Manulife is moving artificial intelligence from isolated tools into the workflows that run a global financial institution. The Canadian insurer is building a runtime platform for agents that can gather information, move between software systems and complete sequences of internal tasks.
The ambition is large. Manulife expects its AI initiatives to generate more than $1 billion in value by 2027 through productivity gains and workflow automation. It already has more than 35 generative AI use cases in production, aims to reach roughly 70 and says about 75% of its global workforce uses generative AI in some form.
Those figures show broad adoption. The agent platform tests something harder: whether AI can act inside regulated operations without weakening control.
From answers to operational action
Insurance work moves through many systems. Claims records, policy information, underwriting assessments and financial reports may pass between several teams before a decision is reached. Employees often spend substantial time collecting and organising information rather than applying judgment.
An agent can potentially retrieve records from multiple internal sources, assemble a case summary and prepare material for a reviewer. Unlike a chatbot responding to one prompt, it can execute a series of steps across tools and maintain the state of the task.
That distinction is where much of the value lies. Writing assistance and document summaries improve individual productivity; workflow agents can change how an entire process operates. Claims management, policy administration, reporting and support all contain high-volume activities that are candidates for carefully bounded automation.
The wider market is still early in this transition. Research cited in the source found that generative AI use had spread rapidly across organisations, but only a small share of deployments had reached broad production. Connecting an impressive model to real systems, permissions and exception handling remains the difficult part.
Regulation makes governance part of runtime
Financial agents cannot operate on capability alone. Underwriting, claims and investment processes are subject to requirements around data use, fairness, transparency and accountability. A mistake can be amplified if an automated workflow repeats it across thousands of cases.
Manulife says its platform includes security and governance controls that manage how agents interact with internal systems. These controls are intended to track data use, monitor how outputs are produced and keep activity within company policy.
Auditability must extend beyond the model’s final answer. Operators need to know which systems an agent accessed, what actions it took, where human approval occurred and how an exception was handled. Permissions should reflect the task rather than grant broad access simply because an agent might find it convenient.
Gradual deployment provides another safeguard. Internal workflows allow a company to measure accuracy, escalation rates and employee behaviour before exposing the same mechanisms directly to customers. People remain responsible for decisions whose consequences require professional or regulatory judgment.
The billion-dollar target needs operational evidence
Forecasting $1 billion in value creates a clear test. Benefits must appear in processing time, reduced manual work, improved data quality or better decisions—not merely in the number of employees with access to a tool.
External research referenced in the source suggests AI automation could reduce financial institutions’ operating costs substantially over time. The outcome depends on the process selected. Automating a fragmented workflow may only move errors faster, while a well-governed agent can remove tedious collection work and let specialists focus on complex cases.
Manulife’s scale gives it an important advantage: many repeated processes and enough usage to learn where agents help or fail. It also increases the stakes, because a poorly designed action can travel quickly through a large organisation.
The insurer is therefore testing the defining enterprise AI question of 2026. Models can already generate useful content. Can a company turn that intelligence into dependable action while preserving oversight?
If Manulife can connect agents to live financial systems, demonstrate measurable returns and retain a complete accountability trail, the technology will become less like an assistant and more like operational infrastructure. That is where the largest productivity gains may sit—and where governance has to be strongest.
