With 78% of firms saying no more than one in ten AI use cases runs autonomously, the next phase is “controlled delegation”, not full autonomy.
AI is now part of daily post-trade operations, but humans still lead it. That may be what carries it deeper into core workflows in 2027.
According to AI in the Real World 2026, published by Broadridge with The ValueExchange, International Securities Services Association (ISSA) and AI specialist DeepSee, 78% of firms say no more than 10% of their AI use cases operate autonomously. Some 49% report generative AI in production, against just 11% for agentic AI.
The next frontier is the back office. In asset servicing, 29% of respondents use AI today and 47% plan to, while 70% still describe the process as fully or mostly manual. Settlement, fund accounting and liquidity management also show more planned activity than live activity, and 80% expect to use AI in external interactions by the end of 2027.
The report draws on more than 200 survey responses and interviews. It calls the next stage “controlled delegation”: AI does bounded work within defined permissions, documents its actions and escalates to a person where judgement is needed.
“In the middle and back office, a wrong answer is a settlement fail or a break someone has to chase,” said Steve Shillingford, CEO of DeepSee. “… When every step is verified against the firm’s own process, AI becomes dependable enough to take on more responsibility.”
Governance has to keep up. While 75% of firms have a responsible-use policy, only 46% carry out regular audits or require explainability, and just 24% have processes for building stakeholder trust.
Clients know what they want: 91% cite better accuracy and fewer errors, ahead of faster processing at 85%.
“The test of AI is not how many tools a firm deploys, but whether it delivers fewer errors, better client service and increased capacity,” said David Kirby, head of U.S. Consulting Services at Broadridge.











