> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tetherdata.io/llms.txt
> Use this file to discover all available pages before exploring further.
# Forecast Admin
> Configure forecast generation, models, versioning, uploads, and availability in Tether
# Forecast Admin
The Forecast Admin surface provides tools for configuring how demand forecasts are generated, managing forecast models, submitting versions, and importing bulk data. These settings significantly impact forecast quality.
## Accessing Forecast Admin
The same surface is available from two entry points:
| Entry point | Page title | Access |
| ---------------------------------------------------------------- | --------------------- | ------------------------------------------------- |
| **Demand** → **Admin** in the sidebar (`/demand-forecast/admin`) | Forecast Admin | Non-admin users see only the **Availability** tab |
| **Settings** → **Forecast Admin** | Demand Forecast Admin | Admin users only |
Both entry points show the same five tabs:
| Tab | Purpose |
| ---------------- | ---------------------------------------------------------------------------------------- |
| **Admin** | Forecast date range configuration, forecast generation, and historical sales calculation |
| **Models** | Create and manage forecast models |
| **Versioning** | Submit and download forecast versions |
| **Upload** | Bulk forecast and sales data imports |
| **Availability** | SKU-channel availability matrix |
The **Admin**, **Models**, **Versioning**, and **Upload** tabs are restricted to Admin users. **Availability** is visible to all users with demand access.
## Admin tab
### Forecast date range configuration
The **Forecast Date Range Configuration** card controls how far forward forecasts extend:
* **Forecast end date mode** — choose one year out, end of the current year, end of the next year, or a custom end date. When custom periods are configured, a quick-select lets you pick a period's end date directly.
* **New period initialization** — choose whether newly added periods start at zero or copy values forward.
Click **Save Configuration** to apply. A confirmation appears: "Configuration updated. Regenerate forecasts to apply changes." — configuration changes do not take effect in forecast values until the next generation run.
### Forecast generation status
The **Forecast Generation Status** card shows the current state of demand forecast generation and lets you trigger a run:
* A status indicator shows **Generating...**, **Last generation successful**, or **Last generation failed**.
* Click **Generate Forecasts** to start a run. While a run is in progress the button label switches to **Generating...** and a progress bar appears.
* After a run completes, the card shows the **Last Generated** timestamp and the **Total Forecasts** count.
* If the last run failed, a Generation Error alert shows the error message.
**Generate Forecasts** regenerates the *demand forecast* (model outputs for SKU-channel combinations). It is a separate system from the supply-planning solve, which is triggered with **Run now** on the Connected Planning page. Generating forecasts does not create a new supply plan — see the [Supply Planning overview](/supply-planning/overview) for how planning consumes forecasts.
### Calculation coverage summary
This card summarizes the calculated historical sales data that feeds the forecast algorithms: total records, active channels, active SKUs, coverage percentage, the covered date range, and the calculation status.
### Calculate historical sales
Use the **Calculate Historical Sales** card to aggregate sales orders into daily historical sales records for a date range:
Set the **Start Date** and **End Date**. The default range is roughly the last year, ending yesterday.
The calculation aggregates sales by date, channel, and SKU. It may take several minutes for large datasets.
A confirmation reports how many records were calculated, and the coverage summary updates.
## Models tab
The Models tab (**Model Management**) is where you create and manage forecast models. A **model** is a method for generating a possible forecast. Each model uses an **algorithm** that defines how forecast values are calculated. Your organization can define multiple models with different algorithms and compare their outputs.
For a deep dive into how each algorithm works, including mathematical formulas and parameter details, see the [Forecast Algorithms](/demand-forecasting/algorithms) page.
### Available algorithms
Tether provides 11 forecast algorithms you can assign to models:
| Algorithm | Description | Best for | Supports auto-tune |
| ------------------------------------------------ | ----------------------------------------------------------------------------------------- | ------------------------------------------------------ | :----------------: |
| **90-Day Rolling Average** | Simple arithmetic mean of the last 90 days of sales | Stable, predictable demand | No |
| **120-Day Rolling Average (Recursive)** | Sliding window average that incorporates prior forecast values as it projects forward | Volatile products, strategic planning | No |
| **Exponentially Weighted Moving Average (EWMA)** | Applies exponential decay weighting to emphasize recent sales | Trending products, fashion items | Yes |
| **Linear Trend** | Least-squares linear regression projected forward | Products with consistent growth or decline | No |
| **Simplified Seasonal** | Weighted combination of weekly (80%) and monthly (20%) seasonal patterns | Seasonal and holiday-driven products | Yes |
| **Simple Seasonal with Trend Adjustment** | Seasonal patterns blended with a year-over-year growth factor | Seasonal products with growth trends | Yes |
| **Seasonal Year-over-Year Growth** | Seasonal baseline multiplied by observed year-over-year growth rate | Seasonal products with strong annual growth | Yes |
| **Rolling Momentum** | Detects acceleration or deceleration by comparing short-term vs long-term moving averages | Viral, accelerating, or declining products | Yes |
| **Holt-Winters (Triple Exponential Smoothing)** | Captures level, trend, and seasonality with separate smoothing factors | Complex seasonal patterns with trends | Yes |
| **SARIMA** | Seasonal AutoRegressive Integrated Moving Average for complex time series | Complex time series with seasonal and trend components | Yes |
| **Static Baseline** | Directly uses uploaded baseline forecast values with no algorithmic processing | External forecasts, manual imports, hybrid workflows | No |
### Creating a model
Navigate to **Demand** → **Admin** and open the **Models** tab.
