How We Generated 341 B2B Leads Across 33 Cities Using an Autonomous AI Agent

This case study documents a 6-week outbound pilot where one autonomous AI agent sourced and enriched B2B opportunities across multiple local markets. The goal was not just list-building volume. We optimized for qualified leads that a sales operator could immediately act on.

By the end of the pilot, the agent generated 341 leads across 33 cities with clear qualification signals and outreach-ready context.

Context: What We Needed to Fix#

Before this pilot, lead generation happened through a mix of manual searches, browser tabs, and spreadsheet copy/paste. The process had three recurring problems:

1. It was slow.
2. It was inconsistent between operators.
3. Quality filters were applied too late.

This meant time was spent chasing prospects that were not a fit. We needed a repeatable system that could run daily and hand over better opportunities.

Pilot Scope and Constraints#

To keep the test realistic, we constrained the workflow deliberately:

• Target segment: local service businesses in specific verticals.
• Geography: 33 cities with pre-defined priority tiers.
• Cadence: one scheduled cycle every 30 minutes.
• Output format: outreach-ready records with contact and context fields.
• Human review: quality sampling at the end of each day.

This scope prevented the common mistake of launching an over-generalized agent.

Agent Workflow Design#

The autonomous loop had five stages.

1) Prospect Discovery#

The agent queried public business sources and city-specific searches using constrained criteria. It ranked candidates by relevance before moving to enrichment.

2) Data Enrichment#

For each candidate, the agent collected:

• business name,
• location,
• website,
• service focus,
• available contact signals,
• simple qualification notes.

3) Qualification Pass#

Records were scored using explicit fit criteria (service category, activity signals, and location alignment). Low-confidence records were flagged for optional review instead of silently accepted.

4) Outreach Drafting#

For qualified prospects, the agent generated first-touch outreach drafts aligned to the value proposition. Drafts included personalized context from the enrichment stage.

5) Delivery and Reporting#

Each cycle pushed summary updates and batched leads into the operator workflow. Daily logs included counts, quality flags, and city-level distribution.

Results Summary#

After six weeks, we measured:

• 341 total leads generated
• 265 outreach drafts produced automatically
• 33 cities covered
• Median cycle duration: under 3 minutes
• Significant reduction in manual research time

The key finding was not just throughput. It was that operators could spend more time on follow-up quality and less time on first-pass data gathering.

Performance by City Tier#

We segmented cities into Tier 1, Tier 2, and Tier 3 based on market size and competitive density.

Tier 1 Markets#

Highest total output but also the most noise. Qualification rules mattered most here to avoid volume traps.

Tier 2 Markets#

Strong consistency and better fit rates. These markets produced the best balance between quality and speed.

Tier 3 Markets#

Lower volume, but often higher response quality due to less saturated outreach environments.

What Improved Quality the Most#

Three adjustments produced outsized impact.

Tightening Discovery Queries#

Early query templates were too broad. Narrowing intent language reduced irrelevant records and improved downstream draft quality.

Moving Qualification Earlier#

When qualification happened later in the loop, wasted enrichment effort increased. Moving fit checks earlier saved compute and operator time.

Standardizing Rejection Reasons#

We added explicit reason codes for disqualified leads. This made weekly tuning faster because patterns were visible instead of anecdotal.

Mistakes and Corrections#

No case study is useful without failures. These were the main ones:

1. Over-reliance on a single source in week one created blind spots.
2. Loose confidence thresholds admitted weak records.
3. Inconsistent city naming caused duplicate handling problems.

Corrections were straightforward: source diversification, stricter confidence rules, and canonical location formatting.

Team Impact#

The most practical business impact came from role clarity. The agent handled repetitive prospecting mechanics while human operators focused on:

• messaging nuance,
• account prioritization,
• relationship-building follow-up.

This reduced context switching and improved morale because the team spent less time on low-leverage admin tasks.

Cost and Efficiency Notes#

We tracked execution efficiency alongside output volume. The pilot stayed within expected operational cost bounds because the loop was constrained and quality filters were explicit.

The broader lesson: autonomous systems are economical when task scope and guardrails are clear. Unbounded loops are where costs become unpredictable.

Lessons for Teams Running Similar Pilots#

Start Narrow, Then Expand#

A narrow pilot gives cleaner feedback and faster iteration. Once quality is stable, expansion is much safer.

Measure Quality, Not Just Count#

Lead count is easy to celebrate and easy to misuse. Track acceptance rate, correction load, and response outcomes.

Keep Human Review in the Loop#

Even high-performing agents benefit from periodic quality sampling. Review protects standards and improves future prompts.

How This Connects to Your Rollout#

If you are still defining your first agent, read the broader implementation framework in The Complete Guide to AI Agents for Small Business in 2024. If your immediate goal is setup, follow this step-by-step lead generation tutorial.

You can also browse AgentTeal templates to accelerate your first deployment.

Final Takeaways#

This pilot confirmed a pattern we keep seeing:

• Autonomous agents are strongest when the task is repetitive and measurable.
• Quality improves when qualification happens early.
• Human teams perform better when repetitive research is removed from daily workload.

Generating 341 leads across 33 cities was meaningful, but the strategic win was operational consistency. The system became predictable, improvable, and easier to scale.

That is what turns AI lead generation from an experiment into a reliable growth channel.