Financial Forecasting Methods: How to Choose the Right One

I have never understood why companies spend enormous amounts of time building sophisticated forecasts and then use essentially the same forecasting method for everything in them.

Revenue?

Historical growth plus an adjustment.

Travel?

Historical growth plus an adjustment.

Software?

Historical growth plus an adjustment.

Headcount?

Last year plus the people we think we’re hiring.

Coffee?

At this point, probably last year plus 5%.

Then somebody puts it all into one beautifully formatted model and we call it a forecast.

Technically, yes.

But if you’re trying to decide which financial forecasting method your company should use, I think there’s a better place to start.

Don’t start with the method.

Start with the number you’re trying to forecast.

Because your revenue and your office supplies do not behave the same way.

Your forecast shouldn’t pretend they do.

What Are the Main Financial Forecasting Methods?

If you came here looking for the traditional answer, financial forecasting methods generally include straight-line forecasting, moving averages and time-series methods, regression, driver-based forecasting, qualitative or judgment-based forecasting, and scenario analysis. Finance teams may also use top-down and bottom-up approaches, while AI and machine-learning techniques are increasingly being incorporated into forecasting. IBM

There.

We did the Google part.

Now let’s talk about the part I actually care about.

When should you use each one?

Because I don’t think your CFO is particularly interested in whether you can define multiple linear regression.

Your CFO wants to know whether we’re going to hit the number.

And if we’re not, why.

Straight-Line Forecasting: Useful Until It Isn’t

Straight-line forecasting is exactly what it sounds like.

Something has been growing at roughly 5% per year, so you assume it continues growing at 5%.

Simple.

There is absolutely nothing wrong with simple.

Finance has a tendency to confuse complexity with intelligence, and Excel has been enabling this behavior for decades.

If something in your business is relatively stable and the past is reasonably predictive of the future, a straight-line forecast may be perfectly adequate.

Let’s say an administrative expense has been running around $22,000 a month and nothing material is changing.

You could spend three hours building a model for it.

Or you could forecast roughly $22,000.

I know which one I’m choosing.

Straight-line forecasting works best when the business driver is stable and you have no compelling reason to believe the relationship is about to change. IBM similarly describes it as most appropriate for stable businesses with reliable historical patterns. IBM

But now let’s change the situation.

Suppose revenue grew 12% last year.

Can we forecast another 12%?

Maybe.

What caused the 12%?

If you don’t know, I don’t want to roll it forward yet.

Historical growth isn’t a driver.

Something caused the growth.

That’s what I want to find.

Moving Averages: When the Last Few Months Tell You Something

Moving averages smooth out short-term noise by averaging results over a selected period.

You might use a three-month average.

Or six months.

Or another period that makes sense for the business.

They’re particularly useful when recent history is informative but individual periods bounce around enough that using only last month would be misleading. CFI and IBM both identify moving averages as a common forecasting technique for historical patterns and short-term projections. IBM

Imagine an expense that went:

$92,000.

$108,000.

$97,000.

$103,000.

You probably don’t need an emergency meeting.

You may just have a somewhat noisy $100,000 expense.

A moving average can help you see that.

But here’s the question I’d ask you:

Why is it moving?

If the answer is ordinary timing and noise, fine.

If the answer is that the business fundamentally changed three months ago, averaging old behavior with new behavior can make your forecast beautifully wrong.

That’s going to become a theme here.

The math matters.

The business matters more.

Driver-Based Forecasting: Usually Where I Want FP&A to Spend Its Time

This is the method I find most useful for the important parts of a forecast.

Driver-based forecasting connects financial outcomes to the operating activities that actually cause them.

Revenue might be:

Customers × average revenue per customer.

Or:

Sales reps × opportunities × conversion rate × average deal size.

Headcount expense might be:

Employees × compensation × benefits.

Shipping might be:

Orders × average shipping cost.

Now if the forecast changes, you can explain why.

