Aitium

Forecast Configuration Guide

Forecast Configuration Guide

Aitium uses Amazon Forecast — a fully managed machine learning service from AWS — to generate sales and inventory predictions. This guide explains how to configure, launch, and interpret forecasts.

Overview

Amazon Forecast uses historical time-series data to train ML models and produce predictions. Aitium automates the entire pipeline: data preparation, model training, forecast generation, and result visualization — so you can focus on acting on the insights rather than managing infrastructure.

Forecast Types

  • Sales Forecasting — Predicts future sales volume based on historical sales and traffic data.
  • Inventory Forecasting — Forecasts inventory needs and optimal stock levels based on historical inventory and sales patterns.

Launching a Forecast

  1. Navigate to the Forecasts page from the sidebar.
  2. Click Launch New Forecast Job.
  3. Select the forecast type (Sales or Inventory).
  4. Choose the marketplace and date range for the training data.
  5. Configure forecast parameters (forecast horizon, quantiles, etc.).
  6. Click Submit to start the forecast job.

How It Works

Behind the scenes, Aitium orchestrates a multi-step workflow:

  1. An AWS Lambda function creates a ForecastJob record in Amazon DynamoDB to track status.
  2. For inventory forecasts, fresh inventory data is collected first to ensure accuracy.
  3. A dedicated launcher function submits the job to Amazon Forecast with your dataset.
  4. AWS Step Functions state machines manage the pipeline: dataset import → predictor training → forecast generation → result export.
  5. Results are stored in Amazon S3 and displayed in the Aitium dashboard.

Monitoring Forecast Jobs

The Forecasts page shows all your forecast jobs with their current status:

StatusDescription
PENDINGJob is queued and waiting to start.
PROCESSINGAmazon Forecast is training the model or generating predictions.
COMPLETEResults are ready to view.
FAILEDAn error occurred. Open the job metrics modal for details.

Click on a completed forecast to visualize predictions with interactive charts. The forecast chart legend helps you interpret confidence intervals and predicted values.

Automated Forecasting

Aitium supports automated forecast scheduling via AWS Step Functions. When enabled, forecasts are regenerated automatically as new data arrives, keeping your predictions current without manual intervention.

Data Requirements

  • At least 30 days of historical data is recommended for meaningful forecasts.
  • More historical data generally produces better predictions — Aitium ingests up to 720 days.
  • Ensure your SP-API authorization is active and data ingestion is running before launching a forecast.

Subscription Requirements

Forecast functionality requires an active subscription. The system validates your subscription status (via Stripe or AWS Marketplace) before allowing forecast creation. Organization members inherit the admin’s subscription tier automatically.

Troubleshooting

  • Forecast stuck in PENDING — Verify your subscription is active and that historical data ingestion has completed.
  • Forecast FAILED — Open the job metrics modal for error details. Common causes include insufficient historical data or Amazon Forecast service quotas.
  • No inventory data for forecast — Ensure inventory ingestion has run successfully. Check the Inventory page for current data before launching an inventory forecast.
  • Forecast takes a long time — Amazon Forecast model training can take 30 minutes to several hours depending on dataset size. This is normal.