๐ฆ SKU Batch Forecaster
Upload SKU data and generate forecasts at scale using univariate models
1๏ธโฃ Upload Data
๐
Drag & drop your file here
CSV or Excel ยท SKU + Date + Value columns required ยท Location / Size optional
๐ฆ
0
Unique SKUs
๐ฆ
-
Date Range
๐
-
Frequency
๐ข
0
Total Rows
Showing first 10 rows
๐ Large Dataset Loaded
โ ๏ธ
Large Dataset Mode
This dataset is too large to load into the browser memory.
Use Submit Background Job to process forecasts on the server.
The job will run in the background and you can close this browser tab.
The job will run in the background and you can close this browser tab.
๐
-
Dataset
๐ฆ
0
Unique SKUs
๐
0
Total Rows
๐ฆ
-
Date Range
๐ Data Preview (Sample)
First 100 rows
โ๏ธ Forecast Configuration
๐ฆ SKUs to Forecast
๐ก Background jobs process on the server. You can close this tab and check results later.
โถ
๐ Dataset Intelligence (optional)
๐ Data Health Check
Click "Run Check" for instant data quality analysis
Detects: frequency, duplicates, outliers, stale SKUs, demand patterns
-
SKUs
-
Rows
-
Periods
-
Frequency
Issues & Recommendations
Demand Pattern Distribution
โ ๏ธ Duplicate SKU-Dates
๐ Potential Outliers
โฑ๏ธ Stale SKU Breakdown
๐ค AI Deep Analysis (optional)
Get contextual AI recommendations based on your data profile
Takes 10-30 seconds โข Uses health check results for context
Analyzing 0 SKUs...
This may take 10-30 seconds
0
Total SKUs
0
Quality Score
0
Forecastable
-
Date Range
๐ Overview
๐ Patterns
โฆ Quality
๐ Related SKUs
๐ค AI Insights
๐ฆ Volume Distribution
๐ Trend Summary
๐ Top 10 SKUs by Volume
๐ฏ Demand Classification
๐ฎ Forecastability
๐ Series Length
โ ๏ธ Problem SKUs
๐
Identify Related SKUs
Find SKUs with correlated demand patterns and detect lead/lag relationships that can improve forecasting accuracy.
Analyzing cross-correlations...
This may take 30-60 seconds for large datasets
0
Related Pairs
0
With Lead/Lag
0
SKU Clusters
0
Avg Correlation
2๏ธโฃ Configure Forecast โถ
๐ Column Mapping
Ctrl+click to select multiple. Creates per-combination forecasts (e.g. SKU | Store)
๐ฆ SKUs to Forecast
๐ฆ Horizon
๐ค Model
๐ง Chronos: Pre-trained neural model - great for complex patterns
โก Quick Mode: Uses Holt-Winters with minimal validation for maximum speed
๐ผ Ensemble: Runs all available models and blends their forecasts using accuracy-based weights
๐
Seasonal Component: Matches each SKU's attributes to a learned 52-week profile, estimates BSR from history, and projects forward โ no Python needed
๐ Centrifuge: Blends level baseline + seasonal + ML (+ optional events) with weights learned by scipy SLSQP to minimise holdout MAPE.
Seasonal needs a Seasonality Library + the SKU's profile match. Event needs a trained Event Uplift model (configured separately).
๐๏ธ Hierarchy
drag to reorder
Which level to generate forecasts for
Forecast at this level, aggregate up and disaggregate down
Optimal (MinT): Forecasts all levels simultaneously and reconciles using minimum-error weighting. Best accuracy but requires sufficient history per node.
โ๏ธ Advanced
๐ Data Handling
How to handle missing dates between first and last sale
SKUs with no sales for this many days before dataset end are considered discontinued
Dataset Range: -
Potentially Discontinued: - SKUs
Replace zero or near-zero sales periods caused by out-of-stock events with interpolated values before forecasting.
Cap or interpolate unexpected demand spikes before forecasting, using IQR-based fencing. Only upward outliers are corrected โ stock-outs are handled separately above.
Mark date ranges where demand data is unreliable (warehouse closures, outages, lockdowns). Values in these periods are interpolated over before forecasting.
Trains the tree & regression models on history with event indicators & log price as features. Affects backtest MAPE (unlike legacy event-uplift multipliers).
โถ
๐ External Regressors (Covariates)
NEW
?
What are Covariates?
External factors that influence demand:
โข Continuous: Price, temperature, GDP
โข Binary: Promotion flag, holiday
โข Categorical: Region, channel
Supported Models:
โ Prophet, LightGBM, ARIMA
โ ETS, Holt-Winters, Croston
โข Continuous: Price, temperature, GDP
โข Binary: Promotion flag, holiday
โข Categorical: Region, channel
Supported Models:
โ Prophet, LightGBM, ARIMA
โ ETS, Holt-Winters, Croston
๐
No additional columns detected for covariates.Upload data with price, promo, or other factor columns.
๐ฆ Future Values for Forecast Horizon
No future values configured - will use extrapolation
โ ๏ธ Selected model does not support covariates. Choose Prophet, LightGBM, or ARIMA, or use Auto-Select.
๐ข New Product Forecasting
0
๐ฆ New Product Mode
Configure attributes for a new product to find similar existing SKUs and generate a forecast based on their demand patterns.
Configure attributes for a new product to find similar existing SKUs and generate a forecast based on their demand patterns.
Load a dataset to see available attributes
Analog selection method
AI explains; Rule-based is deterministic
๐ Sell-Through Curve
Distributes a total life-volume estimate across periods using a decay shape โ ideal for fashion, seasonal, and short-lifecycle items with no history.
๐ง Attribute-Based Launch Model
Trains a regression model on all SKUs in your dataset (attributes โ first-period sales), then predicts launch volume for your new item. No time series needed โ just fill in the product attributes above.
๐
Seasonal Component Forecast
Match your new product's attributes to a pre-built 52-week seasonality profile, then multiply by a Base Sales Rate (BSR) to generate a weekly launch forecast. Requires an imported seasonality library.
๐ Size Profile Library
Learn per-attribute size shares from history and split a forecast across sizes (S/M/L/XL/...). Supports Bayesian blending of the library prior with a SKU's actual size sales.
๐ Day-Level Disaggregation
Learn a MonโSun day-of-week profile (global, per-SKU, or for specific special weeks) and split weekly forecasts into daily values. Includes per-ISO-week override support for events like Black Friday.
๐ Forecasting...
0%
0
Completed
0
Total
0
Errors
--:--
ETA
Initializing...
๐ Forecast Results
0
SKUs Forecasted
-
Avg MAPE (excl. intermittent)
-
Accuracy
-
Time
๐ Table
๐๏ธ Hierarchy
๐ Models
๐ Charts
๐ฌ Variance
๐ค AI Assistant
๐ Demand Planning
| SKU ID | Sparkline | Model | Classification | Covariates | MAPE | RMSE | Trend | Actions |
|---|
-
Model Distribution
MAPE by Model
Demand Classification
Accuracy Distribution
๐๏ธ Hierarchical View
No hierarchy detected
Showing top 20 forecasts by accuracy
โ ๏ธ
Anomaly Detection
Scan for unusual patterns, outliers, and potential issues
๐ก
Smart Recommendations
Get AI-powered inventory and ordering suggestions
๐ฏ
Priority SKU Selection
Identify which SKUs need detailed review
๐
Executive Summary
Get a comprehensive AI analysis of all results
๐ฌ
Loading workbench...