Revenue forecasting: Christian Wattig's 5-method playbook

The short version

Most FP&A teams pick one forecasting method and stick with it. Maybe it's incremental because it's fast, market-based because the data looks authoritative, or pipeline because the CRM is already there. Then the quarter closes, the number misses, and nobody can explain why. The single-method habit is the problem.

Christian Wattig, director of the Wharton School's FP&A program, uses five revenue forecasting techniques in practice, challenges three common myths, and pinpoints where relying on a single method costs you accuracy. You'll walk away knowing how to layer these methods together so your next forecast actually holds up.

What is revenue forecasting?

Revenue forecasting is the practice of projecting a company's future income over a defined period (monthly, quarterly, or annually) using historical data, sales pipeline activity, market conditions, and operational drivers. It informs budgeting, hiring plans, capital allocation, and the conversations between finance, sales, and leadership about what the next quarter or year actually looks like.

A strong revenue forecast does three things at once. It produces a number leadership can plan against, surfaces the assumptions behind that number so they can be challenged, and gives finance a starting point for the next month's variance analysis. The weakest forecasts skip the second step. Leadership gets a clean projection but no way to pressure-test it, which is how a single missed quarter turns into a credibility problem.

Christian Wattig's central thesis is that no single method does all three jobs well. Layering two or three methods is what separates a forecast that holds up from a forecast that has to be rebuilt mid-quarter.

The 5 main revenue forecasting methods

There are five techniques worth knowing across the spectrum from quick-and-dirty to data-heavy. The order roughly tracks how often they're used by finance teams today.

1. Incremental forecasting: Take last period's revenue and apply a growth rate (5% year-over-year, for example). Fast, easy to defend in a meeting, and used by most companies as a baseline. Christian's caution below is that incremental forecasting forces no actual analysis of your business, which is why it's the weakest method on its own.
2. Market-based forecasting: Project revenue from external signals: industry growth rate, share of a known TAM, competitor benchmarks. Useful when you have credible third-party data and the market structure is stable. Risky when the market is shifting or the data source can't be sense-checked.
3. Sales pipeline forecasting: Build the number from deals in the CRM, weighted by stage and historical conversion rates. The most common method for B2B sales-led companies. Goes deep on the next 1-2 quarters but can't see further out than the pipeline reaches.
4. Driver-based forecasting: Forecast the operational drivers behind revenue (sales reps with ramp curves, conversion rates, units sold, average order value) and let revenue fall out of the math. Captures planned changes the other methods miss. Christian's deep-dive in tip 3 later in this recap covers when this works and when it falls apart.
5. Statistical methods: ARIMA, moving average, exponential smoothing, regression, Monte Carlo. Time-series models that learn from your own data. Remove human bias but can't explain themselves to a CFO asking why the number moved.

Most strong forecasts layer two or three of these on purpose.

Why does single-method forecasting keep missing?

Your instinct is probably to pick the forecasting method that fits your available data and run with it. Christian's central thesis is that every method has a blind spot another method covers. You'll forecast more accurately if you combine two or three on purpose.

Here's where each method falls short on its own:

• Incremental forecasting is fast but forces no analysis of your business
• Market-based forecasting pulls in external reality but assumes a market that doesn't change and leans on third-party data you can't sense-check
• Sales pipeline forecasting gets granular but only covers a short window
• Driver-based forecasting captures planned changes but is vulnerable to human overconfidence
• Statistical methods remove bias but can't explain themselves to a CFO asking why the number moved

The right question isn't "which method is best" but "which two or three methods cover each other's blind spots."

Christian’s 5 lessons about revenue forecasting

1. The top-down vs. bottom-up gap is the budget's most valuable artifact

The gap between your top-down and bottom-up plans isn't a problem to fix. It's the most useful signal your budget produces.

In 11 annual budgeting cycles, Christian saw that gap 10 times. The one year the numbers matched, department heads had copied the top-down figure to avoid the back-and-forth. That wasn't alignment. It was avoidance.

Try withholding the top-down plan until bottom-up submissions land, but share budget guardrails so nobody works in the dark. In Christian's example, the marketing team was told their budget would land between $3M and $5M and asked what they could do with that range. The bottom-up plan came back more realistic, and the gap with the top-down forced a concrete conversation about which assumptions were actually defensible.

That gap is where you learn what operators believe is possible and where leadership's assumptions break. Eliminating it eliminates the analysis, which is why Christian's rule of thumb is to protect the gap rather than close it prematurely.

