They answer different questions. The mistake is running one of them as if it were the other.
The argument gets framed as a choice: keep the annual budget, or move to rolling forecasts. Whole consulting engagements are sold on it.
It is the wrong frame. They do different jobs, and almost every problem attributed to one of them is actually the result of asking it to do the other's.
The annual budget is a commitment. It is the agreement between the business and its owners about what will be delivered and what may be spent. It is deliberately fixed, because the whole point is that it does not move when the month is difficult. A budget that changes when performance changes is not a commitment, it is a description.
The rolling forecast is an estimate. It is the current best view of where the year lands, updated as reality arrives. It exists to inform decisions — hiring, spend, cash — and it is supposed to move. A forecast that never changes is not being maintained.
The failures follow directly:
The second is much more common and much more damaging, and it is a behavioural problem rather than a modelling one. If people are held to account for their forecast accuracy in the same conversation where they are held to account for performance, they will forecast the target. Every time.
A rolling forecast always looks the same distance ahead. In January you forecast to the following December; in February you forecast to the following January.
This is the actual mechanism and it is more than a technicality. A year-end-fixed forecast gets shorter as the year goes on, and by October it covers eight weeks — at which point it can no longer inform any decision with a lead time, which is most decisions worth making. Teams stop using it, correctly, and then everyone concludes forecasting does not work here.
Four to six quarters ahead is the usual range. Beyond six the numbers are scenarios rather than forecasts, and there is no point maintaining monthly granularity on them.
The reason most forecasts take three weeks to re-cut is that they are rebuilt at line-item level: someone in each cost centre revisits every row.
A driver-based forecast changes the inputs that actually move: headcount by month, average salary, volume, price, conversion rate, days sales outstanding. Everything else is calculated from them.
The practical test: can one person re-cut the whole forecast in an afternoon? If not, it will not be updated monthly, whatever the policy says — and a rolling forecast that is not updated is just an old forecast with an ambitious name.
This has a second benefit. When the forecast moves, a driver model tells you why: volume assumption down 4%, start dates pushed a month. A line-item model tells you the number changed.
Once you have both, report three columns: budget, latest forecast, actual.
The third column is the one most organisations do not keep, and it is the one that tells you whether to believe the other two.
Sometimes the budget genuinely is dead — a business is acquired, a market closes, a major assumption is invalidated.
Re-baseline formally when that happens: a decision, a date, a written reason, and the old baseline kept alongside the new one. What destroys variance reporting is not re-baselining, it is re-baselining quietly. Six months later nobody can say what was originally committed to, and the year's performance becomes unassessable.
The [Rolling Forecast Model](/products/rolling-forecast-model.html) is driver-based, rolls a fixed horizon forward, and reports budget, forecast, prior forecast and actual side by side — with a worked example where the forecast has been walking towards the truth for three months. £45.