CFI recently made driver-based forecasting part of its FP&A Excel Modeling specialization, emphasizing driver trees and relationships between operating inputs and financial outputs. Corporate Finance Institute Help Center

That makes sense to me.

Because this is where forecasting starts becoming useful for management.

Let’s say revenue is $3 million below forecast.

Okay.

Why?

Fewer opportunities?

Lower conversion?

Smaller deals?

Hiring delays?

Higher churn?

Pricing?

Those are very different business problems hiding behind the same $3 million variance.

A driver-based model gives you somewhere to look.

But Please Don’t Driver-Base the Entire General Ledger

This is where Finance can become enthusiastic.

We discover drivers.

Suddenly everything needs one.

Now somebody is building a seven-variable model for office supplies.

I would like everyone to go home.

Not every line deserves the same modeling effort.

If you’ve got a $40,000 expense in a $200 million company that has behaved predictably for four years, I am not particularly interested in discovering its deepest operational truth.

Forecast it reasonably.

Move on.

Spend your analytical time where being wrong changes a decision.

That’s the part of financial forecasting methodology that I think gets missed.

The sophistication of the method should be proportional to the importance and uncertainty of the number.

Important + uncertain?

Model it.

Unimportant + predictable?

Please leave it alone.

Finance already has enough work.

Regression: When There Really Is a Relationship

Regression forecasting looks at the statistical relationship between one or more variables and the thing you’re trying to predict.

Maybe sales correlate strongly with customer traffic.

Maybe energy expense moves with production volume.

Maybe support staffing tracks active customers.

Regression can help quantify those relationships rather than relying entirely on judgment.

CFI lists simple and multiple linear regression among its core forecasting methods, while Workday includes regression-based approaches among common financial forecasting models. Corporate Finance Institute

This can be powerful.

It can also produce a dangerous sentence:

“The model says…”

I immediately want to know what went into the model.

Correlation doesn’t relieve Finance of the responsibility to understand the business.

If your regression says something is predictive, ask whether there’s a business reason for the relationship.

And then ask whether that relationship is likely to continue.

Because history is very cooperative right up until the business changes.

Qualitative Forecasting: Sometimes You Have to Talk to Humans

This method doesn’t always get the respect it deserves because there isn’t necessarily a magnificent spreadsheet attached to it.

Sometimes the best information available comes from people.

Sales knows a large customer is hesitating.

Operations knows a supplier is struggling.

HR knows those 15 hires aren’t arriving next month.

The product team knows a launch is slipping.

Your historical data knows none of this yet.

Quantitative forecasting uses numerical data; qualitative approaches incorporate judgment, expert knowledge and other information that may not be captured in historical results. Most practical forecasts combine the two. AccountingTools

This is why I don’t believe FP&A should sit in Finance, collect numbers and emerge with a forecast.

Go talk to people.

Preferably before the forecast meeting.

You learn considerably more that way.

Scenario Analysis: When One Answer Would Be Dishonest

Sometimes the problem isn’t determining the most likely number.

It’s acknowledging that several materially different outcomes are plausible.

That’s when scenario analysis becomes useful.

Suppose a major contract is being negotiated.

If it closes, revenue and hiring change.

If it slips six months, something else happens.

If the customer walks away, now you have another problem.

One forecast can hide that uncertainty.

Scenarios expose it.

Sage describes scenario planning as particularly useful for strategic planning and uncertain environments, while IBM includes scenario-based methods among the ways companies can handle changing conditions. Sage

I like scenarios.

I just don’t need 47 of them.

If management makes the same decision under Scenario 14 and Scenario 15, I’m not convinced we needed both.

The purpose isn’t to prove that Finance can imagine many futures.

It’s to help management prepare for the futures that would cause us to do something differently.

Top-Down vs. Bottom-Up Forecasting

People sometimes talk about top-down and bottom-up forecasting as though one of them is correct.

I use both.

A top-down forecast starts with the larger picture and works downward.

