TL;DR. The advice to hold a five-year or ten-year vision is almost never accompanied by evidence that five-year or ten-year forecasts hold. We went and checked, using only forecasts whose horizon has already closed. Four findings. First, the World Economic Forum has now published five consecutive five-year forecasts of global job creation and destruction, and the net direction flipped sign three times; the two editions that state a net figure as a share of their own base, published two years apart, say minus 2% and plus 7%. Second, the 2025 edition contains no reference to the 85 million and 97 million figures its 2020 edition published for 2025, even though that horizon has now closed. Third, the one long-range forecast that can be scored cleanly, the US Bureau of Labor Statistics ten-year projection for 2014 to 2024, projected 6.5% growth against a realised 13.0%, an error of exactly a factor of two, and the 2019 to 2029 projection was passed by reality four years early. Fourth, six consecutive BLS ten-year forecasts of the same economy span 3.06% to 6.51%, a 2.13-fold range, which is a direct measurement of how much a ten-year view moves when you simply re-ask the question a year later. The practical conclusion is not that planning is useless. It is that direction and commitment need different horizons, and most people collapse them into one.
The question nobody asks about the five-year plan
There is a peculiar asymmetry in how we treat the future. We demand evidence for almost every claim about the present, and then accept, without a shred of it, that a person should have a five-year vision and a ten-year goal.
The honest version of the question is empirical: at what horizon do forecasts about work still carry information? That is answerable, because a lot of long-range forecasts have now aged past their target dates. We can open them and score them.
This piece does that with two bodies of evidence that are unusually well suited to the task. The World Economic Forum Future of Jobs Report has asked employers the same five-year question in five editions since 2016. The US Bureau of Labor Statistics has produced a ten-year employment projection for the same economy, on the same methodology family, every year for decades, and publishes its realised base years, which means each forecast can be scored against the agency’s own later data.
If you want the companion piece that scores individual expert predictions about AI rather than institutional ones, we did that separately in the forecast scorecard for a decade of expert AI predictions.
Five consecutive five-year forecasts, and three sign flips
The Future of Jobs Report is the most cited single source on the future of work. Each edition surveys employers and projects job creation and displacement over the following five years. Here is what the five editions actually say, quoted from the reports themselves.
| Edition | Horizon | Jobs created | Jobs displaced | Net | Stated base |
|---|---|---|---|---|---|
| 2016 | 2015 to 2020 | 2 million | 7.1 million | minus 5.1 million | 15 economies |
| 2018 | to 2022 | 133 million | 75 million | plus 58 million | large firms, non-agricultural |
| 2020 | to 2025 | 97 million | 85 million | plus 12 million | company respondents |
| 2023 | 2023 to 2027 | 69 million | 83 million | minus 14 million | 673 million employees |
| 2025 | to 2030 | 170 million | 92 million | plus 78 million | 1.2 billion formal jobs |
Source: World Economic Forum Future of Jobs Report, editions 2016, 2018, 2020, 2023 and 2025. Net column is the report’s own creation figure minus its own displacement figure.
Read the net column downward: minus, plus, plus, minus, plus. Three sign changes in five consecutive attempts to answer the same question.
Now the honest caveat, which matters enormously and which is itself part of the finding. The absolute millions in that table are not comparable across editions. The 2016 edition covered 15 economies. The 2018 and 2020 editions extrapolated from large-firm respondents. The 2023 edition used a base of 673 million employees, and the 2025 edition a base of 1.2 billion formal jobs. The denominators nearly doubled between the last two editions alone. Anyone who lines these numbers up as a trend line is making an error.
But that caveat has a sharp edge. The single most cited forecast series about the future of work cannot be assembled into a consistent time series, even by reading every edition in full. That is not a criticism of the Forum, which is transparent about its changing samples. It is a fact about what a five-year labour forecast is: a snapshot of employer sentiment at a moment, not a measurement that can be tracked.
There is one comparison the reports do license. Two editions state the net figure as a share of their own base. The 2023 edition reports a “net decrease of 14 million jobs, or 2% of current employment.” The 2025 edition reports “net growth of 78 million jobs (7% of today’s total employment).” Same institution, same five-year question, published two years apart, on a normalised basis: minus 2% and plus 7%. A nine percentage point swing, including a change of direction, from two years of re-asking.
The forecast that nobody scored
The 2020 edition’s horizon was 2025. That year has now closed. So the obvious question is whether 85 million jobs were displaced and 97 million created.
