From Pilot to Scale: How Smarsh Unlocked 60% Resolution Rates with Salesforce Agentforce

When compliance technology leader Smarsh set out to transform its customer experience, it didn't just deploy a chatbot. It built a multi‑agent AI ecosystem that now resolves 60% of customer inquiries autonomously, and it did so while reimagining the economics of support. For support leaders watching from the sidelines, the Smarsh story offers a rare, metrics‑backed blueprint for scaling AI without breaking the bank or alienating customers.

In this deep dive, we’ll unpack the Smarsh‑Agentforce partnership, dissect the $2/conversation pricing model, and extract the frameworks your team needs to replicate this success. Whether you’re running a lean SaaS support team or a 500‑seat B2B contact center, the lessons here will help you build a business case your CFO will love.

The Smarsh Success Story: Real Results with AI‑Powered Support

Smarsh, a leader in digital communications compliance, knows that regulatory complexity never sleeps. Its clients, banks, broker‑dealers, and government agencies, demand fast, accurate answers to questions about e‑discovery, retention policies, and 3,000+ data sources. Traditional tier‑1 support teams were drowning in volume, while the expertise required for tier‑2 and tier‑3 issues made scaling headcount prohibitively expensive.

Enter Salesforce’s Agentforce, Einstein GPT, and a deliberate, phased rollout. Smarsh didn’t just switch on an AI copilot; it hand‑crafted specialized autonomous agents that mirror the way its human teams operate. As Patterson, a key architect of the strategy, explains, the key wasn’t one all‑purpose bot, it was a suite of purpose‑built agents that could understand context, pull from knowledge bases, and even hand off between themselves without human intervention.

"We distinguished specialized agents from Salesforce’s broader orchestration platform. Fin is a customer‑service‑focused agent, while Qualified’s Piper is a marketing agent designed to qualify website visitors and book sales meetings. Each plays a distinct role in our unified journey."

The results were immediate and quantifiable. Within three months, Agentforce was handling thousands of conversations, deflecting routine tickets, and surfacing answers from Smarsh’s proprietary knowledge graph. Customer satisfaction didn’t drop, it rose 12 points, because response times went from hours to seconds and answers were more consistent.

The Economics of AI Agents: Understanding $2/Conversation Pricing

Salesforce has made Agentforce pricing refreshingly transparent: $2 per completed, autonomous conversation. But simple math reveals why finance teams need to look beyond the per‑conversation sticker price.

Doing the Real Math on AI Support Costs

Let’s assume Smarsh’s use case: at 10,000 monthly conversations with a 60% resolution rate, here’s what the cost picture looks like.

At $2 per conversation, the AI layer alone costs $20,000 per month for 10,000 conversations, before any Salesforce license fees. But comparing that to the $60,000+ spent on fully manual handling, the net savings exceed 40%. And that doesn’t account for the customer experience lift: faster answers, 24/7 availability, and no queue frustration.

For a more granular view, look at cost per resolved conversation. Only 6,000 of the 10,000 conversations get purely AI resolution, so the effective cost per AI‑resolved interaction is $2. Meanwhile, the 4,000 that escalate to humans still cost $6–$8 each. That blended cost per resolved interaction becomes roughly $3.67, still a fraction of the pre‑AI $7 average.

Building a Scalable AI Support Stack: Lessons for B2B SaaS Companies

Smarsh’s success isn’t magic. It’s built on deliberate design choices that any support operations leader can adopt.

Specialization beats generalization

Rather than training one oversized model, Smarsh deployed specialized agents, Fin for customer support, Piper for marketing qualification. This allowed each agent to master a narrow domain, reducing hallucinations and increasing containment.

• Fin sits inside the support console, reads tickets, searches knowledge articles, and suggests or auto‑completes responses. It knows compliance terminology and can even navigate the complexities of financial regulations.
• Piper lives on the website, identifies high‑intent visitors, answers pre‑sales questions, and books meetings directly into sales reps’ calendars. It speaks the language of use cases and product differentiators.

Tight integration with your knowledge base is non‑negotiable

Agentforce doesn’t work out‑of‑the‑box unless your knowledge base is clean, structured, and comprehensive. Smarsh invested months mapping its 10,000+ help articles, troubleshooting guides, and policy documents into a graph‑based knowledge foundation. The result? The AI could confidently answer 80% of the questions that previously required a level‑2 engineer.

Phased rollout with human‑in‑the‑loop fallback

Smarsh started Agentforce in "suggestion mode" where human agents approved every AI response. After two months of supervised learning and retraining, it transitioned to autonomous mode for high‑confidence topics. Even today, any conversation that dips below a 90% confidence threshold is instantly routed to a human with full context, a model that keeps customer trust intact.

"We didn’t replace our support team; we gave them a co‑pilot that handles the routine so they can focus on the complex. That’s the future of B2B support.", Smarsh support operations lead

Measuring the Impact: Metrics That Matter for AI‑Driven Support

If you’re building a business case for AI agents, your CFO will ask three questions: How much do we save? How much does experience improve? And how do we measure ongoing success?

Here’s the Smarsh‑inspired KPI framework you should adopt.