"Having a gap is actually a good thing because it tells you something about where a possible range may be, and it's a starting point for conversations around, okay, what is possible here? Where can we really take the company performance?"

2. A true rolling forecast projects the same horizon every month

Most teams that say they run a rolling forecast are actually running a monthly-updated annual forecast. Those are different things.

If it's October and your fiscal year ends December 31, a traditional forecast asks you to project 3 more months. A rolling forecast asks you to project 12 months forward, through September of the following year. The horizon stays fixed. The end date moves with you.

Christian's rolling forecast test is simple: if forecasting gets easier the closer you get to year-end, you're not running one. The same horizon every month builds the muscle and gives you a running start on next year's budget. It also kills the December scramble where you're suddenly asked to project an entire fiscal year from scratch.

For teams making the switch, the muscle builds with repetition. Every month you forecast the same horizon, it gets a little easier.

"Why would you, if it's October, just forecast three months? Why not build that muscle of always forecasting the same period? Every month you do that, it gets a little bit easier. If forecasting gets easier the closer you get to the end of the fiscal year, then you're not really building up muscle."

3. Driver-based forecasting only works when capped at 5 to 10 drivers

A driver-based plan can't have 50 line items. Fifty drivers get no leadership attention, dilute the analysis, and turn the model into a spreadsheet exercise nobody acts on.

Each driver should represent a tactic or strategy, not a P&L line item. "Marketing expense" isn't a driver. The specific brand awareness campaign behind it is.

Christian walked through a driver-based plan from a subscription company, starting with a 3.4M subscription base in 2025 and forecasting 2026. His drivers included unaided awareness campaigns, direct response budget, inflation impact, a new promo strategy, and "do nothing different" conversion decay.

For each driver, he built three scenarios: conservative, balanced, and aggressive. The balanced scenario was the base case. If you want to hit a higher target, you can move individual drivers to aggressive and see exactly how much additional revenue you're betting on each lever.

The output was a one-page view you could put in front of the board, showing exactly how much revenue risk each initiative carried, instead of a single number everyone argues over.

"What are the five to 10 initiatives, the five to 10 big rocks, levers that you're trying to pull to move the company forward? Because a driver based plan can't have 50 things. It just doesn't work. It doesn't get the leadership attention it deserves if it has 50 drivers."

4. In B2B pipeline forecasting, conversion rate method depends on deal volume

Pipeline forecasting only works if you've decided how to set conversion rates at each stage of the funnel, and the right approach depends on your deal volume.

• Standardized rates: Derived from historical CRM data work best for high-volume pipelines with many similar, low-to-medium value deals. The data set is large enough to segment by prospect type and produce probabilities you can defend.
• Ad hoc rates: Assigned by individual salespeople work best for small numbers of high-value deals. Reps spend more hands-on time with each prospect and have contextual knowledge that historical averages can't capture.

Christian's caution applies to both approaches: pipeline forecasting is only as good as pipeline hygiene. Filter unqualified leads early and refresh standardized rates as new data comes in. Ad hoc rates introduce subjectivity, and the less you know about a rep's judgment, the riskier their estimates become.

Always test either approach against actual conversions. Treat the variance between forecast and actual as a diagnostic, not just a miss.

"It definitely is an art and a science, but that's why they pay you the big bucks in finance, to help navigate that, to help figure out what those probabilities should be."

5. Statistical methods earn their place as the "do nothing different" baseline

Every other method on this list is vulnerable to human bias. Salespeople sandbag, department heads submit conservative plans to beat them, and leadership sets aggressive top-downs in response. You've probably seen this cycle. Statistical methods don't care. They follow the data.

Five statistical methods are worth adding to your toolkit:

• Moving average: Take the last 3 months of revenue, average them, and use that as next month's forecast
• Exponential smoothing: Weights recent data more heavily. Useful for stripping one-time outliers from history.
• ARIMA (AutoRegressive Integrated Moving Average): Learns from its own prediction errors over time. Strong when your business has clear trends and seasonality.
• Regression models: Show the relationship between revenue and one or more independent variables like ad spend or headcount
• Monte Carlo simulation: Runs thousands of randomized scenarios to produce a probability distribution rather than a point estimate

Statistical output answers one question precisely: if you change nothing about how you operate, where will revenue land? That becomes the floor. Driver-based forecasting layers the financial impact of planned changes on top. For B2B companies, a sales pipeline forecast adds a third short-term layer, because one method alone isn't enough.

