Discover how leading CFOs tackle forecasting challenges with agile strategies, advanced analytics, and data-driven insights to drive business success.
Forecasting is the part of the finance function that almost everybody says is important and almost nobody does well. The annual budget gets built, it is wrong within a quarter, and the business runs on a document everybody has privately stopped believing.
The problem is rarely the technique. Finance teams know how to build a forecast. The problem is that the process is structured in a way that guarantees it will be stale, and that the capacity required to keep it current is consumed by producing last month's numbers.
This article covers why traditional forecasting fails, what the finance leaders who have solved it actually do differently, and what has to change structurally for it to work.
The annual budget is the default forecasting mechanism in most businesses and it has structural problems that effort cannot fix.
It is built once, months before the year starts, against assumptions that were reasonable at the time and are unlikely to survive.
It is built by negotiation rather than by analysis, since every department has an incentive to secure resource and a reason to be conservative about delivery.
It becomes a performance measure, which means people manage to it rather than reporting honestly against it, and the information quality degrades accordingly.
And it has no mechanism for updating, so by the second quarter everybody knows it is wrong and there is no process for saying so.
The alternative most finance leaders eventually arrive at is a rolling forecast, updated continuously, always looking the same distance forward.
Because it is updated regularly, it reflects what is actually happening rather than what was assumed nine months ago.
Because it always looks the same distance ahead, the planning horizon does not shrink as the year progresses, which is a genuine problem with an annual budget in the final quarter.
Because it is separate from the performance measure, people can be honest in it, which is the single largest determinant of forecast quality.
And because it updates, the conversation shifts from explaining variance to deciding what to do, which is what forecasting is for.
The second structural change is building the forecast on what actually causes the numbers rather than on the numbers themselves.
Most forecasts are last year plus a percentage, which encodes no understanding and cannot explain itself when it is wrong.
A driver based forecast starts from the operational quantities. Units, customers, transactions, hours, headcount, whatever the business actually runs on.
Revenue then follows from volume and price. Cost follows from volume, headcount and rate. The financial statements are an output rather than an input.
The practical benefit is diagnostic. When reality diverges, you can see which driver moved, which tells you what is actually happening in the business rather than that the number was wrong.
A common failure is building a driver model so detailed that maintaining it becomes its own job.
Most businesses are adequately described by a small number of drivers. Five to ten is typical, and beyond that the additional precision is usually illusory.
Choose them by testing which quantities actually move the result. A driver that varies by a few percent and affects one line is not worth modelling separately.
And choose them for observability. A driver that nobody measures cannot be forecast and cannot be tracked against, which makes it a guess with extra steps.
The discipline of keeping the model small is what makes it maintainable, and a maintainable model that is approximately right beats a detailed one that is updated annually.
A single point forecast is always wrong, and presenting it as a plan creates a false sense of precision.
Building three cases, a base, a downside and an upside, communicates the uncertainty honestly and is considerably more useful for decisions.
The downside case is the one that matters most, because it answers the question a board actually has, which is what happens if things go badly and can we survive it.
Building scenarios also forces you to state the assumptions explicitly, since a scenario is defined by which assumptions change.
And it changes the conversation from whether the forecast is right to which assumptions are most likely to break and what the business would do about each.
A forecast that projects profit and not cash misses the constraint that actually binds most growing businesses.
Profit is recorded when the work is done. Cash arrives when the customer pays. In a growing business the gap widens continuously, because you are funding more work in progress and more receivables every month.
Which means growth consumes cash, and businesses that fail while growing usually fail for this reason rather than for any failure of the underlying model.
A rolling thirteen week cash forecast, built from the actual receivables and payables ledgers rather than from the profit forecast, is the artefact that most changes how a business makes decisions.
It shifts the question from whether you can afford something to when, which is a considerably more useful question.
The reason most finance functions do not forecast well is not technique. It is that the capacity required is consumed by producing historical reporting.
Where month end takes most of a fortnight, the remaining time goes to compliance and the queries the close generated. There is no space left.
Which means improving forecasting usually starts by compressing the close, which is a systems and process problem rather than a forecasting one.
Automating the rules based work, eliminating reconciliation between systems, and building reporting that refreshes rather than being rebuilt each month all free the capacity that forecasting needs.
Finance leaders who have solved forecasting almost always solved the close first, and describe the order as unavoidable rather than optional.
Beyond capacity, the structure of your data determines whether driver based forecasting is feasible at all.
Driver based models require the operational quantities and the financial results to be joinable, so that the historical relationship between them can be established.
