The CFO asks how accurate the revenue forecast was.

Finance pulls up the numbers. Forecast revenue was $400,000. Actual revenue was $400,000.

A perfect match.

Someone is probably preparing a PowerPoint slide about the team’s forecasting excellence.

There’s just one small problem. Every individual month was wrong.

January was overforecast. February was underforecast. March was overforecast. April was underforecast.

The mistakes canceled each other out so beautifully that the total variance came to zero.

Apparently, we’ve discovered a new forecasting methodology: be wrong in equal and opposite directions.

This is exactly why measuring forecast accuracy using the total variance can be misleading. The company might have landed on the right number, but that doesn’t mean it knew what was going to happen.

And if management used those monthly forecasts to make hiring, spending, or cash decisions, the difference matters.

Let’s look at how to calculate forecast accuracy properly, how to measure bias, and why the prettiest percentage in the management report might be the least useful.

The Forecast Accuracy Formula Finance Should Actually Understand

For positive-valued revenue and expense forecasts, one useful measure is Weighted Absolute Percentage Error, or WAPE.

The formula is:

WAPE = Sum of Absolute Forecast Errors ÷ Sum of Absolute Actual Values × 100

In ordinary English, WAPE measures how far your forecasts missed without allowing overestimates and underestimates to cancel.

That’s the important part.

If you overforecast January revenue by $20,000 and underforecast February revenue by $20,000, the net error is zero.

But you still missed by $20,000 in January and another $20,000 in February.

WAPE counts both errors.

Some companies convert WAPE into a forecast accuracy percentage:

Forecast Accuracy = 100% − WAPE

It’s easy to explain in a dashboard, although it becomes awkward when WAPE exceeds 100% and the resulting accuracy score turns negative.

I prefer reporting WAPE directly.

A lower percentage means smaller aggregate absolute errors relative to the actual financial base.

And unlike a vague claim that the forecast was 95% accurate, everyone can see exactly what was measured.

Let’s Calculate It With Actual Numbers

Consider a hypothetical business forecasting monthly revenue.

All amounts below are in thousands of dollars.

Month Actual revenue Forecast revenue
January $100 $120
February $120 $100
March $80 $100
April $100 $80
Total $400 $400

If finance compares the four-month totals, the variance is zero.

Wonderful.

But look at the individual months.

January missed by $20,000.

February missed by $20,000.

March missed by $20,000.

April missed by $20,000.

The forecast wasn’t perfect. It was consistently $20,000 away from reality, just with alternating signs.

To measure that, start with the forecast error:

Forecast Error = Actual − Forecast

Under this convention, a positive number means the forecast was too low. A negative number means the forecast was too high.

Now calculate the absolute error:

Absolute Error = ABS(Actual − Forecast)

The absolute value removes the sign.

Here’s what happens:

Month Actual Forecast Signed error Absolute error
January $100 $120 −$20 $20
February $120 $100 +$20 $20
March $80 $100 −$20 $20
April $100 $80 +$20 $20
Total $400 $400 $0 $80

Amounts are in thousands of dollars.

The signed errors total zero.

The absolute errors total $80,000.

Same forecast. Very different story.

Now calculate WAPE:

WAPE = $80,000 ÷ $400,000 × 100 = 20%

So the forecast has a 20% weighted absolute percentage error.

That’s a much more revealing number than the zero variance finance originally celebrated.

How to Calculate WAPE in Excel

Assume your spreadsheet has:

  • Column B: Actual revenue
  • Column C: Forecast revenue
  • Column D: Signed error
  • Column E: Absolute error

With monthly data in rows 2 through 5, enter:

Signed error, cell D2:

=B2-C2

Absolute error, cell E2:

=ABS(D2)

Copy both formulas down through row 5.

Then calculate WAPE:

=SUM(E2:E5)/SUM(B2:B5)

Format the result as a percentage.

You should get 20%.

This shortcut assumes actual revenue values are positive. For a dataset with negative actuals, the denominator needs additional consideration, and WAPE may not be the right metric at all.

For measures such as net income or net cash flow, where values can cross zero, applying a percentage-error formula mechanically can produce results that are difficult to interpret.

Excel will calculate almost anything you ask it to.

That doesn’t mean the answer belongs in a board presentation.

WAPE vs. MAPE: Why Two Accuracy Formulas Give Different Answers

Another common measure is Mean Absolute Percentage Error, or MAPE.

The formula is:

MAPE = Average of [ABS(Actual − Forecast) ÷ ABS(Actual)] × 100

MAPE calculates the percentage error for each period and then averages those percentages.

Using the same revenue example:

Month Absolute error Actual revenue Percentage error
January $20 $100 20.00%
February $20 $120 16.67%
March $20 $80 25.00%
April $20 $100 20.00%

Dollar amounts are in thousands.

