Driver-Based Forecasting: How to Find the Right Business Drivers

I get suspicious when someone tells me they have 47 drivers in their forecast.

At that point, I don’t think you have drivers.

I think you have a very complicated spreadsheet.

There is a difference.

Driver-based forecasting is one of those Finance ideas that sounds almost impossible to disagree with. Instead of taking last year’s number, adding 7%, and hoping the business cooperates, you build the forecast around the things that actually cause financial results to change.

Wonderful.

I’m in.

The problem starts when we have to decide what those things actually are.

Because suddenly everything becomes a driver.

Headcount is a driver.

Price is a driver.

Pipeline is a driver.

Marketing spend is a driver.

Weather is apparently a driver.

Someone eventually mentions macroeconomic conditions.

And before you know it, we’ve recreated the entire global economy in Excel.

So if you’re trying to build a driver-based forecast, I wouldn’t start with the model.

I’d start with a much more annoying question:

What actually makes this number move?

That’s where the useful work begins.

What Is Driver-Based Forecasting?

Driver-based forecasting builds financial projections from the operational and financial factors that cause business results rather than simply extending historical financial results.

Instead of:

Revenue last year: $100 million
Growth assumption: 10%
Forecast: $110 million

you might build revenue from:

Customers × purchase frequency × average selling price

Or:

Sales reps × opportunities per rep × conversion rate × average deal size

Or, for a professional services company:

Billable employees × available hours × utilization × billing rate

The exact drivers depend on the business.

That’s the whole point.

Current FP&A guidance generally describes driver-based forecasting this way: connect financial outcomes to underlying operating inputs so Finance can understand not only where the forecast lands, but why. CFI now includes identifying drivers, building driver trees, validating relationships, and scenario analysis in its FP&A modeling curriculum. Corporate Finance Institute

None of that is particularly controversial.

The harder part is figuring out whether the thing you’ve put into your model is actually a driver.

A Driver Is Not Just Something That Moves

This sounds obvious.

It isn’t.

Suppose revenue increased every time marketing spending increased over the last two years.

Great.

Does marketing spending drive revenue?

Maybe.

Now I want to know what happened between those two numbers.

Did marketing create more leads?

Did those leads become opportunities?

Did conversion improve?

Did customers spend more?

Was marketing spending increasing because management already expected demand to increase?

Or did both numbers happen to go up at the same time?

You can find relationships everywhere if you stare at data long enough.

Finance people are particularly vulnerable to this because we have spreadsheets and confidence.

That’s a dangerous combination.

A useful driver should have an understandable relationship to the financial outcome you’re forecasting.

And ideally, the business should recognize it too.

If Sales looks at your revenue model and says, “That’s not how our business works,” I would not immediately explain to Sales why your regression says otherwise.

I’d get curious.

Start With the Business, Not the General Ledger

This is probably the biggest mistake I see in forecasting.

Finance opens the P&L and starts going down the accounts.

Revenue.

COGS.

Payroll.

Travel.

Software.

Marketing.

Rent.

Other operating expenses.

Then we ask:

How should I forecast each line?

I understand why.

The P&L is sitting right there.

Very organized.

Very tempting.

But the business doesn’t operate according to your chart of accounts.

Customers don’t wake up thinking about GL structure.

Neither does Operations.

If I were helping you build a driver-based forecast, I’d want to leave the P&L alone for a little while.

I’d go talk to the business.

How do we acquire customers?

What causes customers to buy more?

Why do they leave?

What limits our capacity?

When do we need another employee?

What causes overtime?

What determines production volume?

What changes gross margin?

What makes us need another location?

What causes cash to move differently from earnings?

Now we’re getting somewhere.

KPMG’s work on driver-based planning similarly emphasizes cross-functional development of the driver framework, connecting Finance with operating inputs from areas such as Sales, Marketing, Operations and HR rather than treating planning as an isolated Finance exercise. KPMG

This is one reason I think good FP&A people spend so much time wandering around other departments asking questions.

We’re professionally nosy.

It turns out to be useful.

Let’s Build a Revenue Forecast

Suppose your company generated $50 million of revenue this year.

Management expects $60 million next year.

Okay.

Where does the additional $10 million come from?

Please don’t say growth.

Growth is what happened.

I want to know what caused it.

Let’s say this is a B2B company.

Maybe revenue depends on:

number of sales reps,

opportunities per rep,

conversion rate,

average deal size,

renewals,

and churn.

Now we’ve moved from:

Revenue grows 20%.

to something management can actually argue about.

And I mean argue affectionately.

