29
USPS
HandWritten Address Interpretation System
Impact
$100M labor costs saved in first deployed year (1997)
Over $1B cumulative savings since adoption
Status
Over 83% of all handwritten USPS
mail sorted automatically (55M
pieces/day)
Above 98% accuracy
Adopted now in other countries
New direction: writer identification
Automatically adds Postnet Bar Code to >83% of all handwritten US Mail with <2% error rate
An application of machine learning
and knowledge-guided
interpretation
Spin-Offs