Field Notes

Notes from real AI-automation builds — methodology, costs, and failure modes. No hype.

What a GA4 Audit Actually Covers

The checks I run on a GA4 property before trusting a number in it — event provenance, purchase integrity, the server-side path, Meta CAPI mirrors — and the three failures that never surface as errors.

Stripe Payment Links Don't Do Carts. Here's a Cart Anyway.

Payment Links are the fastest way to take money and the worst way to sell two things at once — the thin endpoint that fixes it, and the three bugs that fail silently.

Wiring an AI Agent to Your CRM Without Leaking Customer Data

The leak paths are boring — logs, prompts, and analytics events, not hackers. The server-side upsert pattern that gets purchase data into HubSpot with the minimum possible surface area.

Automating Weekly Reporting with Claude and Google Sheets

The hard part of an automated report isn't computing the numbers — it's making them trustworthy. The ordering that keeps a weekly report auditable.

When Your Checkout Isn't Yours: Server-Side Tracking That Survives It

How a live-events business recovered purchase visibility across a third-party checkout — Stape-hosted server-side GTM, Enhanced Conversions, and a Stripe→GA4 fallback route that logs before it sends.

n8n vs Make vs Custom Python: Choosing Without Regret

The four-question rubric I use to pick an automation stack before writing a line of code — volume math at production scale, branching complexity, failure cost, and who maintains it in month six.

What an AI-Automation Audit Actually Looks Like (Checklist Included)

The fixed-scope diagnostic to run before writing automation code — rubric + three failure modes.