"It minimizes human bias because as much as I like driver based forecasting, if you have people that are just super overconfident with everything they do, the forecast quality may suffer. The machine doesn't have those same biases."

How do you apply Christian's advice this quarter?

You can put this into practice in three different ways.

1. Run the rolling forecast test on your own process: Open your current forecast. If the horizon shrinks as you get closer to year-end, you don't have a rolling forecast. Pick a fixed 12-month horizon and commit to projecting that far forward every month, regardless of fiscal year position
2. Pick your top revenue stream and list its 5 to 10 drivers: Not P&L line items, but actual tactics and strategies, like a specific campaign, a specific headcount plan, or a specific pricing change. If you can't get under 10, you're not being ruthless enough. Hand the list to the department owner and ask them to pressure-test it
3. Run a statistical baseline on your historical revenue data: Christian notes that ChatGPT can build an ARIMA-based forecast. The output becomes your "do nothing different" baseline, the floor you layer driver-based work on top of

Final thoughts

"If you don't communicate this, you lose credibility. Because the one thing that senior leaders dislike is surprises. If you know that there's uncertainty in your forecast, then tell them that, and ideally give them a range. Tell them, look, we think revenue in the next quarter will be 20 million, but actually it's likely a range of 16 to 24 million…. That's much more useful than just giving them a number and crossing your fingers and hoping that it will be accurate."

The instinct is to hand leadership a clean number and hope it holds. Christian's point is the opposite: surfacing uncertainty as a range tells leadership how usable the forecast actually is, which is more useful than a single number and crossed fingers.

Combine two or three methods, and show your leadership the range.

See how Ramp fits in

Every forecasting technique Christian walked through depends on clean, timely, structured financial data. When your expense data, vendor spend, and departmental actuals live in fragmented systems and land on your desk weeks late, even the best layered method gives you outdated numbers.

With Ramp, you can pull consolidated spend, cards, bill pay, and expense data from one real-time source directly into a driver-based model or statistical baseline.

Common questions about revenue forecasting

What is revenue forecasting?

Revenue forecasting is the practice of projecting future income over a defined period (monthly, quarterly, or annually) using historical data, sales pipeline activity, market conditions, and operational drivers. The output informs budgeting, hiring, capital allocation, and the cross-functional conversations between finance, sales, and leadership about what the next quarter or year actually looks like. Strong forecasts layer two or three methods so each one's blind spot gets backstopped.

What is the most accurate revenue forecasting method?

There isn't one. Christian's central thesis is that every method has a blind spot another method covers, which is why layering is more accurate than picking a single approach. For most companies, the strongest combination is a statistical baseline (what happens if nothing changes), a driver-based plan (financial impact of planned changes), and a sales pipeline view (short-term deal-level reality). The variance between those three is the analysis.

How do you create a revenue forecast?

The process is roughly five steps regardless of which methods you choose. Gather 12 to 36 months of historical revenue data. Identify the patterns that matter (seasonality, growth trends, deal cycle length). Adjust for the external factors your model doesn't see (new market entries, competitor moves, macro shifts).

Apply your chosen method or combination to produce the projection. Then review monthly against actuals and revise. The last step is the one most teams skip, which is why a forecast that worked in January often doesn't hold by April.

What's the difference between a budget and a forecast?

A budget is the annual commitment leadership makes to the board. A forecast is the ongoing projection of where your business is actually heading. Once leadership approves the budget, it's fixed. The forecast updates as new information comes in. In Christian's experience, teams that conflate the two end up defending stale assumptions instead of surfacing real changes.

Should we run top-down and bottom-up plans in parallel?

Yes, and the gap between them is the point. Share budget guardrails so operators aren't building in a vacuum, but withhold the top-down number until bottom-up submissions land. The conversation about why the two views disagree is more useful than either view alone.

How often should we revisit our forecasting method?

At least quarterly. As your business evolves, whether it's becoming more stable or the market is getting more unpredictable, you may need a different technique. Christian balances that against the value of building a muscle through repetition, but the cadence to revisit should be regular.

What if we don't have enough historical data for statistical methods?

Use the market-based approach as a temporary substitute. Christian recommends never running market-based forecasting alone. But if you're early-stage, it gives you a defensible projection while you build the operating history you need for ARIMA or regression. Pair it with a driver-based view as the sanity check.