Banco Société Générale Brasil now runs 95% of its foreign exchange contracts through Hyland OnBase, which handles distribution, signature tracking and follow-up reminders and has ended the printing of about 230,000 pages a year (Hyland, 2026). The Société Générale subsidiary connected OnBase to the back-office system that generates the contracts and to the digital signature platform it already used. In the Brazilian state of Espírito Santo, the traffic department DETRAN-ES moved reports of accidents without injuries online, cut the backlog of those reports by 99% and moved 86% of the police officers who had been doing that paperwork back to frontline duty (Hyland, 2026).

These two results come from a set of customer stories Hyland has published across banking, health care, higher education, insurance and government. Hyland presents the work as the foundation for what it calls the content-powered agentic enterprise, in which AI agents read a company's documents to answer questions and complete tasks. I looked at how Hyland connects those documents to other systems in my post on Hyland's Guidewire integration. The customer stories show what organizations had to do with their documents before any agent could use them.

Most of that work had nothing to do with AI.

Hyland's biggest numbers come from organizing documents, years before any agent arrived

MetroHealth, the public safety-net health system in Northeast Ohio, started with OnBase for administrative and clinical workflows connected to its Epic electronic health record. ECM Connection reported the system's Hyland imaging archive deployment in February 2017. MetroHealth later added Hyland Acuo as a central image repository and the NilRead viewer inside clinical workflows. Missing reports fell by more than 80%, and 8,000 staff members accessed more than 4.7 million images over three years (Hyland, 2026). Nearly 33,000 patients viewed their own images through the patient portal in the first seven months (Hyland, 2026).

David Kaelber, MetroHealth's chief medical informatics officer, estimates the project was roughly 10 to 15% about the technology and 85 to 90% about everything else. The everything else was an enterprise imaging steering committee, co-chaired by the chair of radiology, that brought radiology, IT, clinical engineering, clinical informatics and point-of-care ultrasound into one room to agree on how images are stored and who can see them.

A Brazilian steel producer Hyland describes without naming did a different version of the same work. The company moved about 500 document types and 21 million engineering files off OpenText Documentum and onto OnBase on a short deadline (Hyland, 2026). Each of those 500 document types records a decision about what a file is, who may open it and how long the company keeps it. An agent asked about the latest revision of an engineering drawing can only answer correctly if those decisions were made and enforced first, because the agent inherits whatever labels and permissions the repository holds.

"Consistent, real-time answers only work when the information behind them is governed and trusted."

Stephen Watt, SVP and CIO, Hyland

The AI already in production works on one document type and sends hard cases to people

Hyland announced Intelligent Transcripts on August 12, 2026, for college admissions and transfer offices. The product reads multi-page academic transcripts that arrive in many formats, checks them and routes them into OnBase and the student information system. It sends a transcript to a human reviewer only when its confidence falls below a threshold the institution sets. Hyland says a manual transcript evaluation often takes more than 20 minutes per document (Hyland, 2026).

Valerie Campbell, associate registrar at Grand Rapids Community College, said the tool "significantly reduces overall transcript processing time" and that its correction workflow cuts down on input errors. Hyland has not published a time-saved figure for the college yet, which is reasonable for a product announced three months ago.

DETRAN-ES runs the same pattern at state scale. Its vehicle service request system, built on OnBase and connected to the department's own vehicle records platform, uses AI-assisted validation to check the documents citizens submit. Both deployments apply AI to a single, well-defined document type with a check the software can score, and both send low-confidence cases to a person. You can copy that design for a first agent without waiting for a broader platform decision.

Hyland built its first agent for its own HR department

The clearest Content Innovation Cloud example Hyland has published is internal. Hyland's HR team built an open enrollment agent on the platform in 2.5 weeks, using Hyland Knowledge Discovery to index benefits documents and Knowledge Enrichment to interpret employee questions (Hyland, 2026). Employees ask questions in the HR benefits portal or in Microsoft Teams and get answers that link back to the source document.

The first step was consolidation. HR gathered plan guides, comparison charts, deduction tables, policy updates and FAQs into a single OnBase document type, so the agent had one governed source to read from. Hyland reports that the agent handled most routine enrollment questions and that its legal, finance, security and cloud operations teams are exploring the same approach. Hyland has not published usage counts or accuracy figures for the agent.

Hyland says customer adoption of the Content Innovation Cloud grew 220% in the fourth quarter of 2025 (Hyland, 2026). I hope the next round of customer stories includes an outside organization describing a Content Innovation Cloud agent in production, with the kind of measured results Société Générale Brasil and DETRAN-ES reported for their OnBase work. That story would tell buyers how long the agent step takes once the document groundwork is done.

Your repository mix sets how far a Hyland agent can reach

Hyland's agents reach documents across repositories through its Content Federation Service. In its February 2026 announcement, Hyland listed OnBase, Alfresco, Nuxeo and SharePoint 365 among the supported sources. Perceptive Content, the system formerly called ImageNow and still common in university records offices, was not named in that list, though the announcement said the list was not complete.

Colleges and universities weighing products like Intelligent Transcripts will hit this question first, because their admissions and records documents often sit in Perceptive Content. Permissions, retention rules and legal holds stay with each source system under federation, a tradeoff I covered in the Guidewire post. For a first agent, the practical question is narrower: which repository holds the document type you want the agent to read, and does Hyland support that repository today.