Field notes from inside real marketing accounts. No newsletter, no guest gurus, no SEO listicles. A note when there's actually something worth saying — about AI, attribution, pricing, framework discipline, or whatever the week revealed at the desk.
Some of these are recent. Some are old enough to read as period documents — written when GPT-3 was still a research preview and "AI marketing" meant something different. Both sections are here. The archive is preserved as it was.
Filed from inside live engagements. Written when there is a pattern worth naming, a vendor archetype worth flagging, or a piece of operator math worth documenting. Read down.
Seven schema types, three deprecated models, and the operator method that took 27 listings to page one across 38 hyper-local queries. The canonical structured-data stack for real estate operations — what to ship, what to retire, and the implementation order that actually moves rankings.
Cost per lead is the metric most agencies sell and most clients buy. It is also the metric most disconnected from revenue. The dashboards optimize for it; the businesses don't grow because of it. A field note on what to measure instead — and why nobody wants to.
Most LLM agent dashboards show token counts. Token counts are not revenue. Token counts are not even cost — they are an upstream proxy for both. A note on what to instrument when an AI is the one closing the deal, and why most agencies won't ship the dashboard that shows it honestly.
Six recurring species of marketing AI that do not work, classified by appearance, mating call, and natural habitat. The Persona Generator. The Headline Spinner. The Sentiment Dashboard. The Insights Bot. The "Personalization at Scale" Engine. The Always-On Strategy Coach. Read before signing the next agency proposal.
Why the most expensive component of a marketing engagement is not the fee — it is the structure that prevents the work from ending. A note on contract design, exit clauses, and the asymmetry that turns most agency relationships into slow leaks.
Earlier writing — preserved as period documents from 2020 to 2021, before the practice was productized and before the framework was named. The voice is younger; the curiosity is the same. Some predictions held. Some did not. Both are useful.
An early note on AI applied to marketing — written when "AI tools" still meant AWS, IBM Watson, Facebook Lookalike Audiences, and the freshly launched Google BERT update. GPT-3 was eight months from public preview. The piece looks naive in retrospect; the underlying argument held: lower compute costs would compound an enabling environment, and marketing would be one of the first verticals reshaped.
Six emerging professions for the new decade — written three weeks before the world locked down. Photovoltaic installers, big data specialists, SEO specialists, programmers, virtual reality developers, business intelligence analysts. Some of those bets compounded; others did not survive the decade. A useful artifact of how the future looked from late February 2020.
A family-history platform built on AI-driven reconstruction, anthropology, and digital preservation. The Manrara-Gastón family, life in Cuba before the Revolution, exile to the United States — documented through Carmelina Gastón's recollections. The documentary work that informs how the marketing operator thinks about narrative, attention, and what makes a story worth holding.
Notes on distribution, audience-building, and the economics of selling long-form non-fiction in a streaming-saturated market. Written from inside the production of two documentaries (Embargoed, Carmelina). The lessons translate directly to high-ticket service marketing more than most marketers would admit.
A working note on visual coherence across web, print, social, and broadcast — where brand voice meets brand surface. Written before Ferminius productized its practice; preserved here because the underlying observation about consistency under format pressure still holds.
No email capture. New entries are published when there's something to say, not on a calendar. The most reliable way to see them is to bookmark this page or follow the operator on LinkedIn.
If you'd rather skip the reading and start the work, the diagnostic takes thirty minutes and ends with a clearer map than most newsletters deliver in a year.
1 seat remaining · Q2 2026
Thirty minutes, 72 hours, no PDF deck — an honest read of where your current marketing leaks money, and whether the unit economics of your business support a self-paying engagement at all.