Altilia said a leading European and Italian credit servicer managing secured NPL portfolios used Altilia agents on private infrastructure to read an archive of 130 million pages without adding headcount, in a customer story dated 21 May 2026.
Document types spanned mortgage contracts and registrations, guarantees, sale orders, court-appointed appraisals, property registers and statements of liabilities — material that due diligence and recovery strategy depend on reading in full. Per case the pipeline targets about 330 datapoints, including creditor, debtor and guarantor identities, full cadastral description, appraisal values, liens and ranking. The agent pipeline covers classification of every page to document type on servicer-tuned models; extraction of entities and objects; reconciliation across documents with confidence scores and source hierarchy; and semantic, regex and keyword search returning the page as evidence.
A pre-live audit of a 10,000-document sample against manually verified values reached 98.3% accuracy, with low-confidence values already flagged for review. Outcomes cited under the servicer's methodology include manual effort down about 80% and case preparation about 10× faster, with analysts shifting from locating and re-keying to deciding on complete information. The head of credit operations at a leading Italian banking group, anonymised in the story, said: "Their AI agents work with our own data in a private infrastructure, delivering accuracy and control that no generic AI solution could match." Metrics and the anonymised customer quote are company statements; AI-360 has not independently verified them.