A housing authority cleans free-text address fields with AI assistance before geocoding work orders. What data-preparation goal does this serve?

Cleaning free-text address fields with AI assistance before geocoding work orders serves normalization: messy housing-authority strings (misspellings, unit designators, directionals) are standardized so geocoders can match municipal street centerlines more reliably. The preparation goal is higher match rates and fewer false pins for maintenance and inspection routing—not inventing geography or bypassing review on sensitive resident locations. Address cleanup is classic civic data prep upstream of spatial operations.

Inventing new street names the post office does not recognize fails because fabrication breaks postal and GIS truth and can misroute crews. Publishing raw free-text addresses into a public LLM training scrape fails by concept and privacy: housing work-order addresses are often sensitive location data and must stay in approved environments. Skipping human review of any address the model rewrites fails because model “corrections” can silently move a unit to the wrong building; high-impact rewrites need sampling and exception queues.

Exam caveat: Normalization helps geocoding but does not replace a maintained address points layer or alias table. Apartment and basement units remain hard cases; measure match confidence, not just string prettiness.

Operational check: Run AI-assisted cleanup into a staging table, compare geocode hit rates before/after on a gold sample, queue low-confidence rewrites for staff, and block any export of raw addresses to consumer AI. Keep authoritative site addresses in the housing system of record with lineage from prep.