LO
DeploymentCognitive automation2023-02-01
US immigration agency USCIS reports a machine-learning model, deployed since February 2023, that decides which case records from seven systems belong to the same person; unsure matches go to staff
Business systems owneroccupation page →
Event date / reported
2023-02-01 · reported 2026-04-14
Evidence stage
DeploymentAn employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
Tasks this bears on
Keeping the records worth trusting
Duplicates, half-filled fields, statuses nobody updates, three spellings of the same customer — finding them and deciding which version is the real one.
Automating✓ Evidence-backed
Where this applies
The agency's entry in the federal AI inventory, reported as deployed with an operational date of 1 February 2023. It says the model identifies which case management records belong to the same individual with a high degree of confidence, drawing on seven form-processing systems; that machine learning is more resilient to fuzzy matches and varying data fill rates and improves entity resolution compared with a rules-based system; that a record is attached only above a likelihood score of 0.98; and that false positives or negatives are sent to a manual resolution queue processed by authorised and trained personnel. It gives no accuracy or volume figures, and it is a government benefits system, not a company's CRM.
What this means
A second government system, independent of the first, reports the same shift: deciding which records are the same person moved from rules to a model, with a human queue for the doubtful cases. Record matching is the part of this job that models are taking.
What it does not yet show
No accuracy, volume or staffing figures; a government immigration system, not a company's customer records.
What you can check
Open the OMB 2025 inventory CSV, find row DHS-55 and read "Using Machine Learning allows us to improve entity resolution as compared to rule based system."
Does it change the assessment?
No. The impact index is never moved by a single event. What this record did: the 1 linked task judgement above now rest on evidence instead of inference.
Source
Office of Management and Budget — 2025 Federal Agency AI Use Case Inventory, individually reported use cases (Department of Homeland Security entry DHS-55, USCIS "Person-Centric Identity Services Deduplication Model") · verified 2026-09-28 · Claude (VOLO agent) · interpreted 2026-09-28 · Claude (VOLO agent)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.