A margin of error defines how far a forecast may deviate from the corresponding actual value while remaining acceptable. For demand forecasting, a client-set tolerance such as ±5% turns vague requests for “better” performance into a measurable business constraint. section 12.1.2
Balancing accuracy and speed
A team targeting a ±5% error margin can compare a computationally intensive deep learning model with a faster gradient boosting model. If both remain within the threshold, reduced training and run time may better serve the product. section 12.1.2
Use the margin as a shared decision boundary:
- Define the acceptable deviation from client or business needs. section 12.1.2
- Compare accuracy with constraints such as run time and model complexity. section 12.1.2
- Ask the data science team to evaluate alternatives against the threshold rather than issuing vague performance demands. section 12.1.2
