How Smarsh’s AI Agent Results Are Reshaping Agentforce Adoption

Are the AI agent numbers your vendor is quoting you actually achievable in production? That’s the question most RevOps and customer success leaders are quietly asking right now, as Salesforce’s Agentforce platform moves from conference keynote to real-world deployment. Smarsh just gave them a concrete answer.

The Portland-based communications intelligence company has publicly reported results from its Agentforce rollout that, frankly, stand out. Its customer-facing AI agent, Archie, has reached a 72% self-service deflection rate. Its internal workflow agent, Emmy, is saving support representatives more than seven hours per case. These aren’t projected figures from a pilot program – Smarsh says it’s now scaling the deployment.

Why These Numbers Matter More Than the Hype

Self-service deflection rates are one of the clearest signals of whether an AI agent is actually working. A 72% rate means nearly three out of four customer inquiries are resolved without a human touching the ticket. That has a direct effect on churn rate – customers who get fast answers don’t escalate, and they don’t leave.

Seven hours saved per case on the internal side is the number that should get operations leaders to stop scrolling. Support reps working regulated communications data deal with complex, documentation-heavy cases. Cutting seven hours from that workflow doesn’t just reduce cost – it changes what your support team can actually do in a day, effectively expanding capacity without adding headcount.

The broader context matters here. Salesforce has been positioning Agentforce as the centerpiece of its go-to-market strategy through 2026, and Dreamforce – running this week in San Francisco – is amplifying that message hard. Partners like TELUS Digital are hosting dedicated Agentforce industry sessions and demo stations at Hotel Zetta, separate from the main conference floor. The ecosystem is clearly betting that customer proof points like Smarsh’s will accelerate enterprise adoption.

What Smarsh’s Deployment Actually Involved

Smarsh operates in a highly regulated sector – it helps financial institutions and other compliance-sensitive organizations manage communications data. That’s a demanding environment for any AI deployment. Data governance, audit trails, and accuracy under scrutiny aren’t optional extras; they’re baseline requirements.

Deploying two distinct agents – one customer-facing, one internal – reflects a thoughtful architecture rather than a single monolithic bot. Archie handles the customer interaction layer, while Emmy supports the humans managing case complexity behind the scenes. This split targets different parts of the support cycle with purpose-built tools rather than a one-size-fits-all approach.

Smarsh’s AI agent Emmy is saving support representatives more than 7 hours per case, while customer-facing agent Archie has achieved a 72% self-service deflection rate since deployment on Salesforce’s Agentforce platform.

It’s also worth paying attention to what Smarsh hasn’t claimed. There’s no statement about eliminating headcount or replacing human judgment in sensitive compliance decisions. That restraint is a signal of a mature deployment – AI handling volume and triage, humans handling judgment.

What CRM Teams Should Take From This

If you’re evaluating Agentforce or any AI agent platform for your support org, Smarsh’s results give you a useful benchmark – not a guarantee, but a credible data point to pressure-test vendor promises against. A few things stand out as relevant to most CRM teams:

• Deflection rate is a more honest metric than “tickets handled by AI.” It tells you whether customers are actually satisfied with the self-service experience, not just whether the bot responded.
• Internal workflow agents targeting case handling time can have faster, more measurable ROI than customer-facing agents, especially in complex support environments.
• A dual-agent architecture – one for customers, one for staff – is worth considering if your support cases involve significant internal coordination or documentation.
• Regulated industries achieving these results removes one of the most common objections to AI agent adoption: that compliance requirements make meaningful automation impractical.

For teams thinking through customer lifetime value impact, faster resolution times and higher deflection rates are upstream inputs worth modeling. Customers who don’t wait don’t churn at the same rate – that’s a compounding effect that shows up in net revenue retention over time, not just in support cost reduction.

The Agentforce Ecosystem Is Growing Fast

Smarsh’s announcement lands during a week when the Salesforce partner ecosystem is highly visible. TELUS Digital, a Summit Level implementation partner, is running its own Agentforce-focused programming at Dreamforce 2026 – industry sessions, expert demo stations, and invitation-only events designed to help prospective customers understand what deployment actually looks like rather than just what the platform promises.

That kind of partner investment signals confidence. Implementation partners don’t build out conference infrastructure for a product they don’t believe clients will buy. You can track ongoing developments across the CRM news space as more Dreamforce announcements land this week.

For teams doing their own research, our CRM Tools Directory includes current information on Agentforce alongside competing AI-assisted support platforms. If you want a structured view of how to evaluate AI agent deployments against your specific support model, the CRM Guides section covers implementation frameworks worth reviewing before you commit budget.

The open question isn’t whether Agentforce can deliver results – Smarsh’s data makes that harder to dispute. It’s whether most enterprise CRM teams have the data infrastructure and internal process discipline to actually replicate those outcomes, or whether headline metrics like 72% deflection are achievable only in specific, well-instrumented environments. That gap between benchmark and typical deployment is where most AI agent projects quietly fall short.