The annual budget is an artefact of a slower world. Set in the last quarter, approved for twelve months, and treated as the standard against which performance is measured for a year.
For a company growing at 100% or more, it is often obsolete by month four. Not because the planning was poor, but because the assumptions it rested on were superseded by real information.
It anchors to stale assumptions. A budget built in November on a plan to hire twelve engineers assumes a hiring rate. When you hire four by March, the entire cost base is wrong, and every variance report compares against a plan nobody believes.
It encourages spending to the number. Departments that underspend often lose the allocation next year, so they spend. The budget becomes a floor rather than a ceiling.
It creates a negotiation, not a forecast. If the budget determines resources, managers bid for allocation rather than estimating honestly. The output is a political document.
It goes stale silently. Everyone knows the budget is wrong by mid-year. Nobody says so, because re-forecasting is a project.
A forecast that always looks the same distance ahead — typically four to six quarters — and is updated on a regular cycle, usually monthly or quarterly. When Q1 completes, Q1 of the following year is added.
Two properties matter more than the mechanics.
It is always current. The forecast reflects what you know today, not what you believed in November.
It is separated from the commitment. The forecast says what you expect. The budget, if you keep one, says what has been approved. Conflating them is what makes forecasts political.
Forecast drivers, not line items. Do not forecast "AWS cost" directly. Forecast customers, usage per customer and unit cost, and let the number derive. When reality moves, you change the driver and everything downstream updates.
Keep the model coarse. A forecast with four hundred lines is not more accurate; it is more expensive to maintain and harder to explain. Detail should match the materiality of the line.
Fix the cycle. Same days each month, same owners, same inputs. A forecast that happens when someone has time will not happen.
Track forecast accuracy. Compare each forecast to actual and record the variance. This is the discipline that turns forecasting from an exercise into a capability — you learn which drivers you consistently misjudge.
Keep scenarios to three. Base, downside, upside. More than that and nobody can hold them in mind, and the exercise becomes modelling for its own sake.
This is not an argument for abolishing it.
Boards approve annual budgets, and that approval is a governance mechanism with real value. Cost control benefits from a fixed reference point. Some obligations — lease commitments, insurance, statutory costs — genuinely are annual. And a company at a stable growth rate in a predictable market may find the annual cycle entirely adequate.
The pragmatic answer for most growth-stage companies is both: an annual budget as the approved commitment and control reference, and a rolling forecast as the operating view of where things are actually heading.
Planning connected to the ledger matters more here than sophisticated modelling features, because the binding constraint is usually the effort of keeping the forecast current.
We connect planning to actuals during implementation so re-forecasting is a review rather than a rebuild. That is usually what determines whether a rolling forecast survives its first busy quarter.
Get in touch, or see cash flow forecasting for subscription businesses for the cash side.