Pick any company you have worked at for more than three years and ask the FP&A team to lay last year's plan over last year's actuals. Then the year before that. Then the year before that.
What you will see, with a regularity that should alarm everyone involved, is a gap of roughly twenty percent — sometimes more, sometimes less, sometimes high, sometimes low, but very rarely small. The gap will be discussed at the board level once, attributed to "macro," "execution," or "a tough comp," and then quietly absorbed into the next plan, which will also miss, in a different direction, by roughly the same amount.
This is not a failure of any individual planner. It is not a failure of any specific model. It is a structural property of how annual planning is done in most companies — and the fix, it turns out, is also structural. Three changes to the operating rhythm do most of the work. None of them require a new tool. None of them require more headcount. They require, mostly, the willingness to plan honestly instead of plan beautifully.
Why the miss is structural
Before the fixes, the diagnosis. The twenty-percent miss is not random. It has a shape, and the shape is the same in nearly every company we have looked inside.
The plan begins in late Q3, with bottoms-up inputs from each business unit. Each unit is being asked, simultaneously, two questions that cannot be answered with the same number: what do you commit to and what do you actually expect. The commit number, by the time it has been negotiated up by the CRO, sharpened by FP&A, and signed off by the CEO, is sitting somewhere between the unit's median forecast and its 80th-percentile aspirational case. It is the number that will be defended at the board. It is the number compensation gets pegged to. It is also, by construction, biased high.
The aggregation amplifies the bias. Twelve units each delivering an 80th-percentile case do not, on aggregation, produce an 80th-percentile case for the company. They produce something much closer to a 95th-percentile case, because the variance of the sum is less than the sum of the variances. The company has now committed, at the consolidated level, to a number it would never have committed to if the question had been asked directly.
That is one half of the problem. The other half is that the plan is then locked. It becomes the denominator for every variance report for twelve months. Reforecasts, when they happen, are often a one-line slide in the Q2 board pack. The plan does not move. The actuals do. The gap is "explained" rather than absorbed into a more honest view of where the year is heading.
By the time the year ends, the company has run, in effect, a single forecast made nine months before the period began, with no formal mechanism to update it as the world delivered new information. We would not run any other operationally important process this way. We do not insist on closing books once a year. We do not run inventory once a year. But we run the forecast that drives hiring, capex, and capital allocation once a year, and then act surprised when it misses.
Change one: calibrated ranges instead of point forecasts
The first structural change is to stop producing single-number plans and start producing ranges with stated confidence — 80% intervals, on the four or five numbers that actually drive the year.
We covered the philosophy of calibrated forecasting in a separate piece. The operational version, for annual planning, is narrower. It is this: the plan deck, on the page that shows revenue, should show three numbers — a low, a mid, and a high — with a clear note that the low and high together represent an 80% confidence interval. Meaning: the team producing it believes there is a roughly one-in-ten chance the year comes in below the low number, and a one-in-ten chance it comes in above the high.
This sounds modest. It is not. The first time a leadership team does this honestly, two things happen. The ranges are wider than anyone expected. And the conversation about the plan shifts from "is the number right?" to "what would push us toward the low end, and what would push us toward the high end?" The first question has no useful answer. The second is the one the plan should have been about all along.
There is a CFO objection here that deserves to be addressed directly: we still need a number, because compensation, because guidance, because the board wants one. True. The point is not to abolish the number. The point is to ensure the number sits on top of a visible range, so that when the year comes in at the low end, the company is not surprised, and when it comes in at the high end, the company is not over-celebratory. The number remains the commit. The range is what you plan against.
Change two: a reforecasting cadence built into the operating rhythm
The second change is to stop treating the plan as an annual artifact and start treating it as a living view that gets updated on a regular cadence — monthly or quarterly, depending on the volatility of the business, with the same rigor as a financial close.
The plan should be the version of the truth most recently updated against what the world has done. If it is older than your inventory count, it is not a plan. It is a souvenir.
