A forecast that reads the source directly: eleven methods from naive to Holt-Winters, one accuracy test, prediction bands, scenarios and Ask Greg.
"What will revenue be next quarter?" usually means an export, a forecasting add-in and a chart someone rebuilds by hand every month.
In ATX AnalyticsTable a forecast is a dashboard widget that reads the source directly:
๐ฆ Straight from DuckDB. Point it at a table, a date column and a value column. The query returns one row per period, and the model refits on every refresh. No export, no copy of the data. The dashboard's slicers (region, product, channel, order size) change what the forecast learns from.
๐งช Eleven methods, one test. Naive, seasonal naive, drift, moving average, linear trend, linear trend + seasonality, exponential smoothing, Holt (optionally damped) and Holt-Winters (additive and multiplicative). Or pick Automatic: it fits every method the history supports, tests each on the most recent months it holds back, and keeps the one with the smallest scaled error (MASE).
On the Showcase project (10,000 orders, Jan 2025 โ Sep 2026) Automatic picked Holt with a damped trend: MAPE 5.9 %, MASE 0.41. The naive benchmark scores MAPE 11.5 %, MASE 0.78. The method and its accuracy are printed under the chart, so the reader sees how far to trust it.
๐ Bands and scenarios. Up to three prediction intervals (80 % and 95 % here), a Plan +5 % and a Downside โ8 % line, a forecast for each region, and a horizon that can come from a project parameter.
๐ค Or just ask. "Forecast monthly revenue for the next 6 months. Which method wins the backtest, and how accurate is it?" Ask Greg runs the chart's own forecasting model (not a second, guessed one) and answers: Holt, MAPE 5.9 %, October 2026 at 1,247,296 with its 80 % and 95 % ranges. Then it proposes the forecast chart. You place it on a sheet or a dashboard, and the numbers on the chart match the answer.
A forecast is a projection of the pattern in the history, not a promise. That's why every chart says which method it used and how it scored on months it had not seen.
This site needs JavaScript for the full experience. Every page above is reachable without it.