Maybe Sales says conversion will rise from 22% to 28%.

Fine.

Why?

New sales training?

Better leads?

Different product?

Pricing?

New territory?

Divine intervention?

If we can’t explain why a driver changes, we may have simply moved the unsupported assumption one level deeper into the spreadsheet.

That’s not driver-based forecasting.

That’s assumption relocation.

This Is Where Driver-Based Forecasting Gets Useful

Now imagine we’re three months into the year and revenue is below forecast.

With the old model, we know:

Revenue missed by $1.2 million.

Wonderful.

Everyone looks at Finance.

Finance looks at Sales.

Sales looks at the pipeline.

Someone asks whether the number includes the Johnson deal.

It does not.

We spend 14 minutes discussing Johnson.

You know the meeting.

With a good driver-based forecast, we can go further.

Maybe opportunities are exactly where we expected.

Average deal size is fine.

Sales capacity is fine.

But conversion fell from 24% to 19%.

Now we have something useful.

The financial result changed because an operating assumption changed.

And management can investigate the operating assumption.

That’s the real value.

FP&A Trends makes a similar point: driver-based forecasting allows variance analysis to move from simply identifying that results changed toward examining which underlying drivers missed and why. FPA Trends

The model isn’t just producing another number.

It’s helping us locate the problem.

But Not Every Number Deserves a Driver

This is where I tend to ruin the fun.

Once Finance discovers driver-based forecasting, there is a temptation to driver-base everything.

Please don’t.

Imagine a $100 million company.

Office supplies run about $65,000 a year.

They’re reasonably stable.

You could build:

Employees × office attendance × supplies consumed per employee × inflation × printer hostility.

Or you could forecast roughly $65,000.

I know.

Very disappointing.

But your analytical resources aren’t unlimited.

Neither is management’s attention.

I want sophisticated forecasting where sophistication changes something.

So before building a driver model for a line, I’d ask:

Is the number financially meaningful?

Is it uncertain?

Can management influence it?

Would understanding the driver change a decision?

If the answer is no to most of those, historical run rate may be perfectly respectable.

Not everything needs to become a science project.

Finance already has month-end close.

How Do You Find the Right Drivers?

If you’re building this from scratch, don’t start by brainstorming 75 possible variables.

Start with the financial outcome.

Let’s say gross margin is deteriorating.

Ask:

What would have to change in the business for gross margin to move?

Maybe:

product mix,

input costs,

labor efficiency,

discounting,

freight,

waste,

or pricing.

Now talk to the people who actually operate those processes.

This part matters.

Finance may believe labor hours drive production costs.

Operations may tell you overtime is the real issue.

Finance may think customer count drives support expense.

Customer Success may tell you certain customer types consume five times the support of others.

Finance may assume headcount follows revenue.

Management may actually hire in large steps because the business needs capacity before the revenue arrives.

Those conversations are where the model gets better.

Not because Finance has surrendered ownership of the forecast.

Because Finance finally understands what it’s forecasting.

I Like a Simple Driver Test

When someone proposes a driver, I mentally run it through a few questions.

Does it have a logical relationship to the result?

Can we explain why changing this variable should change the financial outcome?

Can we measure it?

A driver that nobody can reliably produce every month is going to become irritating very quickly.

Does it move enough to matter?

Something can technically affect revenue without explaining enough of the movement to deserve a place in the model.

Can someone in the business influence it?

Not every driver needs to be controllable, but controllable drivers are particularly useful because they connect the forecast to action.

Would management care if it changed?

This is my favorite test.

If the driver moves 20% and nobody does anything differently, I question why it’s occupying prime spreadsheet real estate.

Current driver-based forecasting guidance similarly distinguishes between identifying drivers and validating whether the relationships are actually useful enough to model. Corporate Finance Institute

Because a driver model isn’t improved by having more drivers.

It’s improved by having the right ones.

Build a Driver Tree Before You Build the Model

This is one of the few times I will encourage Finance to draw something before opening Excel.

Start with the financial outcome.

Let’s use revenue.

Revenue

↓

Existing customers + new customers

Existing customers might break into:

Beginning customers – churned customers

New customers might break into:

Leads × conversion rate

Revenue per customer might break into:

Units × price

Now you’ve got a simple picture of how the business produces revenue.

That’s your driver tree.

CFI explicitly teaches driver trees as a way to organize the relationship between operating inputs and financial outputs, while KPMG describes multilevel driver frameworks that move from top-level financial results into increasingly detailed underlying inputs. Corporate Finance Institute

Now stop.

Seriously.