1. Containment rate (also called resolution rate): The percentage of conversations fully resolved by AI without human touch. Smarsh targets >55%; the initial 60% exceeded expectations.
2. Cost per ticket (CPT): Pre‑AI, Smarsh’s blended CPT hovered around $7. Post‑AI, it dropped to $3.67. Track this monthly to ensure AI investments pay off.
3. Mean time to resolution (MTTR): From 12+ minutes to under 2 minutes for 60% of conversations, a 6x improvement that directly lifts CSAT.
4. Agent productivity: With AI handling tier‑1, agents now resolve 40% more complex tickets per shift, and escalation rates fell by 35%.

The ROI Multiplier Effect

The financial impact isn’t linear. Deflecting 6,000 conversations per month doesn’t just save $36,000 in labor, it prevents the need to hire additional agents as volume grows. Smarsh, for example, absorbed a 25% increase in support volume over six months without adding a single headcount. That’s the scalability that makes SaaS CFOs sit up and listen.

The Human‑AI Collaboration Model: How Smarsh Empowered Agents

One of the biggest myths about AI in support is that it kills jobs. The Smarsh story proves the opposite. By automating the repetitive, password resets, “how do I find my reports?” questions, basic troubleshooting, AI made the remaining human work more engaging.

From ticket‑takers to problem‑solvers

Before Agentforce, Smarsh’s tier‑1 agents spent 70% of their time on tasks that didn’t require human judgment. After AI took over those conversations, the same agents were upskilled to handle complex compliance interpretations, account configuration, and even proactive customer health outreach. Agent satisfaction scores rose, and turnover dropped 22%.

Cross‑functional AI orchestration

Smarsh’s use of both Fin and Piper shows the power of cross‑department AI collaboration. A visitor who starts a chat on the website can be qualified by Piper, seamlessly handed to Fin for a support question about a potential integration, and then routed to a sales rep, all without the customer repeating a single detail. This orchestration is where the real CX magic happens.

For companies built on Salesforce, this is a natural advantage. But even if your stack is different, the principle holds: AI agents should share context across the customer journey, not live in silos.

Beyond Salesforce: Future‑Proofing Your Support with an AI Layer

Agentforce is powerful, but it’s not the only path. Many SaaS companies run on Intercom, Zendesk, or home‑grown support platforms, and they need AI that plugs in without rip‑and‑replace. That’s where a dedicated AI support automation layer can be a game‑changer.

Platform‑agnostic solutions like Successly let you build custom AI agents that sit on top of your existing ticketing system, knowledge base, and CRM. You can replicate Smarsh’s specialized‑agent model with Fin‑like capabilities, but without being locked into one vendor’s ecosystem. And you can start with a $2‑style pricing that scales with your actual conversation volume.

Successly, for instance, delivers 43% ticket deflection within the first 30 days of deployment by training on your existing help docs and past conversations, a result that directly mirrors the Smarsh trajectory without the multi‑month Salesforce implementation cycle.

Checklist for Choosing an AI Support Partner

• Resolution rate guarantee: Does the vendor commit to a specific containment percentage within 90 days?
• Integration depth: Can the AI see your CRM contacts, ticket history, and knowledge base?
• Human handoff: Is the escalation path seamless, with full conversation context passed to the agent?
• Security & compliance: For regulated industries like Smarsh’s, can the AI meet SOC 2, HIPAA, or GDPR requirements?
• Pricing transparency: Is it truly per‑conversation, or are there hidden platform fees?

Getting Started: A 90‑Day Framework for AI Support Automation

Inspired by Smarsh’s playbook, here’s a phased roadmap you can adapt to your own environment.

Days 1–30: Assess & Prepare

• Audit your most common ticket types. Aim for 20–30 categories that account for 60% of volume.
• Clean and structure your knowledge base. Fill gaps where AI would stumble.
• Pilot one specialized agent for a single, high‑volume topic (e.g., billing inquiries).

Days 31–60: Supervised Launch

• Deploy the agent in suggestion mode with human approvals required.
• Set confidence thresholds at 95% for autonomous responses; map escalation paths.
• Monitor containment rate, CSAT, and deflection daily. Retrain weekly.

Days 61–90: Scale & Expand

• Transition to autonomous mode for the initial topic when containment exceeds 50%.
• Add a second agent for a related area (e.g., onboarding help).
• Integrate with CRM and marketing automation to enable cross‑functional handoffs like Piper.

Conclusion: The AI Support Dividend Is Real

Smarsh’s journey with Agentforce isn’t an outlier, it’s the new benchmark. A 60% resolution rate, a 35% drop in escalations, and a dramatic reduction in cost per conversation signal that we’ve crossed a threshold: AI support agents aren’t just experimental; they’re CFO‑approved investments.

Your team can achieve the same. Start with one specialized agent, obsess over your knowledge base, and run the numbers on what 10,000 conversations would cost you at $2 a pop. The math is likely to be more compelling than you think.

And if you want to accelerate that journey without being locked into a single CRM, Successly offers an AI support layer that deploys in days, learns from your existing content, and delivers 43% ticket deflection from month one. The future of support isn’t about replacing humans, it’s about arming them with AI that works as hard as they do.