Prepared for your SMS lead gen system — April 2026

An agent that reads the room,
not a pipeline that follows a script

The hard part isn't the SMS integration — it's the conversation. Balancing your business objective against not being pushy. That balance shifts with every reply. An agent handles that. A workflow can't.

Who I am

A quick intro before we get into the build.

Alexander Girardet

AI Engineer — Claude Code specialist

I build AI agent systems that run real business operations — lead generation, conversational outreach, CRM integrations. I've built agent teams that operate across WhatsApp and Slack, handling conversations and coordinating business operations autonomously. Same pattern, applied to your SMS lead gen.

I've built this before

An ICP discovery pipeline using the same architecture — an orchestrator coordinating specialist agents, each with custom API tools and CRM integration. Agents handle research, classification, and data sync autonomously.

The system researches companies across multiple data sources, scores them against custom criteria, and surfaces qualified leads. One agent coordinating the operation, specialists doing the work. That's what we'd build for your SMS pipeline.

Two ways to build this

Most proposals will wire up Make.com with an OpenAI call. Here's why that approach breaks down for conversational lead gen.

The workflow approach

  • ⛓︎ n8n / Make.com rigid pipeline, painful to maintain — every edge case is a new workflow branch
  • 💸 OpenAI API tokens per-call billing — costs spiral as volume grows, a variable cost you can't predict
  • 📄 Prompt template one-size-fits-all replies — no depth, no adaptation to the seller's emotional state
  • 🪛 Zapier glue between disconnected tools — brittle, breaks when anything changes

Every conversation path is a workflow branch. Seller gets emotional? New branch. Asks about price? New branch. Goes off-topic? New branch. Each new edge case is a new build. You end up maintaining the pipeline instead of closing deals.

The agent approach

  • ⚙︎ Claude Code runtime state-of-the-art agentic infrastructure, maintained by Anthropic — flexible, not rigid
  • 💰 Flat Claude Pro subscription not per-API-call LLM tokens — fixed cost regardless of volume
  • 📖 Skills + documentation in-depth instructions with edge case handling, combining rule-based logic with AI (not AI-only)
  • 🔧 CLI tools agents use tools directly instead of being trapped in a workflow — more flexible, more robust

The agent has skills, tools, and instructions — but it's free to read the nuance. A grieving probate heir gets empathy. A suspicious seller gets patience. A motivated one gets moved forward. The agent decides when to dig deeper, when to back off, when to escalate to you. Guided by guardrails, not locked in a pipeline.

Cost at scale

API tokens are like gas: costs scale linearly with usage. Claude subscriptions are flat, heavily subsidised by Anthropic. The $200/mo plan is up to 20x cheaper than API pricing at volume.

API token costs
$20/mo subscription
$200/mo subscription

The agent

One agent with dedicated skills and tools. Connects to your platforms directly. You interact with it conversationally or schedule it on a CRON job.
SMS Lead Gen Agent

Conversation Manager

Owns your SMS lead gen operation. Handles inbound replies, holds natural conversations with motivated sellers, classifies leads, syncs to your CRM, and escalates hot leads to you.

🗣 Miner/Voss dialogue 🎯 Lead classification 🔄 CRM sync 🚨 Escalation
📱 FlipForce SMS API 📊 REISift sync 🔔 Escalation alerts
↓   ↓   ↓ every message passes through three layers
Layer 01

🛡 Conversation Skill

Hard rules enforced in the skill definition: never negotiate price, never make offers, never discuss legal matters, always escalate sensitive situations. Not prompt suggestions — structural constraints the agent cannot override.

Layer 02

🔍 Evaluator Agent

A separate agent reviews every outbound message before it sends. Checks tone, rule compliance, and appropriateness. If it flags an issue, the message doesn't go out. Independent judgment, not self-review.

Layer 03

✅ Testing Framework

The conversation skill is evaluated against a test suite before deployment. Edge cases: grieving seller, price question, aggressive response, off-topic, opt-out. Validated against real scenarios, not eyeballed.

How a conversation runs

From inbound SMS to qualified lead — end to end.
00

Outreach campaign created Full Build

You tell the agent who to target — probate leads, pre-foreclosure, a specific market. The agent generates the campaign and sends initial SMS through FlipForce.

↓
01

Inbound SMS received MVP

Seller replies to your outreach through FlipForce. The agent picks it up via webhook.

↓
02

Context loaded MVP

The agent pulls the conversation history and any existing lead data. It knows what was said before and what campaign this seller came from.

↓
03

Response generated MVP

The conversation skill crafts a reply — empathetic, concise (1-3 sentences), aligned with Miner/Voss methodology. Adapts tone based on seller's emotional state and stage in the conversation.

↓
04

Evaluated before sending MVP

The evaluator agent reviews the message. Does it stay on script? Does it break any rules? Is the tone right? If flagged, it's rewritten or escalated.

↓
05

SMS sent + lead classified MVP

Response sent through FlipForce. The classifier categorizes the lead (hot, warm, cold, dead). Tags and status pushed to REISift.

↓
06

Escalation if needed MVP

Hot leads or sensitive situations trigger an alert to you. The agent hands off with a summary of the conversation and its assessment. You take it from there.

↓
07

Follow-up drip sequence Full Build

Non-responders and cold leads enter an automated nurture sequence. The agent follows up at intervals, adapting the message based on previous interactions.

What you get

Everything you need to run the system — and own it after delivery.

Conversation agent

Handles inbound SMS replies end-to-end. Miner/Voss empathetic dialogue, lead classification, CRM sync, hot lead escalation. Runs autonomously or on your command.

CLI tools

Custom-built tools for FlipForce SMS (send/receive) and REISift sync (tags, status updates). You own the integration layer — no platform lock-in, no middleware fees.

Conversation playbook

The Miner/Voss prompt engineering that drives conversation quality. Editable markdown — adjust tone, rules, escalation triggers yourself. Not a black box. You control how the agent talks.

Full documentation

Every skill, every tool, every decision — documented as AI-readable markdown. No vendor lock-in, no black box. Want to improve the system? Claude already knows everything about it. Hand it to any developer and they can extend it on day one.

2 Phase Implementation

Start with a focused MVP. Expand to the full build once it's working.
Phase 1 — MVP
$1,500
  • One campaign type (probate or pre-foreclosure)
  • Inbound SMS conversation handling
  • Lead classification (hot/warm/cold/dead)
  • REISift sync (tags + status)
  • Hot lead escalation alerts
  • Editable conversation playbook
  • CLI tools for FlipForce + REISift
  • 2-3 week delivery
Phase 2 — Full Build
$3,500
  • Everything in the POC
  • Automated campaign creation + sending
  • Multi-campaign management
  • Follow-up drip sequences
  • Dashboard — conversations, lead pipeline
  • Self-optimizing feedback loop
  • 4-6 week delivery

Want to get started?

Reply on Upwork and I'll send over a detailed scope within 24 hours.
I'd love to discuss your FlipForce setup and which list to start with.