TL;DR. “Track everything” and “stop measuring your life” are both wrong. A meta-analysis of 138 randomized studies with 19,951 participants found that monitoring progress improves goal attainment (d = 0.40), and that design decides how much: 0.47 when results were reported to at least one other person, 0.19 when they stayed private, and monitoring a behaviour moved that behaviour (d = 0.79) but not the outcome it was meant to serve (d = 0.14). Measurement also has a price: in six experiments, counting an activity lowered enjoyment, and after reading with a page counter 27.3% chose to keep reading, against 48.5% without one. In a CEOtudent join of US government data, Americans aged 15 and over averaged 2.61 hours a day of television in 2025 and 0.58 hours of socializing, yet in 2021 ratings socializing scored 5.12 for happiness on a 0-6 scale, against 4.24 for television. The CEO move: run five to seven KPIs, each tied to a decision, a threshold, a counter-metric and a sunset rule. The student move: treat every metric as a hypothesis and retire it when it stops predicting what you care about.
This is the measurement layer of the Personal Operating System, and it serves the argument of the cluster hub on life design versus goal setting: systems need instruments, and instruments need rules. The cadence that reads the dashboard lives in the weekly review protocol and the annual personal review.
Why do “track everything” and “measure nothing” both fail?
A well-run company picks a handful of indicators because each one changes a decision, pairs volume metrics with quality metrics so nobody hits one by wrecking the other, and reads them on a fixed rhythm.
The quantified-self approach collects whatever a device can capture. The backlash notices that numbers distort behaviour and drops them. The research treats measurement as a design problem with known failure points.
What does self-monitoring actually do to goal attainment?
The headline result is covered in our comparison of habit tracking with and without an AI assistant. Harkin and colleagues pooled 138 studies (N = 19,951) that randomly assigned people to an intervention prompting progress monitoring. The interventions raised monitoring frequency (d = 1.98) and goal attainment (d = 0.40). For KPI design, the moderators matter more than the average.
Table 1. Progress-monitoring interventions and goal attainment, selected moderators (figures as printed)
| Moderator | Level | Studies (k) | N | Goal attainment d+ | 95% CI |
|---|---|---|---|---|---|
| What is monitored | Behaviour | 78 | 11,461 | 0.43 | 0.33 to 0.52 |
| What is monitored | Outcomes | 83 | 11,360 | 0.42 | 0.31 to 0.52 |
| Who sees it | Public | 3 | 214 | 0.55 | -0.16 to 1.26 |
| Who sees it | Private, reported to someone | 95 | 12,155 | 0.47 | 0.37 to 0.58 |
| Who sees it | Private, not reported | 14 | 3,177 | 0.19 | 0.05 to 0.33 |
| Recording | Physically recorded | 106 | 15,589 | 0.43 | 0.34 to 0.53 |
| Recording | Not recorded | 32 | 2,809 | 0.29 | 0.15 to 0.42 |
| Method | Pedometer | 5 | 1,070 | 0.02 | -0.10 to 0.14 |
| Study quality | Participants blind to condition | 14 | 1,911 | 0.32 | 0.14 to 0.49 |
| Study quality | Participants not blind | 34 | 7,285 | 0.53 | 0.34 to 0.73 |
Source: Harkin and colleagues, Psychological Bulletin (2016), Table 3. Public monitoring rests on three studies and its interval crosses zero.
A number nobody else sees is the weakest version. The paper defines the middle row as progress monitored privately but “reported to at least one other person”. Private monitoring that was not reported produced about two-fifths of that effect (0.19 against 0.47). A KPI needs a reporting line, even if the board is one friend.
A metric moves what it measures. The authors found a “matching effect”: monitoring behaviour had a large effect on behaviour (d = 0.79) but no reliable effect on outcomes (d = 0.14), while monitoring outcomes moved outcomes (d = 0.62) but not behaviour (d = 0.17). A lead indicator (something you do) and a lag indicator (the result you want) are separate instruments, and the link between them is a hypothesis that needs its own check.
Capture is not use. The pedometer studies produced one of the largest increases in monitoring frequency (d = 3.02) and essentially no change in goal attainment (d = 0.02). Expect modest effects: they were smaller in better-quality studies (0.32 with participants blind to condition, against 0.53 without), and 0.33 among adequately powered studies.
Where does measurement start to backfire?
The oldest warning comes from social science. In work first presented in 1974, the psychologist Donald T. Campbell stated what is now called Campbell’s law: “The more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures and the more apt it will be to distort and corrupt the social processes it is intended to monitor.” His illustrations included Soviet production targets: factories judged by weight made their heaviest items, such as the largest nails, and factories judged by item counts made the smallest nails.
