SF's Duplicate Image Problem: What Happens Next and the Key Decisions Ahead

San Francisco's coordinated entry system for homelessness services is sitting on a growing pile of duplicate client records, and administrators at the Department of Homelessness and Supportive Housing are now under pressure to decide how — and how fast — to clean them up. The problem has compounded since the city's Shelter Reservation System was expanded in 2024, adding thousands of new intake photographs and biometric check-ins that weren't consistently deduplicated against legacy records dating to 2019.

The stakes are real. When a single unhoused person appears in a database two or three times under slightly different names or outdated photos, caseworkers at Navigation Centers on 13th Street in SoMa and at the Civic Center area's Tenderloin housing intake offices can end up duplicating outreach efforts — or, worse, missing someone entirely because a slot looks filled when it isn't. With the city's official count of unsheltered residents still in the thousands, nobody can afford to waste a bed on a ghost record.

Why This Summer Is the Decision Point

The timing matters for a specific bureaucratic reason: the Department of Technology's contract with its current Homeless Management Information System vendor is up for renegotiation before October 1, 2026. Whatever data governance rules the city writes into the new contract will define how duplicate images are flagged, quarantined, or merged for the next five-year cycle. Advocates and city officials who track housing policy say the window between now and Labor Day is effectively the only practical moment to reset the standards before they get locked in again.

The San Francisco Human Services Agency, which shares data pipelines with the homelessness department, has been piloting a record-linkage protocol since February 2026 at its offices on Otis Street. That pilot uses probabilistic matching — comparing partial name strings, intake dates, and photo metadata — to surface likely duplicates for human review rather than automated deletion. Early results from the first 90 days of the pilot, according to program documentation circulated to the Board of Supervisors' Government Audit and Oversight Committee in April, flagged more than 1,200 record pairs as probable duplicates out of roughly 28,000 active profiles. That is a duplication rate of just over four percent — high enough to distort bed-availability figures in real time.

The decision about what to do with flagged records is not purely technical. Deleting an image outright risks erasing documentation that a caseworker at, say, Glide Memorial Church on Ellis Street or at the Episcopal Community Services drop-in on Turk Street uploaded as proof of a client interaction. Merging records requires a staff member to manually confirm identity, which costs time that stretched intake teams don't have. Quarantining duplicates preserves the data but doesn't fix the underlying count problem.

Three Paths Forward — and the Political Dimension

City technology staff have outlined three broad options in internal planning documents: an automated merge-and-flag workflow requiring vendor customisation estimated to cost between $400,000 and $600,000; a manual review queue staffed by a new data quality team of four FTEs; or a phased hybrid that starts with automated flagging and routes only high-confidence matches to a smaller review team. Each option requires sign-off from the Department of Homelessness, the Department of Technology, and the City Administrator's Office before the contract deadline.

The Board of Supervisors' budget cycle, which wrapped its main hearings in June, did not include a dedicated line item for database remediation — meaning any approved approach will have to be funded either through existing departmental reserves or a supplemental appropriation request this fall. That adds a political layer: supervisors representing the Tenderloin, SoMa, and the Mission have all pushed publicly for faster shelter placement times, but authorising new spending on back-end data infrastructure is a harder sell in a year when the city is already projecting a significant general fund shortfall.

The practical next step for anyone tracking this issue is the Government Audit and Oversight Committee's scheduled hearing in mid-July, where department heads are expected to present a preliminary recommendation. Community organizations that plug into the coordinated entry system — including Hamilton Families, which operates shelter programs in the Lower Haight, and Compass Family Services on Polk Street — have until that hearing to submit public comment on which approach best protects client data integrity. The choice made in that room will quietly determine how accurately San Francisco can count — and serve — its most vulnerable residents for the rest of the decade.