The Death of the Annual Budget: Continuous 18-Month Rolling Forecasts

Why the annual budget expires

Not because it is wrong, but because it is fixed. The world it was built for lasts a quarter; the document lasts a year.

Consider what a budget is asked to be. It is a target ("hit this revenue"), a resource envelope ("spend no more than this"), a forecast ("this is what we expect"), and a performance yardstick ("you missed by 8%") — four jobs in one spreadsheet. The trouble is that the jobs pull in opposite directions. A target should be stretching; a forecast should be honest. An envelope should be stable; a forecast should move with the facts. When one number tries to be all four, it becomes a negotiation rather than an estimate, and once negotiated it cannot be changed without reopening the negotiation.

Figure 1. The visibility problem, drawn. A fixed annual plan is a countdown; a rolling one is a constant. In the fourth quarter most companies run on no forward plan at all while the next budget is negotiated.

The practical symptoms are familiar to anyone who has sat through a quarterly business review: variance explanations that amount to "the budget was built before we knew X"; managers protecting envelopes they no longer need because the number was fought for; and a leadership team that, in October, has a very detailed view of the eleven weeks to year-end and almost none of the following twelve months.

Decouple the target from the forecast

The fix is not a better budget. It is three artefacts instead of one, each doing a single job, each with its own owner and cadence.

Aspiration · set annually

Targets

Where leadership wants the business to be: revenue, margin, cash, headcount at year-end. Stretching by design. Set once, revised rarely, owned by the executive team. Compensation may reference them; forecasts must not be bent toward them.

Expectation · rolled quarterly

Rolling forecast

The best honest estimate of the next 18 months, refreshed as facts arrive. Driver-based. No penalty for a forecast that moves — the penalty is for a forecast that is wrong and was not updated. Owned by FP&A with functional inputs.

Authorisation · governed

Resource envelopes

What each function may commit — hiring approvals, capex, discretionary spend. Released in tranches against the forecast rather than fixed for twelve months. Owned by the CFO, governed by workflow.

This separation is the core of what the Beyond Budgeting literature has argued since Hope and Fraser's 2003 book of that name, and it is what most high-growth finance teams converge on independently. Its power is that it removes the incentive to lie. When the forecast no longer determines anyone's bonus or budget envelope, people forecast what they expect. When targets are explicit aspirations, nobody confuses "we set out to grow 30%" with "we expect to grow 30%". And when envelopes are released against the forecast, a function that forecasts honestly downward is not punished by losing money it had not yet spent.

The leader's objection: "but I need a number to hold people to"

You still have one — the target. What changes is that you stop holding people to a fifteen-month-old forecast as if it were a commitment. Accountability moves from "did you hit the number we agreed in October" to "did you see it coming, say so, and act". That is a harder standard, not a softer one.

Why 18 months, and how the horizon rolls

Six quarters forward is long enough to see past the current fiscal year at every point in it, and short enough that the far quarters are drivers rather than fiction.

Figure 2. The window moves; the shape inside it does not. Detail is concentrated where decisions are being made this quarter and thins toward the far end, where only the drivers are maintained. At every refresh the leadership team can see past the fiscal-year boundary.

Why not 12, why not 24

Twelve months rolling is the common compromise and it has a flaw: in the last quarter of the fiscal year it shows only one quarter of the next, which is exactly when the board wants to see the full next year. Eighteen months always contains the next full fiscal year. Twenty-four months sounds more strategic and in practice is not maintained — the far quarters become a copy-forward that nobody revises. Six quarters is the longest horizon most mid-size companies will keep honest.

Granularity that thins with distance

The current quarter is planned monthly, line by line, because it is being managed. The next two are monthly but driver-based — headcount, volume, price, not individual invoices. Quarters four and five are quarterly. The sixth is drivers only: it exists so that the model has a view beyond the year, not so that anyone spends a week filling it in. This is the single most important design decision in a rolling forecast, because a rolling forecast that demands twelve-month line-item detail every quarter collapses under its own weight within two cycles.

Cadence and versions

Continuous does not mean constant. It means a light monthly touch and a deliberate quarterly re-plan, each producing a named version that nobody edits afterward.

Figure 3. The rhythm. Monthly work is re-basing, not re-planning. Quarterly work produces a locked version with a name, so that "the forecast" always means a specific, dated artefact. Target-setting is a separate annual event that does not interrupt the rolling cycle.

Version discipline

Every locked forecast keeps its name forever — Q1 RF, Q2 RF — and is never edited after locking. Working copies are just that, and are visibly labelled as such. This sounds bureaucratic and is in fact the thing that makes the whole approach auditable: a board can ask "what did we expect in April, and what changed by July?" and get an answer from two versions rather than from memory. It is also what makes forecast accuracy measurable, which is the only way to improve it.

Drivers, not line items

A rolling forecast that asks managers to re-estimate eight hundred GL lines every quarter will be abandoned by the third cycle. One that asks them to update thirty drivers will survive.

The far quarters in Figure 2 exist only because they are computed, not entered. Revenue is volume times price by segment; personnel cost is headcount times average compensation by band, with hiring lags; cloud cost is usage times unit rate; facilities is square footage times rate. A functional owner updates the handful of numbers they actually know — how many people they plan to hire and when, what volume they expect — and the model does the arithmetic across the horizon. The mechanics of driver modelling are a subject in themselves and have their own article in this library; the point here is that rolling forecasting and driver-based modelling are not two initiatives. The first is impossible without the second.

