What the research can and cannot say about plan decay
Published data measures forecast drift and shrinking budgets, not how long a plan lasts. Here is what those figures can and cannot tell a board about re-basing.
No study in our sources says how long a strategic plan stays valid. We looked for measured decay rates for market, cost and competitor assumptions, and for data on how often boards re-base their plans. The fact sheet contains neither. What it does contain is evidence on a neighbouring question: how far forecasts drift from what happens, and how quickly a fixed allowance runs down. That is useful to a leader, provided it is not mistaken for a measure of plan life.
- Shortfall of 2024 US health spending against 2010 CMS projections, per NBER
- 15 percent
- US medical spending as a share of GDP in 2024, per NBER working paper 35231
- 18.0%
- Likelihood level behind carbon budget estimates, per Our World in Data
- 50%
- Projected fall in global fossil CO2 in 2026, preliminary Global Carbon Project estimate
- 0.4%
What the data shows
The clearest evidence on forecast drift comes from US health spending. According to NBER working paper 35231, by David M. Cutler and Lev R. Klarnet, US medical spending rose from 17.2 per cent to 18.0 per cent of GDP between 2010 and 2024. That growth ran below both the country's own history and the projections made by CMS, whose forecasts the paper uses as its benchmark. By 2024, annual health expenditure was nearly $1 trillion, or 15 per cent, below the 2010 CMS projections, the NBER says. The authors adjust for gaps between anticipated and realised figures. The method is truncated on the page we could read, so we do not describe it further.
A projection is an assumption with a publication date. The health figures show what happens when that assumption is carried forward unchanged for years: the gap between expectation and outcome grows until someone measures it. The lesson for a company is not that forecasters are careless. A plan built on a third party's projection inherits that projection's error, and the error is rarely written down. Ask who made each forecast in your plan, when they made it, and whether anyone has checked it against actual results since.
Carbon budgets show a different kind of decay: a fixed allowance that shrinks while it is being used. According to Our World in Data, the 1.5°C carbon budget runs out around 2030 at current emission rates, and the 2°C budget lasts until mid-century. The budgets shrink each year emissions continue. The site adds that the budgets are 50 per cent likelihood estimates, so a budget guaranteed to be safe would be much smaller. According to a preliminary Global Carbon Project estimate, relayed by Our World in Data, global fossil CO2 is projected to fall about 0.4 per cent in 2026, and emissions have not yet peaked. The direction differs by country. Emissions are projected to rise about 4 per cent in India and to fall in China, the US, the EU and Japan.
The two cases fail in opposite directions. In health spending, the realised figure came in below the projection. In carbon budgets, the margin was shrinking under anyone using it. A plan can be wrong because the world is smaller than expected, or because the room it needs is disappearing. Test for both before you decide which error your plan is most exposed to.
Figures to compare
Each figure is taken from the page named in its source link. None is a measure of how long a strategic plan stays valid.
- Shortfall of 2024 US health spending against 2010 CMS projections (NBER)
- 15 percent
- US medical spending as a share of GDP in 2010, per NBER working paper 35231
- 17.2%
- Year the 1.5°C carbon budget runs out at current emission rates (Our World in Data)
- around 2030
- Projected fall in global fossil CO2 in 2026, preliminary estimate (Our World in Data)
- 0.4%
Source: National Bureau of Economic Research; Our World in Data
What it means for a leader
Treat every assumption in a plan as something with a date on it. Most plans carry three kinds: market growth, cost paths and the position of named competitors. None of our sources gives a typical rate at which these go stale, and we would not invent one. In our judgement, the useful habit is simpler. For each assumption, write down when it was last checked, against what measure, and what observation would show it wrong. An assumption with no checking date is already a candidate for re-basing.
Concentration is a second hidden assumption. The European Central Bank reports that US hyperscalers have rapidly raised their share of long-term euro corporate bonds, leaving investors more exposed to a few issuers. It describes the growing hyperscaler presence as “still in its early stages”, and says it could amplify portfolio losses if AI-related revenues disappoint. A plan that depends on a handful of suppliers, buyers or platforms carries the same exposure. Its growth assumption is one bet, however many line items it appears in.
A forecast review does not need a research team. Take the five or six forecasts that drive the largest decisions in your current plan. For each, pull the figure you used, the figure that actually happened, and the gap between them. Do this for the last two budget cycles. If the gap runs one way every time, your forecasts are biased, and the plan should be re-based on that evidence rather than on the next optimistic round.
Deciding when to stop is the part of planning most boards leave for a later meeting. A National Bureau of Economic Research bulletin reports that entrepreneurship training may help founders recognise when a venture no longer merits further investment, and that it could shut some startups sooner. The study behind it, NBER working paper 34755 by Bailey, Fehder, Floyd, Hochberg and Lee, is a randomised, multi-year, multi-site field experiment with innovation-driven entrepreneurs. The lesson transfers even though the setting does not. Write the stop condition before the money goes out, with a number and a date.
Before you re-base a plan on a headline figure, check what the figure counts. NBER working paper 35551 reports patent filings in China at about three times the US count, and records that observers dispute whether this reflects real innovation. NBER also notes that one group argues headline patent statistics overstate China's innovative progress. Separately, other observers have taken a low-cost AI model from DeepSeek in 2025 as evidence of rapid technological advance. Two readings of one number are a signal to find a third measure before changing the plan.
Set the review rhythm by assumption, not by calendar. The sources give no re-basing frequency, so we cannot say how often a plan should be revisited. Our judgement is that an assumption whose measure moves sharply should trigger a review whatever the date. A quiet assumption can wait for the annual cycle. This is a rule of thumb, not a finding, and a board should test it against its own record of forecast misses.
Expectations are an input too. Pew Research Center reports pessimism about AI's effect on jobs and economic inequality across a 37-country survey. In the United States, 75 per cent of Democrats and 68 per cent of Republicans predict AI will mean fewer jobs over the next 20 years, according to the same survey. A labour-cost assumption built on sentiment rather than data is weak in both directions. Survey answers signal that opinion may shift. They are not forecasts of wages or headcount.
Limits of the data
The brief set out to answer a narrower question than the sources can. The fact sheet contains no strategy-process study and no management-practice survey, so there is no measured decay rate for plan assumptions and no sourced re-basing frequency. The figures above measure national health spending against forecasts, global emissions against budgets, bond holdings, and public opinion. The NBER health method was not available to us in full. The Stanford Graduate School of Business item is a headline only, with no findings. Read these figures as evidence about forecasts in general, not as a measure of how long your plan will hold.

