AI summary by TwinMind
How the US Postal Service reads terrible handwriting
Tom Scott
7m 53s
US Mail Processing Automation Overview
- Almost all US letters processed automatically via OCR technology
- Letters go to processing centers where cameras/computers photograph envelopes
- OCR reads addresses, routes mail to correct trucks/planes
- Handles most mail including handwritten letters
Remote Encoding Center (REC) Operations
- Last remaining REC located in Salt Lake City, Utah
- Handles mail OCR technology cannot read (bad handwriting, blurred ink, damaged envelopes)
- Currently 810 employees at facility
- Processes mail that automated systems reject
Historical Mail Volume Data
- 1997: 55 RECs processed 19 billion images combined
- 2021: Single remaining REC processed 1.2 billion images
- OCR technology improvement: <50% computer-read mail in 1990s → ~99% today
- REC handles ~1% of mail that OCR cannot read
- Successfully processes about half of that 1%
REC Processing Technology and Infrastructure
- Three internet service providers with fiber optic lines at different building entry points
- Connects to 300+ processing plants nationwide
- Serves locations including Guam, Anchorage/Juneau Alaska, San Juan Puerto Rico
- Uses specialized keyboards with modified layouts for faster data entry
- Home row keys (A,S,D,F) function as numbers to avoid reaching top row
Mail Processing Workflow and Timing
- Live mail: 90-second processing deadline before rejection to hand-sort bin
- Letters: Run through machine twice (stored between runs for REC processing)
- Packages/flats: Remain on conveyor belt during processing
- Average processing time: 4 seconds per piece
- Required speed: 7,150 keystrokes per hour
Address Coding System (Siemens 1990s)
- "C coding": ZIP code
- "Outward": City and state
- "Inward": Street address
- 3+1 rule: First 3 characters of first word + first character of second word
- System compares partial entries against complete US address database
Quality and Accuracy Standards
- Keyers type exactly what customer wrote (no corrections/interpretations)
- System matches partial input to known-good addresses in database
- If multiple matches, keyer selects from list
- Failed pieces go to reject bin → manual handling → return to sender or Mail Recovery Center
Current Challenges
- Decreasing mail volumes (quiet during summer)
- Remaining mail increasingly difficult to read (since OCR rejects it)
- Addresses often damaged or contain incorrect information
- Job difficulty increasing as only hardest-to-read mail reaches REC
