Florence introduced Farr, an AI agent for study teams, and expanded the AI workflows it already runs for study startup and monitoring. Each one puts humans over software: the AI drafts, checks, and recommends, and research teams decide.

"Sites and sponsors share one journey, but today they connect it by hand: emails, spreadsheets, re-keyed questionnaires, and documents fixed after the fact," said Shankar Jagannathan, COO of Florence. "These AI workflows take on that work, from the first feasibility questionnaire to the documents an inspector will check, so research teams spend their time on decisions and patients."

Farr: ask a question, get an answer

Farr is the AI agent in TrialSight. It answers plain-English questions about how a trial is progressing, grounded in Florence trial operations data: 12 million events a month across 50,000 protocols and 44 million documents. It suggests next steps for a person to review, so study teams spend less time digging through reports and more time acting on what they find.

Study startup: answer once, not for every sponsor

Study sites answer the same feasibility questions for sponsor after sponsor. Site Feasibility, live since September, turns each site's past answers into a reusable library and uses AI to recommend answers.

• Answer library: Every past answer is stored and searchable, so sites stop retyping the same information.
• AI-recommended answers: AI suggests answers from the site's own library for the site team to review.
• Works where sponsors work: A browser extension brings the same help into sponsor portals and web surveys.

On the sponsor side, Florence is previewing two AI features at Research Revolution:

• AI survey design: AI drafts a feasibility survey from the protocol or synopsis, and the sponsor team makes the final call.
• Ambient AI Assist: AI drafts the site qualification form from the pre-study visit call, for a person to review.

Monitoring and document quality: catch problems when documents arrive

Florence's network holds more than 44 million documents. Across 88,000 study sites, DocQC catches problems when a document arrives, not at the next monitoring visit.

• DocQC: Flags unreadable scans, blank pages, rotation problems, duplicates, and possibly missing pages as documents enter Florence.
• Classification: Recommends each document's type and filing location. The user accepts before anything is final.

"For a 500-page document set, what would normally take us about three hours to review manually was done in around 30 minutes with DocQC. That's a significant amount of time we can give back to our team," said Dr. Louise Rossiter, R&I digital transformation lead at University Hospitals of Leicester NHS Trust.

"Every workflow we're showing this week keeps a person in charge of the final decision," said Shree Shirgurkar, VP of Product at Florence. "The AI drafts, checks, and recommends. Research teams decide."