Skip to main content

Confidence Intervals

Overview​

Model 3 incorporates advanced probabilistic prediction capabilities, enabling inventory management decisions based on statistical confidence levels rather than single-point forecasts. This approach allows for more nuanced stock management based on item importance, lead times, and organizational risk tolerance.

Understanding Confidence Intervals​

Confidence intervals in Model 3 represent the range within which future sales are expected to fall with a specified probability:

  • A 50% confidence interval indicates the median forecast - there is an equal chance that actual sales will be above or below this prediction.
  • A 90% confidence interval provides a more conservative estimate, indicating that there is a 90% probability that actual sales will be at or below this predicted quantity.
  • Higher confidence intervals (e.g., 95%, 99%) provide increasingly conservative estimates for critical items where stockouts must be minimized.

Strategic Inventory Management​

This probabilistic approach enables inventory managers to:

  • Set higher confidence intervals (e.g., 90-99%) for critical items where stockouts would significantly impact operations
  • Use lower confidence intervals (e.g., 50-70%) for less critical items to optimize inventory costs
  • Balance inventory levels based on specific business requirements and constraints

Automated Confidence Interval Calculation​

When a sales file is uploaded, the system automatically calculates optimal confidence intervals based on two key parameters:

  1. Lead time: The time required to replenish inventory (measured in weeks)
  2. Target cumulative coverage: The desired probability of avoiding stockouts during the lead time

Calculation Process​

The algorithm:

  1. Analyzes historical forecast accuracy using test data
  2. Considers the specified lead time for each item
  3. Determines the confidence interval that would have achieved the target cumulative coverage over the test period
  4. Sets this as the default confidence interval for future forecasts

For example, if an item has a 6-week lead time and requires 100% coverage (zero stockout tolerance), the system will determine the confidence interval that would have provided complete coverage over all 6-week periods in the test data.

Item-Specific Configuration​

Lead time and target cumulative coverage values are sourced from item configurations, which can be:

  1. Uploaded via the Item List File Upload process
  2. Set to system defaults if no specific configuration exists (see Item Configurations)

Adjusting Confidence Intervals​

Users can modify confidence intervals through:

  1. Dashboard Controls: When Model 3 is selected, adjust the quantile value and click "Save Value" to update the database
  2. API Reset: To recalculate all confidence intervals based on the latest data, use the Optimize Confidence Intervals API