Skip to main content

​
What This Node Does

The Forecast node generates a forward-looking projection from a historical time series using ETS (exponential smoothing, Holt-Winters). Point a date field at the Rows bucket, the metric you want to project at the Values bucket, and Go Fig fits a model and emits the forecast horizon as additional rows on the same time grain. Optionally split the series into multiple parallel forecasts by adding columns to the Columns bucket (e.g., one forecast per region, product, or customer segment).

​
When to Use This Node

Use the Forecast node when you need to:
  • Project revenue or expenses for the next N months given a clean historical series
  • Forecast unit sales by SKU or region in parallel using a single node
  • Build budget-vs-forecast comparisons in a flow that already aggregates actuals
  • Stress-test a baseline by toggling trend and seasonality components
For complex driver-based modeling (price × volume × discount), pair Forecast with a Formula or Code node downstream. Forecast handles the time series fit, and the downstream nodes handle the business logic.

​
Step-by-Step Usage Guide

1

Configure the buckets

  • Rows: the date field to forecast over (typically month, week, or day).
  • Values: the numeric field to forecast (e.g., revenue, units_sold). Pick an aggregation function (Sum is most common).
  • Columns (optional): dimension(s) to fit one independent model per group, e.g., region to get separate forecasts per region in parallel.
2

Set the forecast horizon

Enter the number of periods to forecast in the Forecast Horizon field. Units match the date grain in the Rows bucket: 6 with monthly data = 6 months ahead. Range: 1 to 365.
3

Choose a growth pattern

The Growth Pattern setting controls how the trend component is modeled:
  • Linear (default): additive trend. Use for series with steady absolute growth (e.g., +$50K/month).
  • Exponential: multiplicative trend. Use for series growing as a percentage (e.g., +5%/month).
  • None: no trend component. Use for stable series with no directional drift.
4

Choose a seasonal pattern

The Seasonal Pattern setting controls how recurring cycles are modeled:
  • Consistent (default): additive seasonality. Use when seasonal swings are similar in absolute size each year (e.g., December always +$200K).
  • Proportional: multiplicative seasonality. Use when seasonal swings scale with the level of the series (e.g., December is always 30% above average).
  • None: no seasonal component. Use for short series (< 2 full cycles) or non-seasonal data.
5

Run and inspect

Run the node. Output extends the input table with periods additional rows. Forecast values appear in the same Values field. A generated is_forecast boolean field distinguishes actuals from projections.
Forecast node output: a table of report_date and amount rows where is_forecast is False for historical periods and True for the projected future periods.

​
Tips and Best Practices

Need at least 2 full cycles for seasonality: If the data covers only one year, set Seasonal Pattern to “None”. Otherwise the model will fit noise as if it were seasonal and the forecast will be unreliable.
Aggregate before forecasting: ETS works best on regular time grains. If your input is transactional, run Aggregation first to roll up to monthly (or weekly) totals before connecting to Forecast.
Multiple forecasts in one node: Adding a column to the Columns bucket fits one independent ETS model per group. Faster and cleaner than building a separate flow per segment.
Linear vs Exponential: If you can’t decide, run both and compare against the most recent 3 to 6 months of held-out actuals. Exponential is appropriate when growth rate is stable as a percentage and Linear when growth rate is stable in absolute terms.
Validate before sharing: Forecasts assume the future resembles the past. Always sanity-check against business context: a Forecast that projects flat revenue through a known launch quarter is mathematically correct but practically wrong.