Smarsh deploys agentic AI in regulated finance with measurable gains
Smarsh, the Portland-based communications-data company whose client base includes nine of the ten largest banks in North America, Europe and Asia, has published operational results from its first year running agentic AI across customer support. The numbers are notable not because they are record-breaking but because they are auditable: a 72% self-service deflection rate across 405 live customer interactions in Q2 2026, and a resolution confidence score of 2.6 out of 3, derived from manual session review rather than self-reported satisfaction. For the financial services operators who make up the bulk of Smarsh's customer base, the distinction matters.
The company deployed two agents built on Salesforce's Agentforce platform. Archie handles inbound customer support queries, resolving more than seven in ten sessions without routing to a human representative. Emmy, the second agent, works internally: it serves support staff with instant account snapshots, ownership histories and case summaries, and offers AI-assisted resolution guidance at case closure. Launched in late March 2026, Emmy reached a 65% adoption rate among the company's 185 support representatives by end of June, against a target of 50%. Teams using Emmy reported saving an average of 7.5 hours per case compared with equivalent cases handled by the same staff before the deployment, exceeding a five-hour internal target.
From experiment to enterprise template
What makes the Smarsh deployment of particular interest to the broader market is the governance scaffolding built around the performance metrics. Financial services firms have been cautious about agentic AI precisely because the regulatory environment, spanning MiFID II oversight, SEC communications-retention rules, and FINRA surveillance requirements, demands that every automated decision carry a defensible audit trail. Smarsh's architecture, which sits on top of a communications-capture and retention platform already trusted by regulated institutions, is positioned as evidence that the two objectives need not conflict. "The most important result isn't simply that Archie can resolve routine needs faster," said Rohit Khanna, Smarsh's Chief Customer Officer. "It's that our customers can trust us to get a consistent experience while our people have more time to focus on the complex issues where their knowledge and judgment matter most."
The company has also filed for a US trademark on the Archie name, receiving a Notice of Allowance from the US Patent and Trademark Office, a move that signals intent to build the agent into a durable product brand rather than treat it as a one-cycle pilot.
Cross-sector read-across: compliance as a competitive moat
The story carries implications that reach well beyond Smarsh's own install base. Across the financial services sector, the pace of agentic AI adoption has been constrained less by capability than by institutional risk appetite: legal teams and chief compliance officers have been slow to approve deployments where AI responses could trigger regulatory liability. Smarsh's published deflection and confidence metrics, backed by a named methodology, provide a reference architecture that compliance functions at peer institutions can point to when building internal business cases.
The dynamic also has capital implications. Enterprise software vendors competing in the regulated-industry AI space, including players in legal-tech, insurance claims processing and government communications management, are watching whether a defensible governance wrapper can become a durable commercial differentiator. If Smarsh's model proves replicable across other high-compliance verticals, it reframes agentic AI not as a productivity tool but as a regulatory-risk management layer, a framing that shifts the buying conversation from IT budgets to the C-suite and board risk committees. Smarsh will present operational lessons from the deployment at Salesforce's Dreamforce conference on 16 September 2026.