We searched the full text of the 2025 edition for the answer. The strings “85 million” and “97 million” appear zero times. The phrase “Future of Jobs Report 2020” appears zero times. The 2025 report does compare itself to its predecessors, eight times, and one of those comparisons is substantive and useful, which we come to next. But the headline job numbers from the edition whose horizon has just closed are not among them.
This is worth stating carefully, because it is a statement about a document rather than about the world. The 2025 edition does not claim its predecessor was right, and does not claim it was wrong. It simply issues a new five-year forecast and moves on.
That pattern, forecast and replace rather than forecast and score, is the normal condition of futures work. It is also precisely why the genre resists correction, and why a reader has to do the scoring themselves. We built a method for spotting weak signals early for the same reason: the institutional pipeline does not reward retrospective accuracy.
The one metric the Forum does track, and what it shows
To its credit, the 2025 edition does compare one measure across editions explicitly. On skill instability, the share of workers’ existing skills expected to be transformed or become outdated over the coming five years, it reports:
“two-fifths (39%) of their existing skill sets will be transformed or become outdated over the 2025-2030 period. However, this measure of ‘skill instability’ has slowed compared to previous editions of the report, from 44% in 2023 and a high point of 57% in 2020 in the wake of the pandemic.”
| Edition | Five-year horizon | Share of skills expected to be disrupted |
|---|---|---|
| 2020 | to 2025 | 57% |
| 2023 | to 2027 | 44% |
| 2025 | to 2030 | 39% |
Source: World Economic Forum Future of Jobs Report 2025, which states the series and attributes the 2020 peak to the pandemic.
This is the cleanest comparable series in the whole literature, and it goes the opposite way to the public conversation. Through the period in which generative AI went from a research demo to a workplace fixture, employers’ expectation of how much of their workforce’s skill base would be disrupted fell by 18 percentage points. The forecast of disruption peaked in 2020, before ChatGPT existed.
The lesson is not that disruption is not happening. It is that forward-looking sentiment measures track the mood of the moment at least as much as the underlying reality, and the mood of 2020 was pandemic, not AI.
Scoring a ten-year forecast against what actually happened
Employer sentiment is soft data. So here is the hardest long-range forecast available: the US Bureau of Labor Statistics ten-year employment projection, built from macroeconomic models, industry staffing patterns and occupational matrices, and published annually.
The 2014 to 2024 round can now be scored completely, because 2024 is a realised base year in a later round.
| Quantity | Value | Source |
|---|---|---|
| Employment, 2014 (base) | 150.5 million | BLS 2014 to 2024 projections |
| Employment projected for 2024 | 160.3 million | BLS 2014 to 2024 projections |
| Projected ten-year growth | plus 9.8 million, plus 6.5% | BLS 2014 to 2024 projections |
| Employment, 2024 (realised base) | 170.0 million | BLS 2024 to 2034 projections |
| Realised ten-year growth | plus 13.0% | BLS 2024 to 2034 projections |
| Forecast error | realised was 2.00 times projected, plus 6.5 points | computed |
Level cross-check: 150.5 million multiplied by 1.130 gives 170.1 million, against the 170.0 million BLS states for 2024. The 0.1 million gap is rounding, which confirms the two figures sit on the same employment measure and that the comparison is legitimate.
The best-resourced ten-year labour forecast in the world, for the world’s most measured economy, in a decade with no world war and one pandemic, was wrong by a factor of exactly two on its single headline number.
It gets sharper. The 2019 to 2029 round projected total employment of 168.8 million for 2029. BLS states realised employment of 170.3 million for 2025. The target was passed four years before the forecast window closed.
How much a ten-year view moves when you just ask again
The most direct measurement of forecast stability is to line up successive rounds. Each of these is a ten-year forecast of the same economy, made one or two years apart by the same agency.
| Projection round | Base employment | Projected end employment | Projected ten-year change |
|---|---|---|---|
| 2014 to 2024 | 150.5 million | 160.3 million | plus 6.51% |
| 2019 to 2029 | 162.8 million | 168.8 million | plus 3.69% |
| 2021 to 2031 | 158.1 million | 166.5 million | plus 5.31% |
| 2023 to 2033 | 167.9 million | 174.6 million | plus 3.99% |
| 2024 to 2034 | 170.0 million | 175.2 million | plus 3.06% |
| 2025 to 2035 | 170.3 million | 176.2 million | plus 3.46% |
Percentages computed from the stated endpoint levels. Two notes on fidelity. BLS states the 2021 to 2031 change as 8.3 million where the stated endpoints differ by 8.4 million, a rounding artefact of unrounded underlying levels. BLS also warns in that release that the 2021 base year “does not reflect much of the employment recovery” after the pandemic, which is the main reason that round sits high. The 2023 base is derived as 174.6 million minus the 6.7 million BLS states.
