> ## Documentation Index
> Fetch the complete documentation index at: https://docs.spherecast.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Baseline forecast

> Run statistical forecast runs, let Spherecast auto-select the best method, and catch products whose sales have shifted

## What this page is for

A **forecast run** generates the statistical **Baseline** — the starting-point number for every product and channel, calculated from sales history. This page lists your past runs, lets you start a new one, and schedules runs to happen automatically.

The key idea: you don't pick a forecasting model by hand. Spherecast tests several methods and **auto-selects the best one per product and channel** based on an error metric you choose.

## What you see

The run list shows one row per run:

| Column           | Meaning                                                                               |
| ---------------- | ------------------------------------------------------------------------------------- |
| **Id**           | The run's reference number                                                            |
| **Status**       | Done, Active, Ready for review, Creating validations…, Saving forecasts…, or Inactive |
| **Triggered by** | Scheduled or manual                                                                   |
| **Scope**        | Which products, channels, or groups the run covered                                   |
| **Created at**   | When it started                                                                       |
| **Duration**     | How long it took                                                                      |
| **Created by**   | The person, or "System" for scheduled runs                                            |

Buttons: **New forecast run** and **Setup schedule**. Before any runs exist you'll see "No forecast runs yet. Create your first!"

## Step by step: create a run

1. Click **New forecast run**.
2. Choose the **scope** — channels, products, groups, or categories.
3. Set the **time horizon** in months.
4. Set the **confidence interval**.
5. Choose the **error metric** used to pick the best model.
6. Choose which **methods** to test.

### Error metrics

These measure how far a method's forecast is from actual sales. Lower is better.

| Metric    | Full name                                |
| --------- | ---------------------------------------- |
| **MAPE**  | Mean Absolute Percentage Error           |
| **MAE**   | Mean Absolute Error                      |
| **RMSE**  | Root Mean Square Error                   |
| **SMAPE** | Symmetric Mean Absolute Percentage Error |

### Methods

| Group                | Methods                                                                                                       |
| -------------------- | ------------------------------------------------------------------------------------------------------------- |
| **AI**               | Time-GPT (an AI model trained on time-series data)                                                            |
| **Machine Learning** | LGBM, XGBoost, Linear Regression, CatBoost                                                                    |
| **Statistical**      | Moving Average (last 90 days), Exponential Smoothing (ETS), Historic Average, ADIDA, IMAPA, Croston Optimized |

Spherecast runs the candidate methods against each product and channel, scores them by your chosen error metric, and keeps the best performer for each one automatically.

## Scheduling runs

Click **Setup schedule** to run forecasts **weekly** or **monthly**, with an optional **auto push-down** so agreed numbers flow into supply planning without a manual step. The page shows "Next run scheduled for …".

## Sales-deviation alerts

A **Sales-deviation alerts** popover shows a count of **Products to review** — items whose last calendar month of sales deviates from the prior 3-month baseline beyond a threshold. Click **Review & re-run forecasts** to start a new run scoped to just those products.

Thresholds are set per **ABC class**, each with a **Threshold %** and an **Enabled** toggle. These are the same "Sales deviation" items you'll see on the Cockpit.

## Example

A planner runs a monthly forecast across all channels using MAPE. For a fast-moving beverage SKU, Spherecast picks Exponential Smoothing; for a slow, lumpy spare part it picks Croston Optimized. A week later the sales-deviation popover flags 14 products whose latest month jumped well above their recent baseline. The planner clicks **Review & re-run forecasts** to refresh just those 14.

> **Tip:** Turn on **auto push-down** only once your team trusts the Baseline. Until then, keep push-down manual so you can review before it reaches supply. See **[Building consensus](/guide/demand-planning/building-consensus)** and **[Exceptions](/guide/home/exceptions)**.