Alpha is the whole argument. Set it near 0.8 and the forecast snaps to the latest month, which makes the model responsive and jumpy. Set it near 0.2 and the forecast holds its line, which makes it stable but slow to admit something changed. Fit alpha against your own history by minimizing error on periods the model did not learn from, rather than picking a value by feel.
The versions that matter for revenue
Simple exponential smoothing assumes the series hovers around a fixed level. It suits a mature support-ticket volume, not a growing SaaS book.
Holt's linear method adds a second smoothing term for trend, so the projection can climb with the business. This is the minimum useful version for bookings at a company that is still growing.
Holt-Winters adds a seasonal term on top of that. It is the version worth running for B2B SaaS, because bookings never arrive evenly across the year.
Fitting the seasonal term to a SaaS calendar
ORM's view is that the SaaS calendar has a consistent shape. Q2 and Q4 close stronger than Q1 and Q3, and the third month of a quarter closes more than the first two. Holt-Winters learns that shape as a set of seasonal indices and reapplies it to matching future periods, so a strong March gets measured against prior Marches instead of against a naturally slow February.
That correction is the practical reason to prefer Holt-Winters over a raw trend line. Without it, every quarter-end looks like acceleration and every quarter-start looks like collapse. The same logic underpins broader time series forecasting, which separates the calendar effect from the underlying growth rate.
Where exponential smoothing stops working
The model reads nothing except the series itself. It has no view of pipeline created, win rate, or average deal size, so a market shift reaches the forecast only after it has already damaged several months of bookings. ORM points to exactly this as the most common reason a forecast misses: the model runs on assumptions that stopped being true. A competitor enters and compresses deal size, or buyers slow down and cycles stretch, and the smoothed line keeps projecting a curve the market has already left.
Run it as a baseline against the rep roll-up rather than as the forecast itself. When the smoothed projection sits well above the roll-up, pipeline has thinned or conversion has dropped. When it sits below, reps are calling a quarter their own history does not support. Either gap is worth more than the point estimate, and tracking which side you land on builds a record of forecast accuracy you can act on.
Frequently Asked Questions
What does alpha do in exponential smoothing?
Alpha sets how fast the weights decay. An alpha near 0.8 makes the forecast track the most recent month closely, which catches real changes early but also chases noise. An alpha near 0.2 holds the line and treats a jump as random variation. Fit alpha against your own history by minimizing error on periods the model did not learn from rather than picking a value by feel.
What is the difference between exponential smoothing and a moving average?
A moving average gives every period in its window the same weight and ignores everything outside the window. Exponential smoothing gives every past period some weight, declining geometrically. That makes smoothing more responsive to a recent shift without the cliff effect a moving average creates when an old month drops out of the window.
How much history does Holt-Winters need?
At least two full years of monthly data for annual seasonality, and three years is better. With one year the model cannot tell a repeating seasonal effect from one unusual quarter, so the seasonal indices become noise wearing the shape of a pattern.
Can exponential smoothing forecast individual deals?
No. It operates on an aggregate series such as monthly bookings or ARR. Individual deals need a probability or scoring approach that reads deal-level attributes. Use smoothing for the top-line shape and a deal-level model for what closes.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like exponential smoothing forecasting into prescriptive action for your team.
Schedule a Demo