Druid AI report finds sector gaps in live agent use

Druid AI has published its 2026 AI Adoption Benchmark Report, based on 15 months of anonymised production telemetry from live AI agents across four sectors.

The study covers healthcare, higher education, financial services, and HR and IT. It examines how AI agents are used after deployment in customer, student, patient, and employee service journeys. The data spans hundreds of enterprise customers and covers production activity from January 2025 to March 2026.

The findings show marked differences between sectors in channel choice, timing of demand, and the share of interactions handled without being passed to a human worker.

Channel split

Healthcare was the only sector in the dataset where voice interactions exceeded chat. The report says 54% of healthcare volume came through voice, compared with 46% through chat.

All other sectors skewed heavily towards text-based channels. Higher education recorded 95% of usage through chat, while financial services showed a 70% share for chat and 30% for messaging. HR and IT logged 94% of demand through chat.

The figures suggest chat has not become the default interface evenly across all sectors in live deployments. In healthcare, phone-based engagement remains central even after AI agents are introduced.

Timing patterns

The data also showed clear differences in when users seek help. Higher education had the highest share of off-hours demand at 39%, followed by financial services at 31% and healthcare at 29%.

HR and IT showed a different pattern. Only 6% of demand arrived outside standard working hours, but requests were sharply concentrated at the start of the day, with 24% of total volume coming between 9 a.m. and 10 a.m.

Peak-hour data showed healthcare reaching its busiest point at 10 a.m., when 8% of volume was recorded. Higher education peaked at 2 p.m. with 8%, financial services at 12 p.m. with 8%, and HR and IT at 9 a.m. with 12%.

Druid AI argues that these patterns create two distinct business cases for AI agents. In healthcare, higher education, and financial services, the case centres on continuity of service because a large share of demand arrives outside normal hours. In HR and IT, it is more closely tied to absorbing concentrated demand during short busy periods.

Resolution rates

The report found a wide spread in containment rates, which measure the share of interactions concluded without a handover to a human. Higher education recorded the highest level at 99.5%, while financial services posted the lowest at 80%.

Healthcare reached 87%, and HR and IT stood at 93%. According to Druid AI, the gap reflects how organisations design escalation rules around issues such as compliance, clinical review, fraud risk, security approvals, and policy exceptions, rather than serving as a simple measure of AI quality.

The composition of demand also varied significantly by sector. In financial services, three workflow types accounted for 90% of all production volume, while in higher education three workflows made up 92% of usage. Across all four sectors, most demand came from front-line service tasks such as customer and student support, patient access, and workplace operations.

That concentration suggests many organisations are deploying AI agents first in a relatively narrow set of high-volume tasks rather than across broad areas of operations. Druid AI says this creates an entry point for later expansion into more complex processes that require integration with internal systems and tighter controls over when cases move to human staff.

Joseph Kim, Chief Executive Officer of Druid AI, outlined the company's rationale for compiling the data. "There have been plenty of 'State of AI' reports based on surveys that illustrate the current sentiments on Agentic AI. At Druid, what we thought might add more value is to share what these agents are actually doing once in production. After analyzing 15 months of AI agent data across four industries and hundreds of enterprise customers, the patterns on what is working and how you can make it work are clear," he said.

Druid AI was founded in 2018 and supports more than 350 clients globally through a partner network of more than 250 organisations.