Add those percentages and divide by four:

MAPE = (20% + 16.67% + 25% + 20%) ÷ 4 = 20.42%

WAPE was 20%.

MAPE is 20.42%.

Not a dramatic difference here, but the distinction becomes important when you’re measuring forecasts across business units of very different sizes.

Imagine one division generates $20 million in annual revenue and another generates $200,000.

Both miss their forecasts by $100,000.

For the larger division, that’s a 0.5% error.

For the smaller division, it’s a 50% error.

MAPE gives those two percentage errors equal weight.

WAPE evaluates the combined dollar error against the combined actual revenue base.

Which is better?

That depends on what management is trying to understand.

If the CFO wants to know the aggregate financial magnitude of forecast misses, WAPE can be useful.

If the goal is to evaluate percentage accuracy across individual observations, MAPE may provide useful information.

But MAPE has a problem when actual values approach zero. The percentage error can become enormous, and it’s undefined when actual equals zero.

A $5,000 error against $10,000 of revenue is 50%.

The same $5,000 error against $100 of revenue is 5,000%.

The arithmetic is correct.

Whether the resulting average tells management anything useful is another question.

For a more technical discussion, FP&A Trends examines these limitations in Forecasting: Should the Mean APE Rule the Accuracy Planet?.

The Excel Formula for MAPE

Using actual revenue in column B and signed error in column D:

=IF(B2=0,NA(),ABS(D2/B2))

Copy the formula through row 5.

Then:

=AVERAGE(F2:F5)

The result is approximately 20.42%, assuming the individual percentage errors are in column F.

The formula flags a zero actual rather than pretending the percentage can be calculated normally.

If zero actuals occur frequently, investigate whether MAPE is appropriate before excluding those observations.

Now Let’s Talk About Forecast Bias

Here’s where forecast accuracy gets interesting.

WAPE tells you how large the errors were.

Bias tells you whether the forecast consistently leans in one direction.

Using our convention of Actual minus Forecast:

Aggregate Forecast Bias = Sum(Actual − Forecast) ÷ Sum(Actual) × 100

In our example:

Bias = $0 ÷ $400,000 × 100 = 0%

Zero bias.

Twenty percent WAPE.

That’s possible because the overforecasts and underforecasts canceled.

The company had no net directional error over those four months, but it still missed every monthly forecast.

This is why I wouldn’t put a single forecast accuracy percentage in front of management and call the job finished.

Calculate Forecast Bias in Excel

If signed errors are in cells D2 through D5 and positive actual revenue is in B2 through B5:

=SUM(D2:D5)/SUM(B2:B5)

Result: 0%

Under this convention:

  • Positive bias means aggregate underforecasting.
  • Negative bias means aggregate overforecasting.
  • Zero means the signed errors offset.

The denominator must be meaningful and nonzero. For financial measures containing positive and negative actuals, this percentage can be misleading or undefined.

Also, don’t confuse zero aggregate bias with proof that the forecasting process is unbiased.

A company could consistently overforecast one division and underforecast another.

At the consolidated level, everything cancels.

The CFO sees zero bias.

Meanwhile, two division leaders are making decisions using unreliable numbers.

The forecast isn’t improving. The consolidation is hiding the problem.

FP&A Trends explores this issue further in Tackling Forecast Bias: Signals and Noise.

Put Accuracy and Bias on the Same Report

Here’s what our example actually tells us.

Metric Result
Actual revenue $400,000
Forecast revenue $400,000
Net forecast error $0
Total absolute error $80,000
WAPE 20.00%
MAPE 20.42%
Aggregate bias 0.00%

The company forecast the total revenue correctly.

It did not forecast the monthly revenue pattern correctly.

And depending on the business, that distinction could be expensive.

If management used those monthly forecasts to determine inventory purchases, staffing requirements, or short-term borrowing, the errors could have affected real decisions.

A quarterly total doesn’t pay February’s bills.

This is also why the right level of measurement matters.

A company-wide forecast may look accurate while individual business units are materially wrong.

An annual forecast may look accurate while quarterly timing is unreliable.

The scorecard needs to reflect the decisions the forecast supports.

What Happens When the Forecast Improves but Bias Gets Worse?

Suppose FP&A introduces a more detailed forecasting method.

Instead of applying a broad revenue growth assumption, the team builds customer-level expectations and reviews them with Sales.

For illustration, assume the following alternative forecasts for the same four actual revenue observations.

Month Actual Alternative forecast Absolute error
January $100 $105 $5
February $120 $125 $5
March $80 $85 $5
April $100 $105 $5
Total $400 $420 $20

Amounts are in thousands. These are hypothetical values used to illustrate the calculations, not evidence of a real forecasting improvement.

Calculate WAPE:

$20,000 ÷ $400,000 = 5%

Calculate bias:

($400,000 − $420,000) ÷ $400,000 = −5%

The alternative forecast has smaller absolute errors.

