Revolutionizing Spam Management in Salesforce Email-to-Case with Agentforce AI

The Challenge

For organizations utilizing Salesforce’s Email-to-Case feature, an often overlooked issue is the influx of spam emails generating fictitious cases, disrupting support workflows. One of our clients, a provider of industry-specific software solutions to small and medium-sized businesses (SMB) and mid-market organizations, faced this challenge head-on. Despite relying on an advanced AI spam filter within their Microsoft 365 email system—effective for general inboxes—the filter produced daily false positives. These misclassifications were untenable for their support team, who lacked visibility into IT systems to clear quarantined emails. Initially, the client opted to bypass Microsoft 365’s filtering due to these false positives, shifting the burden to manual review, which led to increased effort, potential missed customer inquiries, and a strained case management process.

What We Did

To solve this singular yet pervasive problem, we implemented a tailored solution using Salesforce’s Agentforce, leveraging its AI capabilities to enhance spam management within the client’s ecosystem. Here’s our approach: 
  • AI-Powered Spam Detection: We configured Agentforce to analyze incoming emails’ subject lines, bodies, and header information. Using large language models (LLMs), the system assigned a spam score with transparent reasoning, offering contextual spam identification beyond traditional filters. 
  • Prompt Optimization with Feedback: To minimize false positives—critical for the support team—we used Salesforce’s Prompt Builder to iteratively tweak LLM prompts based on agent feedback from the quarantine queue. This ensured the AI adapted to the client’s email patterns, prioritizing accuracy for legitimate support requests. 
  • Quarantine Queue Workflow: Suspected spam emails were routed to a dedicated quarantine queue in Salesforce, enabling the support team to review and classify them without IT dependency, keeping the process seamless within their workflow. 
  • Feedback-Driven Refinement: Agents could mark emails as “not spam” in the quarantine queue, providing reasons for their decision. This feedback guided prompt adjustments via Prompt Builder, creating a self-improving system tailored to the client’s needs. This solution integrated with the client’s existing setup, addressing their constraint of limited IT visibility while bypassing the email provider’s initial filtering challenges.

Outcomes

The results demonstrate the profound impact of this solution, with data tracked from the go-live date of March 21, 2025: 
  • Dramatic Reduction in Manual Effort: Manual spam case handling dropped from an average of 3,850 cases per month pre-implementation to just 853 cases in the first full week post-go-live (week of March 24, 2025), and further stabilized with automation resolving 85% of spam over the next two months. This freed service agents to focus on their “real” work of supporting customers, saving an estimated 5.7 Full-Time Equivalent (FTE) hours. 
  • Robust AI Detection: Agentforce identified and quarantined 2,800 spam cases in April 2025 and 2900 in May 2025, showcasing its ability to handle the bulk of spam traffic effectively after the initial 853 cases in late March. 
  • Managed False Positives: With a deliberate 6.5% false positive rate (adjusted from <1% at a 65 spam score threshold to 6.5% at a 60 threshold to capture more spam), false positives were controlled at 197 in May 2025 and 83 in June 2025. The trained team reviewed these in the quarantine queue, averaging ~1 spam case per user monthly, a significant improvement from pre-implementation levels. 
  • Team Transformation: Of the original team of six, four agents were reassigned to solve legitimate cases, while the remaining two spend approximately 15% of their day (down from a significant portion) reviewing spam, dedicating the rest to case resolution. 
  • Client and Agent Feedback: 
    • “Yuga Shift helped us solve a frustrating problem quickly and effectively. What stood out was their ability to align the solution to our business goals and deliver outcomes our execs noticed.” – Director of Internal Systems.
    • “This has been a huge morale boost. Our agents are engaged again because they’re solving meaningful problems, not just filtering noise.” - CX Ops teams leader.
    • The client described the Case AI project as “a great success,” reflecting enhanced efficiency, reduced workload, and improved customer service quality. This progress, as shown in the monthly trend chart, illustrates a clear shift from manual to AI-driven spam management, stabilizing by June 2025.
 

The Way Forward

This case study addresses a singular yet pervasive challenge many organizations may not initially recognize: managing spam in Salesforce Email-to-Case without disrupting support operations. Moving forward, we recommend: 
  • Ongoing Monitoring: Maintain a Salesforce dashboard to track manual spam, AI-detected spam, and false positive rates, ensuring the solution adapts to evolving spam patterns. 
  • Enhanced Feedback Process: Refine the feedback loop with standardized reasons (e.g., “Known Customer,” “Technical Content”) via a custom field, potentially automating logging with Salesforce Flow to further optimize prompts. 
  • Future Microsoft 365 Integration: As a possible enhancement, re-evaluate integrating with Microsoft 365 email spam filtering to leverage headers or bypass rules, addressing initial false positive issues once data supports a layered defense approach. 
  • Broad Industry Impact: Share this success to inspire other SMB and mid-market software solution providers facing similar issues, positioning Agentforce as a versatile tool for spam management in customer support. 
  • Consult with Your CRM Team: Assess spam impact and pilot this solution to drive efficiency.

Why This Matters

Spam in Email-to-Case is a stealthy obstacle for many organizations, particularly those relying on generic email spam filtering. Our innovative use of Agentforce resolved this client’s issue, offering a replicable framework to save an estimated 5.7 FTE hours—translating to significant cost savings and improved team efficiency—and elevate customer support.