The economist Charles Goodhart made a related point about monetary statistics in 1975. His original was not opened for this piece, and later sources quote it in slightly different words; one version reads “any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” The popular line, “When a measure becomes a target, it ceases to be a good measure,” was written by the anthropologist Marilyn Strathern in 1997 as a paraphrase, according to a 2018 note in Significance that also gives Campbell the earlier claim.
For one person the corruption is quieter: walking laps of the kitchen at night to close a ring, or logging a meeting as focus time. The company fix is the counter-metric, a second number that gets worse when the first is gamed.
The second hazard needs no gaming at all. Jordan Etkin’s six experiments, published in the Journal of Consumer Research, tested what counting does to activities people do for enjoyment.
Table 2. What counting did to output and enjoyment (CEOtudent analysis of Etkin 2016)
| Exp. | Activity and sample | Output, no count vs count | Change | Enjoyment (1-7), no count vs count | Difference |
|---|---|---|---|---|---|
| 1 | Coloring, 105 students | 6.91 vs 8.68 shapes | +25.6% | 5.13 vs 4.63 | -0.50 (p = .062) |
| 2 | Walking, opt-in pedometer, 91 students | 7.01 vs 7.97 (log steps) | higher | 5.33 vs 4.82 | -0.51 (p = .030) |
| 3 | Walking, 100 students | 7.28 vs 8.06 (log steps) | higher | 5.21 vs 4.47 | -0.74 (p = .001) |
| 4 | Reading, 310 students | 13.44 vs 16.14 pages | +20.1% | 5.97 vs 5.19 | -0.78 (p = .003) |
| 5 | Reading, 236 online adults | 10.56 vs 12.58 pages | +19.1% | 5.64 vs 5.27 | -0.37 (p = .057) |
| 6 | Reading, 66 students | 14.36 vs 16.70 pages | +16.3% | 5.37 vs 4.68 | -0.69 (p = .020) |
Source: Etkin, Journal of Consumer Research (2016), author manuscript; changes and differences computed by CEOtudent. Steps were analysed on a log scale. Experiment 4: output pools all framings, enjoyment is unframed; experiment 5: no load. Two differences (p = .062 and .057) were marginal.
Enjoyment fell in every study, and output rose in every main comparison (in experiment 5, only without cognitive load). Etkin’s tested mechanism is that a counter makes an enjoyable activity seem more like work, and the downstream results matter most. After the counter was removed, people who had seen it read fewer pages (3.75 against 4.20). In the sixth study, 27.3% of counted readers chose to keep reading, against 48.5% of uncounted ones. In the walking study, counting lowered overall happiness and satisfaction (4.59 against 5.18), and in the coloring study coders rated counted drawings less creative (2.58 against 3.02 on a 1-5 scale).
This is not a case for never measuring. When reading was framed as useful for learning, the penalty disappeared (5.57 with a counter against 5.18 without, not significant), and with an explicit goal it reversed direction (p = .079). The cost falls on activities done for their own sake, and people accept it willingly: most participants offered a count wanted it (88% for steps, 74% for pages, 70% for shapes). No direct replication was located, so treat the effect size as provisional.
What does a well-calibrated threshold look like?
A threshold should sit where returns change, not at a round number. Daily steps are the clearest public example. Paluch and colleagues pooled 15 cohorts with 47,471 adults and 3,013 deaths over a median follow-up of 7.1 years, and note that the 10,000-steps goal “is not based on evidence, but instead originates from a marketing campaign in Japan.”
Table 3. Daily steps and all-cause mortality (figures as printed, with CEOtudent arithmetic)
| Quartile | Median steps per day | Adjusted hazard ratio vs Q1 (95% CI) | Extra steps vs Q1 (CEOtudent) | Hazard-ratio points per 1,000 extra steps over previous quartile (CEOtudent) |
|---|---|---|---|---|
| Q1 | 3,553 | reference | – | – |
| Q2 | 5,801 | 0.60 (0.51 to 0.71) | 2,248 | 17.8 |
| Q3 | 7,842 | 0.55 (0.49 to 0.62) | 4,289 | 2.4 |
| Q4 | 10,901 | 0.47 (0.39 to 0.57) | 7,348 | 2.6 |
Source: Paluch and colleagues, Lancet Public Health (2022). The spline models put the plateau at about 6,000-8,000 steps a day for adults aged 60 and over and 8,000-10,000 for adults under 60. The last two columns are rough arithmetic on pooled medians of study-specific quartiles and describe an association, not a causal dose.
