Smarsh, a compliance and communications-archiving vendor for financial services, published production results on September 3 for two AI agents it built on Salesforce’s Agentforce platform. Archie, a customer-facing support agent, resolved 72% of interactions without human intervention across 405 sessions in the second quarter, with a resolution-confidence score of 2.6 out of 3. Emmy, an internal agent that gives support representatives account snapshots and case history, reached 65% adoption among representatives, with 120 of 185 reps using it and 31% of cases resolved with its help since a late-March launch, against a 20% target.
“The most important result isn’t simply that Archie can resolve routine needs faster,” said Rohit Khanna, Chief Customer Officer at Smarsh. Salesforce’s own read leaned on the same data: “Smarsh is a standout example of how Agentforce is helping enterprises scale trusted AI,” said Greg Beltzer, SVP-Agentforce at Salesforce.
The development matters because most agentic AI vendors, including the four covered in this publication’s feature on AI moving from add-on to default infrastructure today, announce capability and pricing without disclosing what happens once an agent goes into production. Smarsh’s numbers are one of the few concrete adoption-and-outcome data sets available for an Agentforce deployment at meaningful volume, rather than a pilot.
The original insight for a RevOps leader building a business case for agentic AI is in the gap between the two agents’ numbers. Archie, facing customers directly, needed months of tuning to reach a 2.6-out-of-3 confidence score even at a 72% deflection rate, while Emmy, an internal tool with a lower bar for error, beat its adoption and time-saved targets within one quarter. This publication’s reporting on Salesforce’s Winter 27 release found agents increasingly handed full ownership of workflows rather than assisting a human. Smarsh’s data suggests that shift is happening fastest on internal, lower-stakes workflows first, and RevOps leaders sizing a rollout should sequence pilots the same way rather than starting with customer-facing deployments.
Source: Smarsh
