Success Story · CRM Migration
From Chaos to Fully Automated — In Under 4 Weeks
A general contracting company in Texas was managing leads manually across multiple platforms — contacts scattered, follow-ups missed, and their team had no unified system to track jobs from inquiry to close. They came to me with Follow Up Boss as their CRM and a growing team that needed something better.
What we built together:
n8n Data Automation
Automated Historical Data Pipeline & Knowledge Base
A home services contracting business had years of pricing history locked inside Estimate Rocket — a CRM/estimating platform with no public API and no export option. Every past proposal existed only as an individually rendered PDF, so there was no way to search, filter, or compare what had been charged for similar work before. Rather than opening over a thousand client records by hand, I built an automated pipeline in n8n to handle it end to end.
Phase 1 — the scraper workflow: client sheet → project pages → proposal PDFs, downloaded straight into Google Drive.
Phase 2 — the parsing workflow: reading each downloaded PDF, extracting line items, and writing structured rows into the knowledge base.
The approach
With no official API to work from, the platform's data structure had to be mapped by hand before any automation could run:
Client → Project → Proposal
Mapped the relationship between client records, projects, and saved proposals by inspecting the app's rendered HTML and network requests.
Two authentication paths, not one
- Found a public, unauthenticated customer-portal URL pattern that served the actual proposal PDFs directly — no need to scrape behind a login for document retrieval.
- Authenticated the read-side scraping (client/project pages) using a captured browser session token, since no API keys or OAuth flow existed.
3,325 PDFs, downloaded automatically
Automated traversal from client → project → proposal, pulling every proposal PDF into Google Drive without a single manual download.
Custom regex parser for unstructured text
PDF text extraction doesn't preserve the original table layout, so a custom JavaScript parser was built to pull each line item — description, unit price, quantity, and total — directly out of the raw extracted text.
Batches that fail safely and re-run
Processing ran in chunks of 50 records to stay within the automation platform's execution and memory limits, using an idempotent "append-or-update" write pattern so any batch could be safely re-run without creating duplicates.
Key challenges solved
No public API
Built the entire pipeline against a platform never designed for programmatic access, using session-based requests and careful mapping of internal ID relationships.
Sandboxed execution environment
The automation platform didn't support common HTML-parsing libraries, so hand-written regex parsers stood in as a substitute.
Execution limits at scale
Full-dataset runs initially failed on memory and timeout limits, so the pipeline was redesigned around small, resumable batches instead.
Data integrity at scale
Caught and fixed a parsing bug where quantities were misread from adjacent dollar amounts — an error that would have silently corrupted a large share of the dataset if it had gone unnoticed.
What we built: