Case study · current role

Building an AI-run go-to-market engine as a one-person marketing team

I'm VP of Marketing and the only marketer at a B2B SaaS company. Since April 2026 I've built a "GTM brain": a marketing data warehouse, an analysis layer that writes a weekly insights briefing, and an automated content and paid-search engine.

6.5months, one marketer
500+scoped AI-agent tasks
875merged pull requests, 1,750+ commits
39automated workflows, 29 on a schedule
~14,800automated tests
158blog posts managed in a Git-based site
~40technical articles created, plus multiple content refreshes
252pages covered by 7 internal-linking rounds

How the work is split

I set strategy, positioning, spend and final approval. AI agents (Claude Code, run in parallel by an orchestrating "dispatcher" agent) do the engineering, analysis and drafting. Buyers' research has been moving from search engines into AI assistants, so the system tracks both.

What I built

  • A marketing data warehouse in BigQuery, with scheduled, hands-off loading from 15+ sources: GA4, Search Console, Google Ads, the CRM, forms, product analytics, rank tracking and keyword data, competitor monitoring and AI-answer visibility trackers.
  • A weekly insights briefing, generated automatically and published to Notion for the leadership team every Monday.
  • A content engine: a weekly schedule that ranks about 950 potential articles from about 2,100 tracked keywords; research briefs, drafts, quality and risk checks, then publishing to the website through reviewed changes; regular internal-linking rounds.
  • A rebuilt paid-search account on value-based bidding, changed only through tested tooling that checks every change before applying it, with a one-step rollback and daily guardrail checks.
  • Consent-aware measurement: server-side tagging, first- and last-touch attribution passed to the CRM, de-duplicated conversions, and audits of what consent rules cost in lost data.
  • AI-visibility tracking: about 150 prompts monitored across leading AI assistants, my own Claude-based tracker, and analysis that separates machine-written searches from real people's.

How I worked

  • Every change is small, reviewed, tested and reversible. Nothing that touches spend, the public website or customers goes live without my approval.
  • Every AI and data call is metered for cost. The system was designed around a running cost of about $1,000 a month.
  • Decisions are logged with their reasoning, and a log of site changes records anything that affects how the data should be read.

Want one of these for your team?

A few lines is enough: what you want to grow, and what's in the way. I'll reply by email.

Get in touch