The first 2,248 extra steps came with a hazard ratio 40 points lower; the next 5,100 came with 13 more points. A single 10,000 line sits past the point where the curve flattens for older adults. These are observational associations, and this section is general information, not medical advice. The transferable rule: set green at the point of diminishing returns or at your own baseline, and say which.
Are you measuring output or experience?
Most personal dashboards count output, which the Contribution Ladder handles at work. A life dashboard has a second problem: the hours easiest to count are not the hours rated highest.
The US Bureau of Labor Statistics measures both. The American Time Use Survey 2025 results, released on 25 June 2026, report how people aged 15 and over spend an average day. The survey’s 2021 Well-Being Module, fielded from March to December 2021, asked respondents to rate three randomly selected activities from their diary day for how happy, tired, sad, stressed and in pain they felt (0-6) and how meaningful each was. Table 4 joins the two. The feelings are time-weighted, so each describes an average minute of that activity across the population.
Table 4. Where the hours go and how they feel (CEOtudent analysis of BLS American Time Use Survey data)
| Activity | Hours per day, 2025 | Happy, 2021 | Stressed, 2021 | Tired, 2021 | Meaningful, 2021 | Rated activities |
|---|---|---|---|---|---|---|
| Working | 3.03 | 3.83 | 2.42 | 2.58 | 4.14 | 1,320 |
| Watching television | 2.61 | 4.24 | 0.96 | 2.53 | 3.62 | 2,527 |
| Household activities | 1.99 | 4.21 | 1.39 | 2.33 | 4.31 | 4,197 |
| Eating and drinking | 1.21 | 4.67 | 1.05 | 2.05 | 4.41 | 3,594 |
| Purchasing goods and services | 0.69 | 4.23 | 1.36 | 2.04 | 4.17 | 2,027 |
| Socializing and communicating | 0.58 | 5.12 | 0.84 | 1.94 | 5.05 | 625 |
| Caring for household members | 0.49 | 4.78 | 1.61 | 2.58 | 5.12 | 1,119 |
| Educational activities | 0.44 | 3.49 | 2.65 | 3.10 | 4.08 | 183 |
| Sports, exercise and recreation | 0.34 | 4.84 | 0.76 | 2.42 | 4.88 | 424 |
| Organizational, civic and religious | 0.28 | 4.92 | 0.96 | 1.79 | 5.49 | 475 |
| All rated activities | – | 4.24 | 1.53 | 2.41 | 4.18 | 20,394 |
Sources: BLS, American Time Use Survey 2025 Results, Table 1 (hours averaged over everyone aged 15 and over, including people who did not do the activity); BLS, 2021 Well-Being Module microdata. CEOtudent used the time-based method in the module’s data dictionary and the 160 replicate weights: standard errors are 0.03 to 0.17 except for education (0.16 to 0.25), and the code reproduces the two worked results BLS prints (0.82 and 1.00). Sleep and grooming were not rated. Major categories include related travel, as Table 1 does; working, television, socializing and sports exclude it.
The biggest blocks are not the best-rated ones. Working and television take 5.64 hours of the average day. Working has the lowest happiness and highest stress of any row except education, and television the lowest meaning (3.62, against 4.18 for all rated activities). Socializing, the happiest row at 5.12, gets 0.58 hours; television outspends it by a factor of 4.5.
Happiness and meaning are different instruments. Caring for household members and civic or religious activities score highest on meaning (5.12 and 5.49), while television matches the average for happiness and trails it on meaning. A dashboard with only a happiness reading pushes toward the couch; one with only an output reading pushes toward the desk.
The pattern is not a quirk of one survey. Kahneman and colleagues’ 2004 Day Reconstruction Method study of 909 employed women found working (3.62 positive affect on a 0-6 scale) and commuting (3.45) to be the two lowest-rated activities, while working filled 6.9 hours of the day; socializing scored 4.59. Two methods, 17 years apart, point the same way.
The limits are real: hours from 2025 and feelings from 2021, a pandemic year; US averages from one diary day per person; no 2025 diary days from 30 September to 11 November because of the federal shutdown; and, as BLS notes, no estimate of how the average participant felt. Read it as a map of where time and experience diverge, not a ranking of your life.
The personal version is cheap: for one week, rate the three biggest blocks of each day on the same happy, stressed and meaningful scales, an energy audit applied to experience. For a whole-life lag indicator, official statistics use the Cantril ladder, which the BLS module asked in these words: “on which step of the ladder do you feel you personally stand at the present time?”, with 0 the worst and 10 the best possible life.