Click **Create Model** to open the model creation form.
Provide a name and description, then select an algorithm from the list above.
Save the model. The system calculates the model's output for all SKU-channel combinations.
### Quick actions
The **Quick Actions** card on the Models tab includes:
* **Generate Forecasts** — a second entry point for regenerating model outputs, equivalent to the button on the Admin tab.
* **Optimize All** — runs parameter optimization for every model whose algorithm supports auto-tune.
* **New Model** — opens the model creation form.
The **Current Selection** card shows which model currently drives the official sales forecast, and the **Model Overview** table lists every model with edit, delete, and optimize actions.
### Consensus Model
If your organization has no Consensus Model yet, the Models tab offers **Create Consensus Forecast Model**. The Consensus Model enables direct editing of forecast values for specific SKU-channel combinations, so you can override algorithmic forecasts with business knowledge. See [Editing Forecasts](/demand-forecasting/editing).
### Parameter optimization (auto-tune)
Seven of the 11 algorithms support **automatic parameter optimization**. When enabled, the system finds the best parameter values by minimizing forecast error on historical data.
**Optimization methods:**
* **Nelder-Mead Simplex** — used for algorithms with continuous parameters (e.g., smoothing factors, weights). Iteratively adjusts parameters to minimize error.
* **Grid Search** — used for SARIMA, which has discrete integer parameters (AR/MA orders). Tests all combinations within specified ranges.
**Error metrics available for optimization:**
| Metric | Description |
| ----------------------------------------- | ---------------------------------------- |
| **MSE** (Mean Squared Error) | Penalizes large errors heavily (default) |
| **MAE** (Mean Absolute Error) | More robust to outliers |
| **MAPE** (Mean Absolute Percentage Error) | Scale-independent comparison across SKUs |
**Per-SKU/channel optimization:** Optimized parameters are stored per SKU-channel combination, so each product-channel pair can have its own tuned settings within a model.
For the full list of parameters, defaults, and optimization ranges, see [Forecast Algorithms](/demand-forecasting/algorithms). To assign different models to different SKU-channel pairings, see [Model assignment](/demand-forecasting/model-assignment).
## Versioning tab
The Versioning tab manages forecast version submissions:
* **Versioning Status** — shows whether versioning is enabled and the current period. Admin users can enable versioning and open **Versioning Settings**.
* **Submit Version** — snapshots the current forecast as a named version for the current period.
* **Submission History** — lists past submissions with name, submitted timestamp and user, and version label. Each submission can be downloaded.
Sales representatives cannot submit forecast versions — contact an admin user if the **Submit Version** button is disabled.
## Upload tab
The Upload tab imports forecast or sales data in bulk from CSV.
### Data types
Choose what the upload targets:
| Data type | Effect |
| ---------------------- | -------------------------------------------------------------------------------------------------------------------------- |
| **Consensus Forecast** | Uploads directly to the Consensus Model (overwrites current values) |
| **Baseline Dataset** | Creates or updates a labeled baseline dataset that forecast models can reference (e.g., for the Static Baseline algorithm) |
| **Historical Sales** | Updates actual historical sales data (affects all models that use sales history) |
For baseline uploads, pick an existing label or enter a new one (e.g., "Q4 2024 Optimistic"). Existing baselines are listed on the tab and can be managed there.
### Templates
Click **Download Template** to get a starter CSV. The template format follows your organization's period configuration automatically:
**Standard date template** (default):
| Column | Required | Description |
| -------------- | -------- | ------------------------------------------- |
| `sku_code` | Yes | SKU identifier |
| `channel_name` | Yes | Sales channel name |
| `date` | Yes | Date in `MM/DD/YYYY` or `YYYY-MM-DD` format |
| `quantity` | Yes | Forecast quantity (integer) |
**Custom period template** (when custom time aggregations are enabled):
| Column | Required | Description |
| -------------- | -------- | ------------------------------------ |
| `sku_code` | Yes | SKU identifier |
| `channel_name` | Yes | Sales channel name |
| `period_label` | Yes | Period name (e.g., `P1`, `P10`) |
| `fiscal_year` | Yes | Fiscal year (e.g., `FY2025`, `2024`) |
| `quantity` | Yes | Forecast quantity (integer) |
Do not mix date and period columns in the same file. The system rejects files that contain both a `date` column and `period_label`/`fiscal_year` columns.
