AI Sales Audit System
Automated Lead Analysis & Claude-Powered Diagnostic Pipeline
Built an autonomous end-to-end data pipeline that ingests raw lead prospects, conducts deep web enrichment, prompts Claude API with structured diagnostic prompts, and generates executive sales audit reports.
The Problem & Friction
B2B sales teams were spending hours manually researching prospective clients, scraping company profiles, and writing custom audit observations before discovery calls.
- ✕2 to 3 hours spent per lead researching websites and drafting personalized audit points.
- ✕Inconsistent discovery call preparation leading to lower qualification rates.
- ✕Manual copy-pasting between CSV databases, Google Docs, and internal email draft queues.
Strategy & Approach
Constructed an automated webhook-triggered pipeline using n8n/Make orchestration, modular prompt chains via the Claude API, and automated PDF report generation.
- Structured Extraction Prompts: Implemented XML-tagged prompts ensuring Claude delivers strictly structured JSON observations.
- Rate-Limiting & Retry Architecture: Protected against API limits with automatic exponential backoff queues.
- Modular Multi-Pass Analysis: Separated website teardown from strategic recommendations for higher analysis quality.
Architecture & Data Flow
Autonomous automation webhook flow orchestrating data ingest, LLM parsing, document rendering, and notification dispatch.
Webhook Lead Ingest
New row or form submission triggers webhook containing company URL and target contact.
Automated Data Scraping
Pipeline extracts meta tags, headlines, tech stack signatures, and company positioning.
Claude Reasoning Engine
Evaluates weaknesses, missed conversion opportunities, and crafts tailor-made recommendations.
Document Synthesis & Alert
Compiles formatted PDF audit dossier and pushes direct link to sales team Slack channel.
Technology Stack Breakdown
Verifiable Outcomes & Metrics
Transformed a 2-hour manual prospect research routine into a 90-second autonomous pipeline running 24/7.
Engineering Lessons & Takeaways
- ✦LLM prompts must enforce strict JSON schemas to avoid parsing errors in downstream PDF templates.
- ✦Separating web scraping from LLM reasoning prevents timeouts and makes debugging failure steps instant.