Don’t immediately add 38 more branches because you can.

Ask which ones explain most of the movement.

Those are probably where I’d start.

Drivers Need Owners

This is another place driver-based forecasts quietly fall apart.

Finance builds the model.

Sales supplies pipeline.

HR supplies hiring.

Operations supplies production.

Marketing supplies leads.

Then three months later the forecast changes and nobody agrees on whose assumption it was.

I’ve seen this movie.

I don’t enjoy the ending.

Every important driver should have somebody who understands and owns the operating assumption.

Finance still owns the integrity of the forecast.

But Finance shouldn’t pretend it knows more about sales conversion than Sales or more about manufacturing capacity than Operations.

If Sales believes conversion will be 27%, I want Sales involved in that assumption.

And when it becomes 21%, I want to know what changed.

Not because we’re hunting for someone to blame.

Because the assumption told us something about the business that turned out not to be true.

That’s useful information.

Your Drivers Will Change

This is another reason I don’t like overly complicated driver models.

Businesses change.

The driver that explained revenue two years ago may not explain it now.

Maybe pricing changed.

Maybe the customer mix changed.

Maybe the company moved upmarket.

Maybe a new product launched.

Maybe capacity became the constraint.

Maybe the sales process changed.

Your model can continue working beautifully while the relationship underneath it quietly dies.

That’s unsettling.

I apologize.

I’m naturally a little suspicious.

It’s served me surprisingly well in Finance.

So periodically ask:

Does this driver still explain the business?

Not:

Does the formula still calculate?

Excel will calculate nonsense with tremendous professionalism.

That’s never been the problem.

What About External Drivers?

Some businesses genuinely depend on factors outside management’s control.

Interest rates.

Commodity prices.

Weather.

Exchange rates.

Industry demand.

Regulation.

Economic activity.

Those can belong in a driver-based forecast too.

Wolters Kluwer notes that non-financial drivers such as demand patterns, capacity utilization and machine hours can materially improve forecasting, and newer planning approaches increasingly use AI to help identify relationships across larger sets of operational information. Wolters Kluwer

But I’d treat external drivers differently from operating levers.

Management can’t control the weather.

It can control what it does when weather changes demand.

That’s the part I care about.

A good forecast doesn’t just tell me:

If X happens, EBITDA becomes Y.

It helps management think:

If X happens, what are we going to do?

Now we’re getting out of forecasting and into decision-making.

Which is where I want FP&A anyway.

AI Is Going to Find More Drivers Than We Ever Could

This is where things get interesting.

AI and machine learning can analyze relationships across far more data than an FP&A analyst reasonably can.

Customer behavior.

Pricing.

Sales activity.

Operational data.

External indicators.

Seasonality.

Maybe it finds something we never thought to test.

Good.

I want that.

But we’re going to need some restraint.

Because finding a statistical relationship and understanding a business relationship are not the same thing.

If AI tells me that some obscure operational metric is highly predictive of revenue, I’m interested.

Then I’m going to find someone who understands the business and ask:

Why would that be true?

If nobody knows, I don’t necessarily throw it away.

But I don’t hand it the keys either.

Apparently I have trust issues.

This has been documented.

The Best Driver-Based Forecasts Are Usually Simpler Than You Think

This is the part that can feel counterintuitive.

You start driver-based forecasting because you want a more sophisticated forecast.

Then, if you do it well, the model may actually become easier to understand.

Why?

Because instead of forecasting hundreds of lines independently, you’re identifying a smaller number of variables that explain most of the financial movement.

The forecast becomes:

If volume changes, here’s what happens.

If pricing changes, here’s what happens.

If hiring slips, here’s what happens.

If conversion drops, here’s what happens.

If churn rises, here’s what happens.

That is much more useful to management than:

Cell H147 changed.

I’m sure H147 had its reasons.

Go Look at Your Forecast

Pick one important number.

Revenue is an easy place to start.

Don’t open the formula yet.

Just ask yourself:

What would have to happen in the actual business for this number to be right?

Write those things down.

Then ask:

Which ones matter most?

Which can we measure?

Who understands them?

Who owns them?

How often do they change?

And what would management do if they moved?

Now open the model.

Are those things actually in it?

If they are, good.

If they aren’t, you may have a very nice financial model.

You just might not have a model of the business.

And that’s the distinction I care about.

Because the purpose of driver-based forecasting isn’t to make the spreadsheet more sophisticated.

It’s to make the business less mysterious.

Finance has enough mysteries already.

I host a podcast. Apparently we even have Excel sports now.

I’d rather not turn the forecast into another one.