Mid-Year Financial Forecasting Techniques for CFOs

8/10/2026

Most mid-year reforecasts are unreliable for one structural reason, and it is not a modelling problem.

The forecast and the budget are being treated as the same document.

A budget is a commitment — a target, tied to accountability and frequently to compensation. A forecast is a prediction — the best available estimate of what will actually happen.

Once those merge, the forecast becomes a negotiation. Nobody submits a number below their target, because doing so is an admission; nobody submits a number above it, because it will be taken as a new target. What arrives at the CFO is therefore a document about incentives rather than about the business.

Separate them explicitly, in language and in process. Say out loud that the reforecast will not change anyone's target, and then do not change anyone's target on the basis of it. Everything below depends on that.

Build From Drivers, Not From Last Year's Lines

The technical distinction that determines whether a forecast is useful.

Line-item extrapolation takes each account and adjusts it — up a bit, flat, down a bit. It produces a plausible-looking model that cannot answer a question, because none of the numbers are connected to anything.

Driver-based forecasting models the small number of quantities that actually determine the outcome, and derives the rest. For most businesses that is: units or transactions, price, headcount, and a handful of cost relationships.

The test of whether you have built a driver model: can you answer "what if volume comes in 10% lower" without opening every worksheet? If not, you have a spreadsheet, not a model.

Two related disciplines: keep one source for each input, so a change propagates; and eliminate hardcoded overrides, which are where models silently rot. Practical technique sits in the Essential Excel Skills course, the Excel training for accountants catalog, and the High Impact Excel dashboard session.

The Revenue Line Is the Whole Problem

Everything downstream inherits the revenue assumption, and the revenue assumption comes from the least reliable source in the organization.

Three practices that improve it:

Get it from the people who own the customers, not from a growth percentage applied in finance — but then adjust for known bias, because a sales organization's forecast has a direction and it is consistent. Which brings us to measurement below.

Decompose it. Existing customers at current volumes, plus expansion, minus churn, plus new — each of which behaves differently and is estimable separately. A single growth rate hides all four.

Forecast a range, not a point. A point estimate implies precision the input does not have, and it makes the whole model brittle. Present a range with the drivers that move it.

For a business with contracted or recurring revenue, the base is far more forecastable than the new business, and the model should show them separately so the reader can see how much of the year is already determined.

The Cash Forecast Is a Different Model

The point most often missed, and the one that matters when conditions tighten.

A profit forecast is not a cash forecast, and you cannot derive a usable one from the other by adjusting for depreciation. They answer different questions on different timescales:

The P&L forecast answers "what will the year look like" and is monthly or quarterly.

The cash forecast answers "will we be able to pay for things" and needs to be weekly, built directly from expected receipts and disbursements rather than indirectly from earnings.

Build the short-horizon cash forecast the direct way: opening balance, collections by week based on actual invoice aging and customer payment behaviour, payroll on its actual dates, supplier payments by term, debt service, tax payments, and capital expenditure. It is unglamorous and it is the model that prevents a crisis.

Three things it reveals that a P&L forecast conceals: timing troughs within an otherwise good quarter; the cash consumption of growth, since working capital absorbs cash as a business expands — per our post on working capital analysis; and the effect of a single large customer paying late.

Supporting material sits in the Guide to Cash Management course.

Scenarios With Triggers Attached

Most scenario planning is decoration: three columns at plus and minus some percentage, which nobody uses.

Useful scenarios have three properties:

They are built by changing drivers, not outcomes. "Volume down 15% and price flat" is a scenario. "Revenue down 15%" is an arithmetic exercise.

They have observable triggers. Not "if things get worse" but a specific, watchable indicator — a monthly booking level, a pipeline figure, a customer decision, a covenant headroom threshold.

They have pre-agreed actions. What the business will do in each case, decided now while everyone is calm, with an owner and a lead time. The value of a downside scenario is almost entirely in having decided the response in advance, because the decisions that need making in a downturn need making faster than a management team can agree them.

Three cases is usually enough: the expected case, a downside with actions, and an upside — the last of which matters more than people think, because being unprepared for growth is a real failure mode with its own cash consequences.

Variance Analysis That Says Something

The input to a good reforecast is a decomposed explanation of the first half, not a table of differences.

Separate price, volume, and mix. A revenue shortfall from lower volume, from discounting, or from a shift toward lower-margin products requires three different responses, and a single variance number supports none of them.

Separate timing from permanent. A deferred order and a lost order look identical in a monthly variance and mean entirely different things for the rest of the year. Making this distinction explicitly is the highest-value step in the whole exercise.

Separate controllable from external.

And ask what it implies for the remaining periods, which is the only reason to do the analysis. A variance explanation that does not change the forward forecast has not been used.

Analytical grounding is covered in analyzing financial statements, the Certificate in Financial Reporting and Analysis, and — for the earnings measure itself — understanding EBITDA.

Measure Bias, Not Just Accuracy

The improvement discipline almost nobody runs, and the cheapest one available.

Tracking forecast error tells you the model is imprecise, which you knew. Tracking forecast bias — the direction and consistency of the miss — tells you something actionable:

A persistently optimistic forecast is a management behaviour, not a modelling defect. No model change fixes it. What fixes it is naming it, showing the pattern over several periods, and — if necessary — applying a documented adjustment to that source's submissions.