Maybe the market is $500 million and we believe the company can capture 5%.

That gives us $25 million.

Bottom-up goes the other direction.

How many customers?

How many salespeople?

How many units?

At what price?

What conversion?

Build those pieces and see where you land.

If I were sitting with you reviewing a revenue forecast, I’d want to see both.

Not necessarily because I intend to average them.

I want to know whether they tell wildly different stories.

Suppose the top-down opportunity suggests $50 million.

Wonderful.

Then your bottom-up model says your current sales capacity can produce $31 million.

We have found something worth discussing.

Maybe the $50 million opportunity is real.

Maybe you need more capacity.

Maybe conversion needs to improve.

Maybe the target is nonsense.

I don’t know yet.

But now we’re asking a useful question.

What About AI Forecasting?

This is the section every forecasting article is apparently legally required to contain now.

AI and machine learning can analyze much larger datasets, detect patterns, identify anomalies and update forecasts more quickly than traditional manual processes. Current financial forecasting guidance increasingly includes AI alongside traditional techniques. IBM

I’m interested in that.

Very interested.

But AI doesn’t eliminate the fundamental question:

What are we trying to forecast, and what actually causes it to move?

If the underlying business changed yesterday and nobody told Finance, a magnificent algorithm trained on yesterday’s world still has a problem.

Let AI find patterns.

Let it test relationships.

Let it generate a baseline.

Let it tell me something I didn’t notice.

Then I’m still going to ask:

Why?

I realize this is becoming inconvenient for the machines.

So Which Financial Forecasting Method Should You Use?

Here’s how I’d actually make the decision.

Look at the line you’re forecasting and ask four questions.

How important is it?

If being wrong by 20% doesn’t change anything, don’t build NASA.

How predictable is it?

Stable historical behavior can justify a simpler method.

What actually drives it?

If you can identify meaningful operational drivers, build around them.

What decision does the forecast support?

This is the question I’d put at the top.

Because the best forecasting method isn’t necessarily the one with the greatest statistical sophistication.

It’s the one that gives management enough information to make the decision.

Let’s Do This With an Actual P&L

Pull up your forecast.

Revenue.

I’d probably want drivers.

Compensation.

Headcount, timing, salaries, benefits.

Rent.

If you’ve signed a lease, I have wonderful news. We may not need artificial intelligence.

Software expense.

Contracts, seats, renewals, planned additions.

Utilities.

Depending on the business, perhaps historical trend or an operational driver.

Travel.

Historical run rate adjusted for known business activity may be enough.

Interest expense.

Debt balance and rates.

Now look at what we’ve done.

We didn’t choose a financial forecasting method.

We chose methods based on how different parts of the business behave.

That’s usually how a good forecast actually works.

It’s a collection of methods held together by Finance’s understanding of the company.

Less elegant than saying we use driver-based forecasting.

More useful.

Your Forecast Doesn’t Need to Be Equally Smart Everywhere

This is probably the biggest thing I’d want you to take from this.

Finance teams have limited time.

Your model has limited usefulness.

Management has very limited patience.

Use sophistication selectively.

I want the most thought going into the assumptions where:

the dollars are large,

the uncertainty is meaningful,

and management can actually respond.

For everything else, reasonable may be good enough.

I realize “reasonable may be good enough” will not get me invited to many financial-modeling competitions.

I’ll survive.

One Last Test

Before you finish your next forecast, pick the five most important numbers in it.

Not the five numbers with the most complicated formulas.

The five numbers that would actually change what the company does.

Now ask yourself:

How are we forecasting this?

Why are we using that method?

What causes this number to move?

Has that relationship changed?

What would management do if we’re wrong?

If you can answer those, I probably don’t care whether the formula underneath it has three functions or 300.

If you can’t answer them, another formula isn’t where I’d start.

I’d go find someone who understands the business.

And talk to them.

Finance occasionally hates when the answer is a meeting.

Unfortunately, sometimes it is.