Case Study

How We Helped a B2B Growth Agency Scale Without Adding Headcount and Cut Costs by 80%

Client identity withheld under signed NDA.

80%+
Cost drop: from ~$9,000–$10,000 to under $2,000/month
🤖
Full AI
Outbound team: LinkedIn and Email fully automated
2–4 wks
Live and delivering results within the first month

"Every time we scaled, I kept waiting for the costs to catch up. They never did. That level of scalability, without the cost growing with it, is honestly insane."

Head of Growth, Confidential B2B Agency (name withheld per NDA)
Industry
B2B Lead Generation Agency
Teams
LinkedIn outreach, cold email, Meta ads, sales
Goal
Scale outbound volume without scaling cost or headcount
Stack Before
Claude via MCP, third-party email tools

The agency was already using Claude (the standard desktop app, not a custom integration) across their teams to handle outbound tasks: LinkedIn messages, connection requests, comment engagement, email sequencing. It worked as a starting point. But as they pushed toward scale, three problems compounded fast.

Cost was ballooning across the entire operation. Across all platforms (LinkedIn outreach, cold email, Meta ads, and the full supporting stack) they were spending $9,000 to $10,000 per month to process around 300 leads per day. Their third-party email tool alone was costing significantly more than necessary. The total was unsustainable at the volume they needed to reach.
Speed was a bottleneck. Claude handles complex reasoning well, but it wasn't built for high-throughput, repetitive outbound tasks at volume. Latency piled up across hundreds of simultaneous actions.
Every new hire reset the clock. LinkedIn specialists, email operators, each required full retraining. The playbook lived in people, not systems. Growth meant more training cycles, not more output.

They needed to replace a patchwork of human judgment and paid AI tools with something engineered, cost-controlled, and self-sustaining.

Phase 1
Map the Decision Logic

The first step was auditing every outbound action across all four teams and separating what actually required intelligence from what was repeatable and rule-based. The finding: over 80% of outbound actions are deterministic. Only a small subset (ICP qualification, negotiation, personalised writing) need frontier-level reasoning.

Phase 2
Build: Local AI + Selective Claude Routing

We deployed a local AI orchestrator on the client's own machine: their infrastructure, their control, near-zero marginal cost per inference. The architecture works like this:

Local model (Ollama-based)
Fine-tuned on the client's playbooks, ICP data, and voice. Handles orchestration, deciding what action fires, when, and to whom, at essentially zero cost.
Action modules in code
Every repeatable action (connection requests, comments, follow-up sequences) is executed by dedicated code, not AI. The AI decides; code acts.
Claude routed by task type
When a task genuinely needs high reasoning, the system routes to the right model automatically: Claude for personalised writing and negotiation, Grok for ICP identification and scoring. Routine actions never touch a paid API.
Email infrastructure rebuilt in-house
The agency's third-party email tool was replaced with equivalent infrastructure running on their own stack, same capability, no per-send fees.
Phase 3
Connect and Deploy

The system was integrated across all four teams. LinkedIn and email sequences now run in coordinated cadence: if a prospect is warm on LinkedIn, email timing adjusts automatically. When a lead crosses a warmth threshold (replies, engagement signals), a call task is triggered for the sales rep. One operator initiates the process. The AI runs it.

LinkedIn connection targeting and sending
Comment engagement on ICP accounts
Personalised DM sequences through follow-up
Cold email sending and follow-up cadences
Cross-channel lead tracking and coordination
ICP scoring and lead qualification
Warm prospect flagging and sales handoff
AI model routing by task complexity
Full email infrastructure management
Metric Before After
Total monthly platform cost
~$9,000–$10,000
Under $2,000
Claude dependency
100% of actions
~10–15% only
Operators needed
Multiple specialists
1 to initiate
New hire training
Required every time
System retains knowledge
Time to results
N/A
2–4 weeks

The biggest cost mistake in AI outbound is defaulting every action to the most capable model. Most outbound actions don't need frontier intelligence; they need consistency, speed, and correct execution. Route expensive AI only to the moments that move the needle: personalisation, qualification, negotiation. Run everything else on a local model and code. The result is a system that is simultaneously smarter, faster, and a fraction of the cost.

With the system live and costs stabilised, the agency is expanding automation to Meta ad comment engagement, deeper CRM integration for the sales team, and improving the local model's ICP recognition across new verticals. The foundation is built. The ceiling is gone.