Detecting Spam with Agentforce

Automatically Filtering Spam Website Leads with Agentforce

Inbound website leads are a double-edged sword. On one hand, they’re a crucial part of your sales funnel. On the other, they can attract a flood of spam, bots, and low-quality submissions that eat up your sales team’s time.

At BigSolve, we recently tackled this exact challenge for a client—and came away with a robust, AI-powered solution that stops spam at the door, letting only legitimate leads flow to the sales team. Here’s how we did it (and how you can, too).

‍

The Problem: Website Spam in the Sales Funnel

Anyone with a “Contact Us” or “Get a Quote” form knows the pain: not all website leads are created equal. It’s not uncommon for a significant percentage of inbound leads to be:

• Bots filling out forms with gibberish names and fake companies
• Competitors fishing for information
• Automated SEO spam
• Vendors pitching unrelated services
• Even just the “test” submissions that sneak in from internal or partner teams

Every one of these clogs your pipeline and wastes valuable sales hours.

‍

The Solution: Agentforce + Flow = AI-Powered Spam Filtering

Our approach combines the flexibility of Salesforce Flow with the intelligence of Agentforce’s Large Language Models (LLMs).

‍

Here’s the high-level architecture:

1. When a Lead is created with “Website” as the Lead Source, a Flow is triggered.
2. Agentforce LLM analyzes the Lead’s details (name, company, title, phone, description) and returns a simple verdict: "Spam" or "Legitimate".
3. If “Spam” is detected, the Flow automatically updates the Lead Status to “Unqualified” and logs the reason as “Spam.”
4. Legitimate leads pass through untouched—no human review required.

‍

Why This Works

• No more wasted sales hours—your reps only see leads worth pursuing.
• No manual review queues—the system acts in real time, 24/7.
• Tunable & explainable—you control the Flow logic, and can review “Spam” leads if needed.
• Easily extensible—expand the logic to handle new spam patterns, or apply it to other objects (Cases, Contacts, etc.).

‍

How We Built It (Technical Deep-Dive)

1. Prompt Template:
We designed a prompt for Agentforce with this specific task:

‍

🧑 Role: You are a spam detection assistant for {!$Organization.Name}, responsible for reviewing Leads and identifying spam submissions before they reach the sales team.

Lead Details:
• {!$RecordSnapshot:Lead.Snapshot}

---

📌 Instructions: Review the details above and determine if this Lead is likely spam or legitimate. 
• Spam leads often have gibberish or obviously fake names, nonsense or advertising in the Message, missing or dummy data, or other patterns typical of bots or bad actors.
• If you are unsure, err on the side of “Legitimate” (do not be overly strict).

---

⚠️ Constraints & Formatting:
• Respond with only one word:
 - “Spam” if the Lead is likely spam 
 - “Legitimate” if the Lead is likely legitimate
• Do not include any other words, commentary, or formatting.
• If there is not enough information to decide, respond with “Legitimate”.

‍

‍

2. Record-Triggered Flow:

• Trigger: When a new Website Lead is created
• Agentforce Action: Pass the Lead to the “Detect Spam” prompt
• Decision Element:
  • If the response is "Spam", set Status = Unqualified and Reason = Spam
  • If "Legitimate", do nothing‍
• Update Lead: Automatically update the Lead’s Status and Unqualified Reason if it’s spam—no manual work needed.

‍

Results & Business Value

What happened after launch?

• Spam submissions dropped out of the sales team’s view instantly
• The team’s “first call response time” to real prospects improved noticeably
• No more wasted cycles chasing obviously fake or junk leads