, the intervention either works or it does not, and the question is settled empirically rather than rhetorically. Interventions whose primary outcomes are softer --- attitudes, self-reported intentions, knowledge gains --- can survive in the literature even when they do not produce behavior change, because the soft outcomes can shift even when the hard outcomes do not. SMarT was tested on the hard outcome and survived.
Fifth, the SMarT design is conservative about its own mechanism. The paper does not claim that auto-escalation will solve the entire retirement-savings adequacy problem; it claims that auto-escalation will produce a measurable contribution-rate increase among the subpopulation that enrolls. The claim is appropriately bounded, and the subsequent literature has confirmed the bounded version. Behavioral interventions that overclaim --- that promise to solve large policy problems with small choice-architecture tweaks --- tend to collapse when the overclaim runs into the empirical evidence. SMarT was modest about its own mechanism and has been validated within the modest claim.
The cumulative implication is that SMarT works because it was engineered as a working policy, not as a laboratory demonstration. The four behavioral principles were each known to be robust; the composition was designed to exploit the specific choice-architecture features of corporate retirement plans; the effect size was targeted to be large enough at the individual level to survive implementation noise; and the outcome variable was a hard, easily-measured number rather than a soft, easily-shifted attitude. Each of these design choices is one that a strategist or policymaker can carry over to other domains.
Implications for Retention, Benefits Design, and Behavioral Strategy
For the strategist designing retention, benefits, or compensation architecture, the SMarT case study has several specific implications that go beyond “auto-escalate your 401(k) contributions.”
The first implication is that the right test of a behavioral-design idea is not whether it produces an effect in a laboratory task but whether it composes a stack of robust effects in the actual choice architecture of the workplace. A benefits-design proposal that rests on a single replication-crisis-vintage finding should be evaluated more skeptically than a proposal that composes multiple well-replicated findings into a single intervention. The compositional design philosophy is what produced SMarT, and it is the philosophy that distinguishes interventions that scale from interventions that do not.
The second implication is that the timing of a benefits-design change to coincide with a positively-framed reference event --- a raise, a promotion, an anniversary --- is a structurally important feature of the design, not a cosmetic one. The loss-aversion mechanism is what defeats most “ask employees to give something up” interventions, and the timing-to-coincide-with-a-gain construction is what neutralizes the loss-aversion mechanism. Any benefits change that requires the employee to absorb a felt loss --- a higher contribution rate, a higher deductible, a longer vesting period --- should be evaluated for whether the timing of the change can be aligned with a reference-event gain that masks the loss.
The third implication is that opt-out architecture is a more powerful retention-and-engagement tool than opt-in architecture for any benefit that produces long-term value but requires short-term commitment. The status-quo-bias mechanism preserves the default trajectory, and a default trajectory that is in the employee’s long-term interest will generally be honored even when individual escalation events have an opt-out option. This applies to retirement saving but also to health-savings-account contributions, life-insurance enrollment, professional-development time, and a range of other benefits where the value is long-term and the commitment is short-term.
The fourth implication is that the policy-adoption arc of SMarT is itself a model for how a behavioral-design innovation can move from a single-employer trial to industry-wide adoption. The pattern was: produce a clean empirical demonstration at one employer with proper controls; document the generalization across additional implementations; identify the statutory or regulatory friction that prevents broader adoption; advocate for a statutory safe harbor that removes the friction; and let the safe-harbor structure pull adoption forward across the industry. Other behavioral-design innovations in benefits, healthcare, and consumer protection that follow a similar arc are more likely to achieve real adoption than those that depend on individual employers or providers adopting the innovation without regulatory cover.
The fifth implication is that the right way to read the broader behavioral-economics literature for strategic purposes is to filter for interventions that have been tested in the field, that compose multiple robust findings, that have effect sizes large enough at the individual level to survive implementation noise, and that have produced observable adoption at scale. The findings that pass this filter are a small subset of the behavioral-economics literature, but they are the subset that is reliable enough to invest real organizational effort in. SMarT is one of these. Most of the canonical mid-2000s findings cataloged elsewhere in this hub are not.
Sources
- Thaler, R. H., & Benartzi, S. (2004). Save more tomorrow: Using behavioral economics to increase employee saving. Journal of Political Economy, 112(S1), S164—S187. DOI: 10.1086/380085
- Benartzi, S., & Thaler, R. H. (2007). Heuristics and biases in retirement savings behavior. Journal of Economic Perspectives, 21(3), 81—104. DOI: 10.1257/jep.21.3.81
- Beshears, J., Choi, J. J., Laibson, D., & Madrian, B. C. (2017). The effect of providing peer information on retirement savings decisions. Journal of Finance, 70(3), 1161—1201. DOI: 10.1111/jofi.12258
- Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving Decisions About Health, Wealth, and Happiness. Yale University Press.