How should you choose your personal KPIs?
Table 5. The Personal KPI Charter (CEOtudent editorial framework)
| Field | What it must state | Evidence behind it | Failure it prevents |
|---|---|---|---|
| Decision | The choice this reading changes, written in advance | Campbell’s law | Dashboard as decoration |
| Type | Lead (a behaviour) or lag (a result), and which lag the lead should move | Matching effect | Assuming the lead moves the lag |
| Threshold | Green, yellow and red bands from evidence or your baseline | Step-count plateaus | Round-number targets |
| Cadence and audience | How often it is read, and who sees it | Reported versus private monitoring | Numbers nobody acts on |
| Counter-metric | A second reading that worsens if the first is gamed | Campbell’s nail factories | Hitting the number, missing the point |
| Experience check | For activities done for their own sake, no output count, or an enjoyment rating beside it | Etkin’s six experiments | Turning play into work |
| Sunset rule | The condition that retires or demotes the metric | Pedometer studies | Metrics that outlive their use |
Rules for the whole dashboard: five to seven KPIs, no more than two per domain, at least one lag indicator for the life as a whole, every KPI read in a scheduled review, and a written list of what you deliberately do not quantify:
- Time with the people you love. Counting it turns a relationship into a quota; ask a yes-or-no question about it in the quarterly review instead.
- Hobbies and reading done for pleasure. In Etkin’s reading studies, the count raised pages and cut the wish to continue.
- Creative play. Counted participants produced less creative work.
- Rest. Track sleep duration if useful (see the wearable metrics guide), but a scored weekend is no longer a weekend.
- Anything you could not act on. If no reading would change a decision, the metric is surveillance, not management.
A six-question test for every metric on your dashboard
- What decision does it change? Write the action for red and for green. If both are “carry on”, delete it.
- Lead or lag, and is the link real? Name the lag a lead should move, and check each quarter whether they moved together.
- How would you game it? If the number can rise without the thing improving, add a counter-metric.
- Does counting change the experience? For activities done for their own sake, stop counting output or switch to an enjoyment rating.
- Is the change bigger than the noise? Use multi-week averages and your own baseline, with the humility of the confidence calibration guide.
- Who sees it, and when? A metric with no audience and no review slot belongs in a notebook.
A metric that fails two of the six comes off at the next review. That is the student half of the system: every metric is a hypothesis about what predicts the life you want.
What does an illustrative dashboard look like?
Table 6. One-page personal KPI card (CEOtudent editorial framework; every value is an illustrative assumption)
| Domain | KPI (type) | Green / yellow / red | Audience | Counter-metric | Decision on red | Sunset rule | Reading |
|---|---|---|---|---|---|---|---|
| Health | 7-day average daily steps (lead) | 8,000+ / 6,000-7,999 / under 6,000 | Training partner, weekly | Enjoyed a walk (yes/no) | Walking calls | Monthly after 12 green weeks | 7,400: yellow |
| Sleep | 7-day average sleep vs own 90-day baseline (lead) | within 15 min / 15-45 min short / more | Weekly review | Rested on waking (very to not at all) | Drop an evening commitment | Retire if it stops tracking restedness | 20 min short: yellow |
| Work | Outcomes delivered per week (lag) | 3+ / 2 / 0-1 | Manager or peer, weekly | Work stress (0-6), red at 3.5+ for 3 weeks | Cut scope before adding hours | Replace if the role is judged differently | 3 outcomes; stress 3.5, 3.8, 4.1: counter-metric red |
| Learning | Deliberate practice hours per week (lead) | 4+ / 2-3.9 / under 2 | Weekly | Monthly skill test | Test flat for 2 months: change method, not volume | Retire at target | 2.5 hours (62.5% of target): yellow |
| Life | Cantril ladder, 0-10 (lag) | no fall / 1 step down / 2+ down | Monthly, filed for the annual review | None: it guards the card | Two falls of 2+ steps: re-plan | Never | 7, unchanged: green |
The step bands are illustrative for a reader under 60, placed at the lower edge of the plateau range in Table 3; they are not a health target. The rested-on-waking scale borrows the BLS module’s wording.
Read the work row the way a board would: output is on target, the counter-metric says it is being hit by spending the person, so the rule says cut scope. That is the CEO half, decisions written in advance so a reading triggers an action instead of a mood. The sunset column is the student half.