The range runs from 3.06% to 6.51%, a factor of 2.13. That is not the difference between optimists and pessimists. It is the same institution, the same method, the same country, moving by more than two-fold depending on which year you happened to ask.
And note the direction of the recent rounds against realised history. BLS projects 3.1% for 2024 to 2034 having just recorded 13.0% for 2014 to 2024, and 3.5% for 2025 to 2035 having just recorded 10.9% for 2015 to 2025. The current forecasts are between a quarter and a third of the immediately preceding decade’s outcome: 3.1% against a realised 13.0%, and 3.5% against a realised 10.9%. They may be right. But if you are building a personal ten-year plan on the assumption that official projections are conservative anchors, the recent record is that they have been conservative, badly, in the one round that has fully closed.
What actually survives the horizon
Put the two bodies of evidence together and a pattern appears that is more useful than “forecasts are bad.”
The things that decayed fastest were quantities: how many jobs, what percentage of skills, what net change. The things that survived were structures: which categories of work were growing, which were shrinking, what kind of capability was becoming scarce. The 2016 Future of Jobs Report was badly wrong about the number of jobs lost. It was substantially right that office and administrative work would carry the losses, and BLS now projects that group down 3.9% for 2024 to 2034, the largest decline of any major occupational group.
That asymmetry, direction surviving while magnitude decays, is the whole basis of a workable answer. And it maps onto a distinction most personal planning collapses: the difference between the horizon on which you point yourself and the horizon on which you commit resources you cannot get back.
The Vision Horizon Ladder
This is a CEOtudent editorial framework. It is not a research finding. It is our reading of what the decay evidence above implies for an individual, and you should treat it as an argument rather than a measurement.
| Rung | Horizon | What it is for | What the evidence supports | Failure mode when you use the wrong rung |
|---|---|---|---|---|
| Direction | 10 years and beyond | The kind of work you want to be doing and the kind of person you are becoming | Only structural claims: which categories grow, which capabilities stay scarce | Attaching a number or a title to it, which is the part that decays |
| Position | 3 to 5 years | The field, market or domain you are accumulating advantage in | Directional claims plus your own accumulating evidence | Treating it as fixed; this is the rung that most needs an annual re-read |
| Commitment | 12 to 24 months | Resources you cannot recover: relocation, training spend, equity, a role change | Concrete, checkable conditions rather than forecasts | Committing on a five-year forecast, which is the single most common error |
| Execution | 1 quarter | What you actually do this week | Nothing forecast-dependent at all | Letting the ten-year story dictate the quarter and calling it strategy |
CEOtudent editorial framework. The horizons are argued from the decay evidence in this piece, not measured.
The operating rule that falls out of the table: plan direction long, commit short, and re-read position annually. A ten-year direction costs nothing to hold and nothing to revise. A ten-year commitment is a bet on a forecast class that has a measured two-fold error.
If you want the mechanics of the commitment rung, the useful move is to decide the exit conditions before you enter, which we set out in kill criteria, deciding in advance when to quit, and to stress-test the plan before committing using the pre-mortem protocol.
The CEO and the student read the horizon differently
A chief executive is paid to hold a direction that does not change with the quarter, and simultaneously to refuse to commit capital to anything whose payoff depends on a five-year forecast being right. Those are not in tension. They are the same discipline applied at two horizons. What gets a CEO fired is not being wrong about the ten-year direction. It is committing irreversibly on the strength of it.
The student half is the part that most vision advice omits entirely. A student’s relationship to a five-year plan is that they expect to know more next year than they do now, and they build the plan so that knowing more is an improvement rather than a humiliation. A plan that requires you to have been right is brittle. A plan that gets better as you learn is the only kind worth writing down.
Which is why the annual re-read of the position rung matters more than the quality of the original forecast. On the evidence here, the people who did best over the last decade were not the ones with the most accurate five-year view. They were the ones who noticed, in year two, that office and administrative work was going in a particular direction, and moved. That is a foresight practice, not a prediction, and we ranked the practices that actually work in foresight methods ranked.