Related Articles in This Hub
- The Defaults and Status Quo Anti-Example: The Behavioral Finding That Has Held Up
- Hyperbolic Discounting: The Time-Preference Finding That Held Up
- Loss Aversion: Prospect Theory’s Most Tested Component
- Mental Accounting: Thaler’s Real-But-Conditional Finding
- The Equity Premium Puzzle: The Finance-Theory Anomaly That Survived
Frequently Asked Questions
Is Save More Tomorrow a replicated finding or a casualty of the replication crisis?
Save More Tomorrow is one of the cleanest behavioral-economics interventions to have survived empirical scrutiny. The original Thaler-Benartzi 2004 field trial at Ispat Inland Steel produced a roughly 10-percentage-point increase in contribution rates over 40 months. The subsequent literature, including the Benartzi-Thaler 2007 JEP review and the Beshears-Choi-Laibson-Madrian retirement-savings research program, has consistently documented effects of similar magnitude across a broad range of employer implementations. The intervention has been codified into federal pension law via the 2006 Pension Protection Act and adopted as standard practice in U.S. 401(k) plans. The effect is real, large, durable, and reproducible.
Why does SMarT work when so many other behavioral-economics interventions do not?
SMarT works because it stacks four well-replicated behavioral principles --- hyperbolic discounting, loss aversion, status quo bias, and bounded commitment via a cap --- into a single intervention engineered for the specific choice architecture of corporate 401(k) plans. The compositional design is robust in a way that single-principle interventions are not. The intervention also targets a hard, easily-measured outcome (the contribution rate), produces effect sizes large enough at the individual level to survive implementation noise, and was tested in real workplaces rather than laboratory settings.
What was the magnitude of the effect in the Ispat Inland Steel field trial?
Among the workers who enrolled in the SMarT plan, the average contribution rate rose from approximately 3.5 percent of pay at the start of the program to approximately 13.6 percent of pay over the following 40 months, across four scheduled pay raises and four scheduled contribution-rate increases. The control comparisons --- workers who attended the financial-planning meeting but did not enroll, and workers who did not attend at all --- showed essentially flat contribution rates over the same window. The roughly 10-percentage-point increase was specifically attributable to the auto-escalation mechanism.
How did Save More Tomorrow get codified into federal law?
The 2006 Pension Protection Act created a statutory safe harbor for employers offering 401(k) plans that incorporated auto-enrollment defaults and auto-escalation schedules conforming to certain minimums. Employers who adopted SMarT-style designs were granted relief from certain non-discrimination testing requirements and certain fiduciary exposures, which substantially lowered the implementation cost of adoption. The post-PPA adoption arc moved the SMarT design from a single-employer trial to the de facto default architecture for new U.S. 401(k) plans over roughly a decade.
Does SMarT solve the entire retirement-savings adequacy problem?
No, and the original paper does not claim that it does. The intervention produces a measurable contribution-rate increase among the subpopulation of workers who enroll in it. Workers who do not enroll, workers at employers without auto-escalation plans, and workers whose contribution rates remain below the program cap even after escalation are not fully covered. The intervention is one component of a broader retirement-savings policy stack, and the conservative framing of the original claim is part of what has allowed the empirical record to hold up over time.
What are the practical implications for benefits design beyond 401(k) plans?
The compositional design philosophy --- timing benefits changes to coincide with positively-framed reference events, using opt-out rather than opt-in architecture for benefits with long-term value and short-term commitment, capping commitments to make them psychologically tractable, and exploiting hyperbolic discounting via future-commitment rather than immediate-action asks --- can be applied to health-savings-account contributions, life-insurance enrollment, professional-development time, and a range of other benefits where the value is long-term and the perceived short-term cost is what defeats opt-in adoption.
How should a strategist evaluate other behavioral-economics interventions for adoption?
The SMarT case study provides a working diagnostic checklist. Look for interventions that compose multiple well-replicated underlying findings rather than resting on a single fragile effect, that have been tested in real-world choice environments rather than laboratory tasks, that produce effect sizes large enough at the individual level to survive implementation noise, that target hard easily-measured outcomes rather than soft attitudinal variables, and that have produced observable adoption at scale rather than remaining at the laboratory-demonstration stage. Interventions that pass all five filters are reliable enough to invest organizational effort in. Most do not.
Why is the timing-to-coincide-with-a-pay-raise feature so important?
Loss aversion predicts that a felt loss is weighted roughly twice as heavily as an equivalent felt gain. If a worker’s contribution rate goes up and their take-home pay falls, the fall in take-home pay is experienced as a loss and triggers the loss-aversion barrier that defeats most “raise your contribution rate today” asks. If the contribution rate goes up at the same moment as a pay raise, the worker’s take-home pay still goes up --- just by less than it would have otherwise --- and there is no felt loss. The contribution increase is invisible against the backdrop of the raise. This structural feature is what allows the auto-escalation to proceed at each scheduled event without triggering the opt-out that would otherwise destroy the program.