What this does not mean
- Reactivity is both mechanism and risk. Monitoring changes behaviour, which is why it works, and can change how an activity feels.
- Personal data are small samples. In Table 4, education rests on 183 rated activities and carries a standard error of 0.17 on happiness, against 0.03 for leisure with 5,816. One person’s week is seven observations, far noisier than either, so use rolling averages.
- Population averages are not personal norms. Your own ratings may rank activities differently.
- Health figures here are general information. The step findings are associations from observational cohorts, not a prescription.
- A dashboard is not a review. If routines keep breaking while the numbers look fine, the habit failure-mode taxonomy is the next diagnostic.
Frequently asked questions
How many personal KPIs are enough?
Five to seven, no more than two per domain. Each needs a decision rule and a review slot, which larger dashboards rarely get.
Should progress be shared with someone?
Where possible. In Harkin and colleagues’ meta-analysis, monitoring reported to at least one other person had an effect of 0.47 on goal attainment, against 0.19 for private monitoring that was not reported.
Does tracking an activity make it less enjoyable?
It can. In Etkin’s six experiments, counting lowered enjoyment every time while usually raising output; the penalty disappeared when the activity was framed as useful rather than fun.
What is the difference between a lead and a lag indicator for a person?
A lead is a behaviour you control, such as practice hours; a lag is the result it should produce, such as a skill test. Monitoring behaviour moved behaviour (d = 0.79) but not outcomes (d = 0.14), so check every lead against its lag.
Is 10,000 steps a day the right target?
Paluch and colleagues trace the figure to a marketing campaign in Japan; in their cohorts the mortality association levelled off at about 6,000-8,000 steps for adults aged 60 and over and 8,000-10,000 for younger adults. This is general information, not a recommendation.
When should a metric be retired?
When it fails two of the six questions, usually because it no longer changes a decision or no longer predicts the outcome it was chosen for.
Sources
- Harkin, Webb, Chang, Prestwich, Conner, Kellar, Benn and Sheeran (2016). Does Monitoring Goal Progress Promote Goal Attainment? A Meta-Analysis of the Experimental Evidence. Psychological Bulletin, 142(2), 198-229.
- Etkin (2016). The Hidden Cost of Personal Quantification. Journal of Consumer Research, 42(6), 967-984. Author manuscript.
- Campbell (1976). Assessing the Impact of Planned Social Change. Occasional Paper 8, Public Affairs Center, Dartmouth College; reprinted in Journal of MultiDisciplinary Evaluation, 7(15), 2011. The 1979 version in Evaluation and Program Planning was not opened.
- Rodamar (2018). There Ought to Be a Law! Campbell versus Goodhart. Significance, 15(6), 9. Magazine column, third-party copy; quotes Strathern (1997) and Goodhart (1975).
- Manheim and Garrabrant (2018, revised 2019). Categorizing Variants of Goodhart’s Law. arXiv preprint 1803.04585. Quotes Goodhart (1975).
- Paluch and colleagues, Steps for Health Collaborative (2022). Daily Steps and All-Cause Mortality: A Meta-Analysis of 15 International Cohorts. Lancet Public Health, 7(3), e219-e228. NIH author manuscript.
- Kahneman, Krueger, Schkade, Schwarz and Stone (2004). A Survey Method for Characterizing Daily Life Experience: The Day Reconstruction Method. Science, 306(5702), 1776-1780.
- US Bureau of Labor Statistics (2026). American Time Use Survey: 2025 Results, news release USDL-26-1022, 25 June 2026, including Table 1.
- US Bureau of Labor Statistics. ATUS 2021 Well-Being Module and 2021 ATUS microdata files.
- US Bureau of Labor Statistics (2022). ATUS 2021 Well-Being Module Data Dictionary.
- US Bureau of Labor Statistics (2026). American Time Use Survey User’s Guide, September 2026.
- US Bureau of Labor Statistics (2026). Federal Government Shutdown: 2025 American Time Use Survey (notice).
Tables 2 and 4 are CEOtudent analyses (Table 4 joins BLS 2025 hours with feelings computed from 2021 BLS microdata). Tables 5 and 6 are CEOtudent editorial frameworks; every value in Table 6 is illustrative. Tables 1 and 3 report figures as printed, with CEOtudent arithmetic labelled. Deliberately not used: the 7,000-steps figures from Ding and colleagues’ 2025 Lancet Public Health meta-analysis, because a published correction to it could not be opened; Goodhart’s 1975 wording as a quotation from the original, which was not opened and appears in two versions in later sources; and any claim that Etkin’s results have been replicated.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