What this does not show
Three limits, stated plainly.
The BLS scoring covers one country and one completed round. A single completed forecast with a two-fold error is a strong data point, not a distribution; the honest claim is that the best available long-range labour forecast missed badly on the one occasion we can fully check, not that it always will.
The Future of Jobs comparison is deliberately restricted to what the reports say about themselves. We did not attempt to score the 2020 forecast against realised global employment, because the report’s base is a survey-weighted extrapolation with no published reconciliation to any official statistic. The absence of that reconciliation is the finding; inventing one would not be.
And none of this speaks to whether AI specifically will be larger or smaller than these forecasts imply. The evidence here is about the reliability of the horizon, not about the size of the shock. A world can be genuinely transformed and still be badly forecast, and in fact that combination is exactly what the 2014 to 2024 round looks like.
FAQ
So should I stop making five-year plans?
No. Keep the five-year rung for position, meaning the field and domain you are accumulating advantage in, and re-read it annually. Stop attaching specific numbers, titles or timings to it, because the quantity is the part with the measured error.
Is ten years too long for any kind of goal?
Not for direction. The evidence that decayed was quantitative. Structural direction, which categories of work grow and which capabilities stay scarce, held up reasonably well across the same decade. A ten-year direction is cheap to hold and cheap to revise. A ten-year commitment is not.
If official forecasts under-predicted growth, should I just assume things will be better than forecast?
That is the wrong lesson, and a dangerous one. The completed round under-predicted by a factor of two. The other rounds have not closed. One directional error is not a correction factor. The defensible conclusion is that the uncertainty band around a ten-year labour forecast is wide enough to include double the central estimate, in either direction.
Why did employer expectations of skill disruption fall while AI capability rose?
The Forum’s own explanation is that the 2020 peak of 57% reflected the pandemic. The broader reading is that these are sentiment measures, and sentiment tracks the salient disruption of the moment. That is a reason to weight them as mood data rather than as measurement.
What is the single change worth making after reading this?
Separate the horizon on which you point from the horizon on which you spend. Most people run one horizon for both, and it is usually five years, which is long enough for the forecast to be unreliable and short enough to feel concrete. That combination is the worst of both.
Sources
- World Economic Forum, The Future of Jobs Report 2025. Chapter 2.1 total job growth and loss: creation of 170 million jobs equivalent to 14% of today’s total employment, displacement of 92 million or 8%, net growth of 78 million or 7%, against a base of 1.2 billion formal jobs; Figure 2.1; the skill instability passage stating 39% for 2025 to 2030, 44% in 2023 and a high point of 57% in 2020.
- World Economic Forum, The Future of Jobs Report 2023. Executive summary and Figure 3.1: structural job growth of 69 million and decline of 83 million against a base of 673 million employees, a net decrease of 14 million or 2% of current employment, and churn of 152 million or 23%.
- World Economic Forum, The Future of Jobs Report 2020. Estimate that by 2025, 85 million jobs may be displaced and 97 million new roles may emerge.
- World Economic Forum, The Future of Jobs Report 2018. Estimate of 75 million jobs displaced and 133 million new roles emerging in the period to 2022, with the report’s own caveat that the projections represent the remit of large multinational employers.
- World Economic Forum, The Future of Jobs, 2016. Net employment impact of more than 5.1 million jobs lost over 2015 to 2020, comprising a total loss of 7.1 million, two thirds concentrated in the office and administrative job family, and a total gain of 2 million.
- US Bureau of Labor Statistics, Employment Projections, occupational employment projections to 2024, Monthly Labor Review, December 2015. Base year employment of 150.5 million, projected 160.3 million for 2024, growth of 9.8 million or 6.5%, and the major occupational group table.
- US Bureau of Labor Statistics, Industry and occupational employment projections overview and highlights, 2024 to 2034, Monthly Labor Review, 2026. Base year employment of 170.0 million, projected 175.2 million for 2034, and the statement of 13.0% employment growth recorded over the 2014 to 2024 decade; office and administrative support projected down 3.9%.
- US Bureau of Labor Statistics, Employment Projections news releases for the 2019 to 2029, 2021 to 2031, 2023 to 2033 and 2025 to 2035 rounds. Base and projected employment levels for each round, the note that the 2021 base year does not reflect much of the post-pandemic employment recovery, and the statement that 2015 to 2025 recorded 10.9% growth against 3.5% projected for 2025 to 2035.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
