liveBook · Manning

Uncovered demand

As explained in Machine Learning System Design

Customer demand that is not fulfilled because insufficient inventory is available.

Uncovered demand occurs when available inventory cannot fulfill customer demand. In retail forecasting, it is more harmful than overstock because it combines missed profit with dissatisfaction and possible customer loss. section 9.4.1

Use uncovered demand as a central concern when reviewing forecast errors:

  • Prefer slight overprediction to underprediction, because negative residuals indicate demand was missed. section 9.4.1
  • Inspect residual bias and spread; a negative skew is a deployment red flag, while excessive positive bias also creates overstock risk. section 9.4.1
  • Agree forecasting granularity and resulting stock-management use cases with product managers, category managers, and stakeholders. section 9.4.1

Forecast residuals

A model with small positive bias and low residual variance approaches the preferred scenario. A model skewed toward negative residuals increases the risk of uncovered demand and should not be deployed in its current version. section 9.4.1

How Machine Learning System Design covers it

How deep it goes
Connects uncovered demand to residual distributions, bias, variance, and deployment red flags, but stops short of prescribing algorithms.
Examples
Uses week-store-item forecasting and residual-bias scenarios to illustrate missed demand and deployment risk.
What you can do after
Diagnose forecasting risk by checking residual spread, preferring slight positive bias, and flagging models skewed toward underprediction.
How it connects
Links uncovered demand to demand forecasting, inventory, out-of-stock problems, overstock, residual analysis, and customer dissatisfaction.

This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.