Where sales sits in one system, operations in another and finance in a third, establishing that relationship is a project rather than a query, which means it happens once and then stops being maintained.
Where the business runs on one platform with consistent dimensional coding, the same analysis is a saved search, which means the model can be recalibrated against actuals routinely.
That difference determines whether a rolling forecast is sustainable or whether it becomes another initiative that ran for two quarters.
A forecast built entirely by finance is a finance opinion, and it is usually worse than one built with the people who actually influence the drivers.
Sales knows the pipeline and the likelihood of individual opportunities better than any model does.
Operations knows capacity constraints, lead times and what is genuinely achievable.
The difficulty is that involving them introduces bias, since people forecast optimistically about their own area, and the mitigation is calibration rather than exclusion.
Track how each contributor's forecasts have compared to actuals over time, and adjust for the observed bias. That is more useful and considerably less confrontational than arguing about individual numbers.
This is the single change that most improves forecast quality and the one businesses find hardest.
Where the forecast is also the performance measure, people forecast what they want to be held to rather than what they expect, and the information is systematically distorted.
Where the two are separate, the forecast can be an honest expectation and the target can be an ambition, and everybody understands which is which.
That requires leadership discipline, because the temptation to hold somebody to their forecast is strong and giving in to it destroys the mechanism permanently.
Businesses that make this separation report an immediate improvement in forecast accuracy, which is unsurprising since they have stopped punishing honesty.
Most businesses never measure how good their forecasts are, which means they never improve.
Track the variance between forecast and actual by period and by driver, over time, and look for pattern rather than for individual misses.
Consistent bias in one direction is the most useful finding, because it is correctable. A forecast that is always fifteen percent optimistic can be adjusted.
Random error is less correctable and it tells you something about how predictable the business genuinely is, which is worth knowing.
And measure accuracy by horizon, since a forecast that is good one month out and poor six months out is a different problem from one that is uniformly poor.
Most forecasts stop at the profit and loss, which leaves out the part that determines whether the plan is fundable.
Working capital is the largest omission. Receivables, payables and inventory all move with volume, and a growth forecast that does not model them understates the cash required to deliver it.
Capital expenditure is the second, since growth frequently requires capacity, and the timing of that spend matters more than the amount.
Debt and covenant headroom is the third, and it is the one that turns a forecasting exercise into a financing conversation. A plan that breaches a covenant in month eight is worth knowing about in month one.
None of this requires a sophisticated model. Three or four working capital assumptions expressed as days, applied to the driver based revenue and cost, produces a usable balance sheet forecast in an afternoon.
A forecast that nobody reads has no effect, and the presentation matters more than the modelling.
Keep it short. A small number of drivers, the resulting financial position, the cash profile and the key assumptions.
Show the change since last time, because what moved and why is more informative than the current position.
State the assumptions explicitly and name which one is most likely to break, since that is where the useful conversation is.
And write commentary that explains rather than restates, because a variance table describes what happened and a sentence about why is what turns it into management information.
Forecasting fails more often through neglect than through poor technique, so the cadence matters.
Monthly is the usual cycle for the full forecast, aligned to the close so the actuals are available to recalibrate against.
Weekly for the cash forecast, since cash moves faster and the value of the artefact depends on it being current.
Quarterly for a deeper review of the drivers themselves, asking whether the model still describes the business.
And immediately whenever something material changes, since a forecast that does not respond to a significant event is not being used as a decision tool.
Some variables are genuinely unpredictable and pretending otherwise damages credibility.
Name them explicitly rather than burying an assumption in a model. Exchange rates, commodity prices, the timing of a large contract, a regulatory decision.
Model the range rather than the point, and show what the business looks like at each end.
Identify the trigger points, meaning the level at which the business would need to act, and decide in advance what that action would be.
That converts uncertainty from something that produces anxiety into something with a plan attached, which is the most useful thing forecasting does.
Forecasting improves through structural change rather than through effort, and the sequence is fairly consistent across the finance functions that have improved it.
Compress the close first, because that is what creates the capacity, and it is achievable through automation and process discipline without a large project.
Then build the rolling cash forecast, since it is the artefact with the fastest and most visible payoff.
Then move to driver based forecasting, which requires the data to support it and is worth the foundation work.
And separate the forecast from the target, which costs nothing and improves quality more than any modelling technique.
Our pieces on automation and the finance function and the challenges facing finance leaders cover the surrounding ground, and planning and budgeting covers the mechanics.
If you would like help building a forecasting discipline in your own business, get in touch.