But it consistently overforecasts revenue.

So the first forecast had 20% WAPE and zero aggregate bias.

The alternative has 5% WAPE and negative 5% bias.

Would I prefer the alternative?

Potentially. Its monthly predictions are much closer to actual revenue.

But I’d want to understand why every month was overforecast before using that method to approve additional spending or hiring.

And I certainly wouldn’t declare the new method superior based on four illustrative observations.

To demonstrate actual improvement, you need comparable forecasts made before the results were known, evaluated over enough out-of-sample periods to judge performance.

Building a model around known actuals and then congratulating yourself on its accuracy is a backtest with a very generous grading policy.

Your Forecast Horizon Matters More Than You Think

An annual budget prepared in November shouldn’t be expected to have the same accuracy as a forecast updated three weeks before the reporting period.

The business has learned a lot in the meantime.

Customers have changed their plans.

Hiring has slipped.

Pricing assumptions have moved.

Some projects have been delayed, while others have appeared out of nowhere with the apparent expectation that Finance should have known about them six months ago.

That’s why I would measure forecast accuracy separately by horizon.

Forecast version What it measures
Original annual budget Reliability of the original operating plan
Three-month-ahead forecast Medium-term forecasting performance
One-month-ahead forecast Near-term forecasting performance
Latest pre-period forecast Quality of the most recent estimate

Compare the same actual periods where possible.

Preserve the dates when forecasts were issued.

And use consistent definitions.

If the one-month-ahead forecast is consistently no better than the annual budget, that’s worth investigating.

The company has gained months of information, but the forecast hasn’t become more reliable.

Either the new information isn’t reaching Finance, or the model isn’t responding to it.

Both are management problems.

The Number Matters. The Reason Matters More.

Suppose the company underforecasts revenue by $500,000.

Management wants to know why.

Was it one unexpected customer contract?

Higher prices?

More transactions?

A favorable product mix?

Revenue recognized earlier than expected?

Those are different explanations with different implications.

If the miss came from a major contract that Sales already expected to close, Finance may have a communication problem.

If revenue consistently exceeds forecasts across hundreds of customers, the demand assumptions may be too conservative.

If the difference came from revenue timing, the underlying business may be performing exactly as expected while the monthly forecast remains unreliable.

The WAPE score won’t tell you which explanation is correct.

That’s where financial analysis begins.

A useful forecast review should connect material errors to business drivers and determine whether the assumptions need to change.

Otherwise, the team is measuring the forecast without improving it.

What Is a Good Forecast Accuracy Percentage?

I wish there were one universal answer.

It would make management reporting considerably easier.

Unfortunately, a company with predictable recurring revenue doesn’t face the same forecasting uncertainty as a project-based business dependent on a handful of large contracts.

Payroll forecasts behave differently from new customer revenue forecasts.

Near-term forecasts should generally benefit from information unavailable when longer-term forecasts were prepared.

And a small percentage error can still be serious if the company is operating close to a cash constraint.

So I wouldn’t start by asking whether 90%, 95%, or 98% accuracy is good.

I’d start with the business.

What are we forecasting?

How far ahead?

How volatile is the underlying driver?

What happens if we’re wrong?

Then establish a historical baseline using comparable forecasts.

Measure WAPE and bias consistently.

Look for recurring errors.

And evaluate whether forecast quality is improving where it matters.

A single accuracy target for every financial statement line is convenient.

It is also a good way to measure unlike things as though they’re identical.

Stop Treating Forecast Accuracy Like a Trophy

The point of measuring forecast accuracy isn’t to give Finance a better score on the monthly dashboard.

It’s to improve decisions.

If revenue forecasts are consistently optimistic, management may approve spending based on revenue that never arrives.

If payroll forecasts ignore hiring delays, the expense outlook may be wrong even when the approved headcount plan hasn’t changed.

If cash forecasts miss collection timing, the company can appear adequately funded on paper while facing a very real liquidity problem.

Those are the consequences worth measuring.

A forecast with a beautiful accuracy percentage can still be unreliable for the decisions management needs to make.

And a forecast with a larger error can still be useful when the uncertainty is understood and management has prepared for different outcomes.

So calculate WAPE.

Measure bias.

Investigate the drivers.

But don’t stop at the percentage.

The real test is whether management makes better decisions because Finance understands what the numbers are telling them.

Because explaining last month’s results perfectly is useful.

Knowing what might happen next month is why the company asked for a forecast in the first place.

If your forecasts keep looking reasonable on paper and surprising management in practice, The Schlott Company’s Budgeting & Forecasting Services can help improve the assumptions, financial models, and operating processes behind the numbers.

THE SCHLOTT COMPANY NEWSLETTER

Never miss a new post.

Get our latest finance articles in one weekly email. No need to remember to check the blog.

We will email a confirmation link first. Privacy policy