Wealth managers are not short of AI use cases. They are short of operating capacity.
As books grow, firms need more adviser capacity and client coverage without letting cost-to-serve, compliance workload and operational complexity grow at the same pace. The first wave of AI promised to close that gap, and task by task it has delivered. Assistants now research clients, summarise documents, prepare meetings and draft communications across the industry.
The limits show up after the assistant finishes. An AI-generated meeting summary still has to be moved into the CRM, converted into follow-up tasks, checked against suitability requirements and routed for approval. One task got faster, but the process that actually determines cost-to-serve remains manual and fragmented.
That is the boundary between the two waves of AI adoption, and it comes down to a single operational test: who completes the process? In the first wave, a person completes it with machine help. In the second, the system plans, reasons and executes the workflow end to end, and the person supervises by exception. Where previous tools assisted, agentic systems execute.
Why the first wave is plateauing
The returns from the first wave are flattening. The tools are not the problem; task-level speed simply doesn’t compound. MIT research reported in 2025 that roughly 95% of enterprise GenAI pilots showed no measurable P&L impact, and a Bank of England/FCA survey found that while 55% of financial-services AI use cases involve some automated decision-making, only 2% run fully autonomously. The gap between those two numbers is where the second wave will be won or lost.
Crossing that gap takes more than deploying another tool, and more than renting a frontier model, because every firm can rent the same intelligence by API. Three things cannot be rented: governed access to the systems where the operational truth lives, the earned judgment of the people who know what a correct outcome looks like, and the engineering discipline to put frontier reasoning safely between the two.
In the whitepaper we have just published with IBS Intelligence, we call that convergence the Agentic Triada: context custody (privileged, governed access to the systems of record, not the custodian bank), embedded judgment (knowing which exceptions matter and what regulators will stand behind), and the reasoning frontier (command of the model rather than the model itself: orchestration, validation, cost-routed reasoning, every decision logged and replayable). One strand alone is a stranded asset. Two stall in pilot purgatory. Only the convergence crowns a winner.
What this looks like in production
Live reconciliation agents, run with the operations team of a Nordic fund administration business on real daily workloads and held to full audit standard, have cut process cost by around 75% and freed roughly 20% of operator capacity, with no missed breaks across all break classes and 100% of decisions logged, each one verifiable through the audit trail. Automation identifies breaks; only reasoning resolves them, and properly routed, most of the tail is cheap.
The regulator is ahead of the market
The FCA’s Mills Review, published in July 2026 as the first regulator-initiated review of its kind, named changes to firm operations as the first of four AI-driven shifts and found 20% of consumers already open to AI acting autonomously within pre-set goals. The FCA has also confirmed there will be no new AI rulebook: existing frameworks such as Consumer Duty and SM&CR govern. In a principles-based regime, the audit trail is the compliance story. That makes supervised, logged, replayable execution the licence to operate for agentic AI.
From assistance to governed execution
Moving to the second wave requires trusted data, integration with CRM, portfolio, compliance and back-office systems, clear decision rights, approval controls and end-to-end auditability.
ProfileOne is designed for this shift. It connects data, decisions, applications, workflows and supervised AI agents in one governed, production-grade orchestration environment, built to coordinate execution across your existing estate rather than replace it.
Read the original article here.