The test of a good driver set

Each driver should be a number a named person can defend, that changes the forecast materially when it moves, and that nobody has an incentive to misstate. Headcount plan by month passes all three. "Other opex as a percentage of revenue" passes none.

Cash elasticity: the forecast that matters most

For a growing mid-size company, the rolling forecast's first customer is the cash line. Everything else is context for it.

Figure 4. Cash elasticity: how much the cash balance moves per unit change in each driver. Hiring timing, bookings and collection days are usually the three that matter; the rolling forecast should show their effect as a band, not a number.

A rolling forecast makes cash planning possible in a way an annual budget never can, because the drivers that move cash — hiring dates, bookings, collection days, capex timing — are exactly the ones being refreshed each quarter. The discipline is to publish, with every version, not just the base cash line but its elasticity: how much cash moves if hiring slips two months, if bookings come in 15% under, if DSO stretches ten days. That turns the cash forecast from a number the board is asked to believe into a range the board can reason about, and it makes the decision to raise, borrow or slow down a calculation rather than an argument.

What it looks like in EAConnect Planning

A rolling forecast needs five things from a platform: a scenario dimension with versions, a time dimension that runs past the fiscal year, driver-based entry, a statement that shows target, actual and forecast together, and a workflow to lock versions. Each is shown below as far as the screens allow.

P&L statement showing Q1 Budget, Q1 Actual, FY Budget, FY Forecast, variance and variance percent columns with colour-coded favourability

Figure 5. The P&L view with target and forecast side by side: Q1 Budget, Q1 Actual, FY Budget, FY Forecast, variance and variance %. This is the "three artefacts" of section 02 on one screen — the budget column is the target, the forecast column is the rolling expectation, and the variance is between them rather than between actual and a stale plan.

Planning entry grid with BUD and FCST columns per month by account, a subsidiary selector and a WORKING version selector

Figure 6. The entry grid: Budget and Forecast columns by month, per account and subsidiary, with a version selector set to WORKING. Forecast cells are entered or driver-computed; the version is locked through the review workflow when the quarter's re-plan is complete.

Scenario and version as dimensions

Actual, Budget and Forecast are members of a scenario dimension; named versions (Q1 RF, WORKING) sit alongside. Selecting a column is a dimension choice, not a model rebuild — which is what makes side-by-side comparison of any two versions routine.

A time dimension from a calendar

The time dimension is generated from a configured fiscal calendar and extends as far forward as the calendar does, so an 18-month horizon crossing the fiscal-year boundary is a property of the calendar, not a workaround.

A statistical baseline for the far quarters

The platform's forecasting engine — described in its own article — produces a baseline with intervals per series. It is the natural source for quarters four to six, where nobody should be typing numbers.

Workflow to lock a version

Submission, review and approval steps run on the same process engine as data loads. A version becomes a named, locked artefact when the approval step completes, and the audit trail records who approved it and when.

Making the transition

Companies do not switch from annual budgets to rolling forecasts in a single step, and should not try. The path that works runs one full cycle in parallel with the existing budget before retiring it.

Stage	What to do	What to expect
1 · Separate the artefacts	Declare this year's budget the target. Start a rolling forecast alongside it, owned by FP&A, with no link to compensation or envelopes yet.	Resistance is low because nothing anyone is measured on changes. The forecast can be honest from the first cycle.
2 · Build the driver set	Thirty drivers or fewer, each with a named owner. Headcount plan, volume, price, collection days, capex dates. Everything else is a rate on one of these.	Two to three weeks of functional conversations. This is the real work and the platform cannot do it for you.
3 · Run two quarterly cycles in parallel	Lock Q1 RF and Q2 RF. Report variance to target and to prior version. Do not yet release envelopes against it.	By the second version the organisation can see the forecast moving for reasons it understands. Trust builds from that, not from a memo.
4 · Move envelopes onto the forecast	Release hiring and discretionary spend quarterly against the locked version rather than annually against the budget.	This is the step that changes behaviour. Managers stop protecting envelopes and start forecasting.
5 · Retire the budget-as-forecast	Next year, set targets in a two-week event. The forecast is already rolling; there is no budget cycle to run.	The autumn budget season — typically six to ten weeks of organisational effort — does not happen.

In the four-week implementation ladder used across this library, stages 1–2 are week two (model and drivers) and the first locked version is week four. Stages 3–5 are a matter of calendar, not effort: they take two quarters because two quarters have to pass.

A practitioner's view for leaders

Opinions from people who have run and implemented planning processes in mid-size and enterprise companies. Not product claims.

The budget is not the problem; using it as a forecast is

Keep an annual target-setting event. Keep envelopes. Stop pretending a number negotiated in October is your best estimate in April. Three artefacts, three jobs.

Decouple before you roll

A rolling forecast that still determines bonuses and budgets will be gamed exactly as the budget was, just four times a year. Separate the incentives first; the honesty follows.

Thin the far quarters or they will thin themselves

Line-item detail eighteen months out is fiction typed by tired people. Drivers only, computed, and reviewed for reasonableness. Save the detail for the quarter you are managing.

Lock versions and never edit them

The value of a rolling forecast compounds only if you can compare versions. A forecast that gets "corrected" after the fact is a budget with a shorter memory.

Publish cash as a band

The board's real question is whether the plausible downside crosses the covenant. Answer it every quarter with elasticities, not a single line and a reassuring tone.

Measure forecast accuracy, and mean it

Version against actual, by driver, by owner, every quarter. Accuracy improves when it is measured and attributed. It does not improve from exhortation.