The mechanics are not complicated. Each quarter, the FP&A team takes the original plan, lays the actuals against it, and produces a refreshed view of the rest of the year — same range format, same named drivers, with explicit notation of what has changed since the last version. The board gets the refreshed view. Compensation continues to run against the original commit, so the politics of "moving the goalposts" are defused. But operationally, hiring, capex, and capital decisions get made against the refreshed view, not against a plan that is now six months stale.
The hardest part of this is not analytical. It is governance. The original plan, in most companies, is sacred. Reforecasting it on a quarterly basis feels, to the people who fought to produce it, like admitting it was wrong. The reframe is: of course it was wrong, every plan is wrong, and the question is whether the company is operating against its best current view or against a stale one. The companies that get this right are usually the ones whose CFO has explicitly broken the conflation between the commit (which doesn't move) and the operating view (which must).
Change three: base-rate humility
The third change is the one almost nobody does, and it is probably the highest-leverage of the three. It comes from Bent Flyvbjerg's work on reference-class forecasting — originally developed for megaproject estimation, applicable directly to annual planning.
The idea, in plain English: before you finalize this year's plan, lay out the last five years of plans and actuals for this business, by this team, and look at the distribution. How much, on average, has the team overshot or undershot? In which direction does it tend to err? What is the standard deviation of the miss?
That distribution is the reference class. It is, statistically, a better starting point for the new plan than any bottoms-up build. Doug Hubbard makes the related point in How to Measure Anything: in the absence of other information, the historical track record of the forecaster is the single most valuable input you have. Most companies have this data. Almost none of them use it.
We have run this exercise with leadership teams more times than we can count. The pattern is consistent. The team's bottoms-up forecast, when overlaid against its five-year history, is almost always optimistic by a familiar amount — the same familiar amount, year after year, because the optimism is structural to how the bottoms-up is built. The reference class shows it. Nobody who has seen it can unsee it.
The operational version of this is a single page in the planning deck. Five columns: prior year plan, prior year actual, miss, reason given at the time, reason now. Five rows, one per year. The page takes a junior analyst a day to build. It changes the next planning conversation more than any other artifact we have seen.
What about reference-class objections?
The objection, when it comes, is always: this year is different. The market is different, the team is different, the product is different, the macro is different. The reference class doesn't apply.
It is worth taking the objection seriously, because sometimes it is correct. Sometimes the business has genuinely changed in a way that breaks the historical pattern. But the burden of proof, in a well-run planning process, should be on the team claiming the break. Why is this year different in a way that previous years weren't also different? Every prior year's planning team also believed their year was different. They were wrong, on average, by twenty percent. What is the specific, falsifiable reason the new plan should be granted an exception?
In our experience, about one year in seven actually justifies a break from the reference class. The other six, the team is doing what every other team in our experience has done — privileging the inside view of this year's plan over the outside view of how plans of this type have actually performed. The fix is not to abandon the inside view. It is to anchor it, explicitly, against the outside view, and to require that any deviation be defended.
A CFO-ready operating model
The composite, for a CFO who wants to install this in time for next year's planning cycle:
The annual plan produces an 80% confidence interval, not a single number, on the four to five numbers that drive the business. The point estimate sits on top of the range as the commit, but the range is what the operating decisions are made against.
The plan is reforecast on a quarterly cadence, with the same format and the same named drivers, against actuals. The reforecast becomes the operating view; the original commit remains the compensation and external-guidance view. The two are explicitly separated, and the language used inside the company makes the separation clear.
Each plan begins with a one-page reference-class analysis — the last five years of plans, actuals, and misses for this business — and the new plan is required to either align with the reference class or to defend, in writing, why it deserves to be treated as an exception.
That is the entire model. It is not technically sophisticated. The math is high-school level. The discipline is what most companies lack, and the cost of installing it is, mostly, the political cost of admitting that the prior planning process was producing a number that nobody actually believed.
A company that does this for three consecutive years will, in our experience, see its forecast miss compress from the structural twenty percent toward something closer to ten — not because anyone got smarter, but because the system stopped systematically lying to itself.
If your last three annual plans have missed in the way described, we help CFOs and FP&A leaders install the rhythm above before next year's cycle starts. The window for changing it is now; by Q3 the political momentum of the existing process will close it.
Writes about decision quality at Bayeseon. Reach the team at hello@bayeseon.com.