### Upload process
Click **Download Template** and fill it in with your values.
Select **Consensus Forecast**, **Baseline Dataset**, or **Historical Sales**.
Select your CSV file to validate it.
Review the preview: new records, updated records, and rows with errors.
Confirm to import the valid rows.
### Upload validation
The system validates all uploads:
* SKU codes must exist in the system
* Channel names must match existing channels
* Quantities must be numeric and non-negative
* Dates must be in a valid format; period labels and fiscal years must match configured periods
* Duplicate SKU + channel + date/period combinations are flagged
Invalid rows are rejected. You can fix errors and re-upload, or proceed with valid rows only.
## Availability tab
The Availability tab shows the SKU availability matrix — which SKUs are available on which channels. Unlike the other tabs, it is visible to all users with demand access, and it is the only tab non-admin users see at **Demand** → **Admin**.
## Data quality
### Data requirements by algorithm
Different algorithms have different minimum data requirements:
| Algorithm | Minimum history |
| -------------------------------------------------------------- | ------------------------------------------------ |
| Rolling Average, EWMA, Linear Trend, Rolling Momentum | 30–120 days (depending on window size) |
| Simplified Seasonal, Seasonal YoY Growth, Holt-Winters, SARIMA | 365+ days (at least 1 full seasonal cycle) |
| Static Baseline | No historical data needed (uses uploaded values) |
Seasonal algorithms (Simplified Seasonal, Holt-Winters, SARIMA, Seasonal YoY Growth) automatically fall back to a 90-day rolling average if fewer than 365 days of history are available.
## Best practices
* Start with simpler algorithms (e.g., 90-Day Rolling Average) and compare against more complex ones (e.g., Holt-Winters, SARIMA).
* Use the [Forecast Comparison](/demand-forecasting/comparison) view to evaluate accuracy before switching.
* Consider different models for different product segments — stable products may perform best with Rolling Average, while seasonal products benefit from Simplified Seasonal or Holt-Winters. See [Model assignment](/demand-forecasting/model-assignment) for per-pairing assignment.
* Enable parameter optimization (auto-tune) for algorithms that support it.
* After saving date range configuration, click **Generate Forecasts** so the change takes effect in forecast values.
* Monitor data freshness — stale sales history leads to outdated forecasts. Recalculate historical sales after large data corrections.
* User edits made through the [Consensus Model editing workflow](/demand-forecasting/editing) are preserved — forecast generation does not overwrite them.
* Submit a version before making major changes to models or configuration.
* Name versions descriptively (e.g., "Before SARIMA switch Q1 2026").
* Download past submissions to compare forecast evolution across planning cycles.
* Validate data before upload — ensure SKU codes and channel names match existing records.
* Always review the preview before applying.
* Pick the right data type: Consensus Forecast uploads overwrite current consensus values.
## Troubleshooting
### Forecasts not updating
**Possible causes:**
* No generation run has happened since the change — configuration changes wait for the next run.
* The modified model is not the currently selected model — only the selected model drives the official forecast.
* Data sync has not completed, so the latest sales history is not yet available.
**Solutions:**
1. Click **Generate Forecasts** on the Admin or Models tab.
2. Check the **Current Selection** card on the Models tab to confirm which model is active.
3. Check that data integrations have synced, then recalculate historical sales if needed.
### Generation fails
**Possible causes:**
* Insufficient historical sales data for the selected algorithm.
* A generation run is already in progress.
**Solutions:**
1. Check the Generation Error alert on the **Forecast Generation Status** card for the specific message.
2. Review the **Calculation Coverage Summary** — if coverage is low, run **Calculate Historical Sales** first.
3. Wait for the in-progress run to finish before starting another.
### Upload errors
**Common issues:**
* Wrong template type (date template vs. period template).
* SKU codes or channel names that don't match existing records.
* Invalid date format (use `MM/DD/YYYY` or `YYYY-MM-DD`).
* Mixing date and period columns in the same file.
**Solutions:**
1. Download the template again — it matches your organization's period configuration.
2. Verify all SKU codes and channel names exist in the system before uploading.
3. Check for non-numeric or negative quantity values.
## Next Steps
Learn how each algorithm works, its parameters, and when to use it
Assign forecast models per SKU-channel pairing
Compare model outputs and accuracy side by side
View and edit forecasts