Revenue AI Strategy
Open the platform
All posts
L4 Maturityoutbound 6 min read

Scaling Outbound: Building an L4 Relevance AI BDR Agent

Learn how to build an L4 Relevance AI BDR agent that researches, personalizes, and handles replies to drive down meeting costs and scale outbound.

Run the playbook

Relevance AI BDR agent (L3-4)

Multi-step agent: research → personalize → email → handle reply → book meeting. Real, not magic; needs constant tuning.

Why this matters

The "spray and pray" era of outbound is dead, yet most companies are still trying to solve the problem by hiring more BDRs to send more volume. This approach is failing because prospects have a 6th sense for generic automation. The alternative—high-touch, manual research—is effective but unscalable.

At the L4 maturity level, we move beyond simple sequence automation into Autonomous Research and Action. Using a Relevance AI BDR agent isn't about sending 10,000 emails; it’s about having an "always-on" assistant that researches every lead with the depth of a human but the speed of a machine.

The cost of doing nothing is a rising Customer Acquisition Cost (CAC) and a burned-out BDR team. A properly tuned Relevance AI agent can reduce your Cost Per Meeting by 20-40% while allowing your human BDRs to focus solely on high-value conversations and closing.

How it works

1. Define an ultra-narrow ICP segment

Before you touch any AI tool, you must narrow your focus. If you tell an AI to target "SaaS companies," it will hallucinate generic value propositions. You need a segment so specific that the "pain point" is identical across every lead.

  • The Goal: "Series B FinTechs in the UK using Salesforce who recently hired a head of revenue."
  • The Work: Spend 3-4 hours in LinkedIn Sales Navigator or Apollo.io. Create a "Gold Standard" list of 20 leads.
  • Why it matters: If a human can’t explain the nuance of the segment in 10 seconds, the AI will fail at personalizing the outreach.

2. Build the Research & Draft Agent

In Relevance AI, you aren't just writing a prompt; you are building a workflow. Create an "Agent" and add a "Search Web" step to scrape the lead’s LinkedIn and latest company news.

  • Logic: Use GPT-4o or Claude 3.5 Sonnet as the engine. Prompt it to identify three specific challenges based on the scraped data.
  • The Rule: High constraints. Tell the agent: "Do not mention their education. Do not use the word 'congratulations.' Focus on their specific move from [Competitor X] to [Target Company]."
  • Output: A 3-sentence "personalization snippet" that sounds like a peer-to-peer insight.

3. Connect the Delivery Stack

Relevance AI shouldn't send the email directly. You need a deliverability layer. Integrate Relevance with Smartlead or Instantly via API.

  • The Setup: Map the agent’s output to a custom variable like {{ai_research_intro}}.
  • Safety First: Use secondary domains (e.g., name@trycompany.com) to protect your main domain. For L4 maturity, ensure your email accounts have been "warmed up" for at least 14 days before the agent goes live.

4. Implement Human-in-the-Loop (HITL)

This is where most RevOps teams fail. They turn the agent on and walk away. For the first 100 leads, you must enable the "Approval" toggle.

  • Refine, Don't Just Fix: If the agent makes a mistake (e.g., mixing up a CEO’s name), don't just edit the email. Go back into the Relevance AI instructions and add a negative constraint to the system prompt to ensure it never happens again.
  • Time Commitment: A BDR Lead should spend 30-60 minutes a day reviewing the "Draft" queue.

5. Deploy Reply Handling

The agent's job doesn't end at the "send." Configure the Relevance "Reply Handling" template to monitor the inbox.

  • Categorization: The agent should bucket replies into Interested, Objective, or Not Interested.
  • The Hand-off: For "Interested" leads, trigger a Slack notification via Momentum.io or a similar tool. This is when the human steps in. Do not let the AI book the meeting directly; it’s too brittle. Have the BDR send a personal Calendly link once the interest is confirmed.

Tools you need

  • Relevance AI: The core orchestration engine.
  • Apollo / Sales Navigator: For lead sourcing.
  • Smartlead / Instantly: For email deliverability and domain rotation.
  • Clay: (Optional) To enrich data before it hits Relevance AI.
  • Slack: For real-time "Interested" alerts.

KPIs to track

  • Meetings per Agent per Week: Expect 2-4 high-quality meetings per agent once tuned.
  • Cost Per Meeting: (Software Subscriptions + LLM Tokens + Human Audit Time) / Meetings Booked.
  • Positive Reply Rate: Aim for >3% in niche segments.
  • Human-to-AI Correction Ratio: How many drafts require manual edits? (Goal: <5%).

Common pitfalls

  • The Broad ICP Trap: Trying to sell to "all Marketing Managers." The AI will default to "I saw your profile and was impressed." This is an instant "delete" for prospects.
  • Set and Forget: Thinking the agent is a finished product. It’s a "living" employee that needs weekly prompt engineering.
  • Ignoring Deliverability: Sending 200 emails a day from a new domain without a warmup. You’ll be in spam within 48 hours.

When to graduate to the next level

Once your Relevance AI agent is booking meetings at a lower Cost Per Meeting than a human BDR, and your "Correction Ratio" is under 5%, you are ready for L5 Maturity. This involves multi-channel agents that can handle LinkedIn voice notes, personalized video generation (via HeyGen), and autonomous CRM qualification without any human eyes on the drafts.

relevancebdragentoutbound

Ready to ship it? Open the playbook

Relevance AI BDR agent (L3-4)

Step-by-step instructions, the tools to use, and the KPIs to watch — already wired into the Revenue AI Strategy workspace.

Open playbook