Forecasting

Exponential Smoothing

Create practical demand forecasts with simple smoothing, Holt's trend method, or Holt-Winters seasonality, then review the forecast table and chart.

Browser-based calculation MAE and forecast chart CSV export support

Why use this tool?

Decision Supported

Use this tool when you have historical demand and need a near-term forecast for planning inventory, capacity, purchasing, or workload.

What It Tells You

It produces forecast values, an error metric, and a chart that shows how the forecast follows level, trend, or seasonal demand patterns.

Example Input

Load the sample to see a chronological demand series that can be tested with simple smoothing, Holt's trend method, or Holt-Winters seasonality.

Sample Output Interpretation

The result shows the next forecast and MAE. Use the chart to check whether the selected method follows the pattern without overreacting to noise.

1. Input Data / Data Import

Prepare your demand-history file

Upload an Excel or CSV file with a header row and one row for each historical period, ordered from oldest to newest.

Period or Date
Recommended for readability and checking sequence. Keep periods regular, such as one row per week or month.
Demand Value
A required numeric column containing demand, sales, units, quantity, or volume for each period.
History Length
At least 2 numeric observations are required. Use enough complete seasonal cycles when applying Holt-Winters seasonality.

Example row: 2026-01 | 850

What ATH does next: Upload the file and map the numeric demand column. ATH reads valid values in row order, places them into Quick Manual Entry for review, and applies the smoothing method and parameters you select.

Drag & drop your Excel (.xlsx, .xls) or CSV demand data here

Or click to browse files from your computer

Data is processed locally in the browser and is not uploaded to a server.

2. Quick Manual Entry

Period Demand Delete

3. Set Parameters

About Exponential Smoothing

Exponential smoothing is a forecasting approach that gives more influence to recent demand while still retaining information from older periods. It is useful when demand changes gradually and the next forecast should react to new data without swinging too sharply after every movement.

The method works by blending the latest actual demand with the previous forecast. The smoothing factor controls how quickly the forecast responds. A higher value follows recent changes more closely, while a lower value creates a steadier forecast.

Guidance

  1. Upload a CSV or Excel file with numeric demand values, or enter demand directly in Quick Manual Entry.
  2. Choose simple smoothing for level demand, Holt's method for a trend, or Holt-Winters when a repeating seasonal pattern is present.
  3. Set alpha to control how much weight recent demand receives.
  4. For Holt's method, set beta to control how quickly the trend estimate updates.
  5. For Holt-Winters, set gamma, the number of periods in one season, and additive or multiplicative seasonality.
  6. Review the next-period forecast, MAE, table, and chart before exporting the results.

Parameter Notes

Alpha

Use a higher alpha when demand has shifted recently and the forecast needs to react faster. Use a lower alpha when demand is noisy and you want a smoother result.

Beta

Beta is used only for Holt's method. It adjusts how strongly the model updates the trend from one period to the next.

Simple Method

Best for stable demand where the main goal is to smooth short-term variation rather than model growth or decline.

Holt's Method

Best when demand has a visible trend and a flat forecast would consistently lag behind the actual pattern.

Gamma

Gamma controls how quickly Holt-Winters updates the recurring seasonal pattern. Lower values keep seasonal estimates steadier.

Holt-Winters

Use additive seasonality when seasonal changes are similar in size, and multiplicative seasonality when they grow or shrink with demand.

Methodology, Assumptions, and Limitations

Formula Logic

Notation
yt = observed demand, lt = level, bt = trend, st = seasonal factor, m = season length

Simple smoothing
Ft+1 = αyt + (1 - α)Ft

Holt level
lt = αyt + (1 - α)(lt-1 + bt-1)

Holt trend
bt = β(lt - lt-1) + (1 - β)bt-1

Additive Holt-Winters
lt = α(yt - st-m) + (1 - α)(lt-1 + bt-1)

Multiplicative Holt-Winters
lt = α(yt / st-m) + (1 - α)(lt-1 + bt-1)

Seasonal update
Additive: st = γ(yt - lt) + (1 - γ)st-m; multiplicative uses yt / lt

Mean absolute error
MAE = (1 / (n - 1)) ∑t=2n |yt - Ft|

The implementation produces a one-period-ahead forecast. Holt-Winters initialises level and seasonality from the first complete season and trend from the first two complete seasons.

Required Inputs

The tool needs a clean numeric demand series in chronological order. Holt-Winters requires enough history to represent at least two full seasonal cycles.

Assumptions

The method assumes recent history is a useful guide to near-term demand and that major structural changes are reviewed by the user before relying on the forecast.

Limitations

Exponential smoothing does not explain why demand changes. Promotional events, stockouts, one-off orders, and market shocks should be reviewed before interpreting the output.

FAQ

What does MAE mean?

MAE stands for mean absolute error. It shows the average size of the forecast misses in the same unit as the demand data.

Can this handle seasonal demand?

Yes. Select Holt-Winters and enter the number of periods in a complete season. The data must contain at least two complete seasons.

Should I always use the lowest MAE?

A lower MAE is helpful, but the forecast should also make business sense. Check whether the chosen settings react appropriately to recent changes.