Bias by source is more useful than bias in aggregate, because it is usually concentrated in one or two contributors.

Bias by line matters too: organizations are typically optimistic on revenue timing and on cost control simultaneously, which compounds.

Keep the history. A forecast archive that shows what was predicted and when is what makes this possible, and it takes no effort beyond not overwriting the file.

Debt Covenants Deserve Their Own Forecast Line

Short and non-negotiable for a leveraged business.

Forecast the covenant calculations explicitly, with headroom shown, on the definitions in the agreement — which are frequently not the definitions in your management reporting, and the difference has caught many finance teams.

Two rules: run the covenant under the downside scenario, not just the expected case; and if a breach appears in the forecast, that is a board and lender conversation early, not a problem to manage quietly and hope out of. Lenders respond very differently to a projected breach disclosed in advance than to one discovered at a reporting date.

Who Owns the Number

A process question that determines the quality of everything above.

The business owns the forecast; finance owns the process. A forecast produced by finance and imposed on operators is not believed and not acted on. One produced by operators with finance providing the model, the challenge, and the consolidation is.

Which requires finance to actually challenge — asking what changed, what the assumption rests on, and what would have to be true — rather than collecting submissions.

And keep it out of the compensation calculation. A forecast that affects a bonus is a negotiating position. This is the same point as the opening one and it is worth making twice, because organizations announce the separation and then quietly violate it.

The Mid-Year Specifics

What makes the mid-year reforecast different from any other:

You have half the year in actuals, so the full-year landing zone is largely determined. Show it that way: actual to date plus forecast to go, with the proportion of the year already locked.

The decisions still open are identifiable — hiring, capital expenditure, discretionary spend, pricing actions, and the timing of initiatives — and those are what the reforecast exists to inform.

There is time for the actions to matter. A cost action taken in July affects the year; the same action in November does not.

And next year's planning starts from this, so a reforecast built properly is most of the work of the annual budget rather than a separate exercise.

Where Reforecasts Fail

Treating the forecast as a commitment, so submissions become negotiations
Tying it to compensation, guaranteeing bias
Extrapolating line items instead of modelling drivers
A model that cannot answer a "what if" without manual rework
Hardcoded overrides that break propagation
A single revenue growth rate, hiding base, expansion, churn, and new business
A point estimate where the input does not support precision
Deriving cash from profit rather than building a direct weekly cash forecast
Ignoring working capital consumption in a growth forecast
Scenarios that change outcomes rather than drivers
Scenarios with no observable trigger and no pre-agreed action
No upside plan, leaving the business unprepared for growth
Variance tables with no price/volume/mix decomposition
Not separating timing from permanent variances
Variance analysis that never changes the forward forecast
Measuring error but never bias, so a systematic optimism persists
No forecast archive, making bias unmeasurable
Covenants forecast on management definitions rather than the agreement's
A projected breach managed quietly instead of disclosed early
Finance producing the forecast and operators not owning it

The summary for a CFO in July: say explicitly that the reforecast will not move anyone's target, and then do not move it — because that one commitment determines whether the numbers you receive are predictions or negotiations. Build from drivers so the model can answer a question, keep a separate weekly direct cash forecast, attach triggers and pre-agreed actions to the downside case, and start measuring forecast bias by source, since systematic optimism is a management problem no model change will fix.

Frequently Asked Questions

Why are mid-year reforecasts usually unreliable?

Because the forecast and the budget are treated as one document. A budget is a commitment tied to accountability and often compensation; a forecast is a prediction. Once merged, nobody submits a number below their target because that is an admission, or above it because it becomes a new target — so what reaches the CFO is a document about incentives rather than the business.

What makes a forecast model actually useful?

Driver-based construction rather than line-item extrapolation. The test is whether you can answer "what if volume comes in 10% lower" without opening every worksheet. If not, it is a spreadsheet rather than a model. Keeping one source per input and eliminating hardcoded overrides is what keeps it that way.

Can a cash forecast be derived from the profit forecast?

Not usefully. They answer different questions on different timescales, and adjusting earnings for depreciation does not produce a usable cash view. The short-horizon cash forecast should be built directly and weekly — collections from actual invoice aging and payment behaviour, payroll on its real dates, supplier payments by term, debt service, tax, and capital expenditure.

What makes a scenario worth building?

Three things: it changes drivers rather than outcomes, so "volume down 15% and price flat" rather than "revenue down 15%"; it has an observable trigger, such as a booking level or covenant headroom threshold rather than "if things get worse"; and it has pre-agreed actions with owners, decided now, because downturn decisions need making faster than a management team can agree them.

What should be measured to improve forecasting?

Bias rather than accuracy. Error tells you the model is imprecise, which you already knew. Bias — the direction and consistency of the miss, tracked by source and by line — reveals systematic optimism, which is a management behaviour that no model change fixes. Naming it and showing the pattern across several periods is what changes it.

What is different about the mid-year reforecast specifically?

Half the year is in actuals, so the full-year landing zone is largely determined and should be presented that way — actual to date plus forecast to go. The decisions still open are identifiable and are the reason the exercise exists, there is still time for cost or pricing actions to affect the year, and a properly built reforecast is most of the work of next year's budget.