Fletcher Keeley

Theories — No. 02

Working frameworks

Not predictions. Arguments I keep testing against real work, plus data reports drawn from my own infrastructure and from years of operating work. Each one earned by building, not borrowed.

theory No. 01

How AI actually reshapes work

The cost of load-bearing writing is collapsing, so the bottleneck moves to operationalizing it. Software goes from metal to clay, businesses can finally own how they operate, and adoption climbs four distinct levels: process, the mech-suited individual, the connected organization, and system management.

metal → claythe mech suitfour levels8 codified patterns
→ read
theory No. 02

The Retreat: the economics of attention

Digital ads have never cost more, or been worth less. Every impression decomposes into volume, attention, density, and legitimacy; fourteen channels parameterized over a decade show volume masking substrate decay everywhere the substrate can be manufactured. Published as a live interactive model.

EI = I × A × D × L14 channelsprojected to 2040live model
→ read
theory No. 03

How I model a consumer business

Every consumer business is two populations: customers you have and customers you haven't met yet. Model the base from cohort behavior first, buy the gap at a price the margin can absorb, give every channel three numbers, respect the CAC ceiling and the owned-channel band, and deliver the model through a daily cadence. Nearly six years of running one, distilled.

two populationsthe CAC ceilingoptimal bandbase health
→ read
report No. 01

Five months inside 226 brand inboxes

A DTC email panel snapshot from my own pipeline: 6,120 marketing emails, Feb–Jul 2026. Two-thirds carry an offer, half the discounts run 40% or deeper, weak offers shout the loudest, and the top tenth of senders own half the inbox. A bounded dataset, reported as exactly that.

6,120 emails226 brandsown pipelinehonest caveats
→ read
report No. 02

Anatomy of a funnel contraction

Five years inside one DTC P&L, indexed. Flow email revenue fell 60% and it looked like a strategy failure; reconciling order, spend, session, and message-level data showed a structural CAC wall touched three times, a forced spend pullback, and every downstream channel starving in sequence while every efficiency metric improved.

20 quartersthe CAC wall ×3spend −56% → sends −76%indexed data
→ read
report No. 03

How I run an agency with AI

A field report on running a boutique agency with AI as named staff. The AI sees the email, calendar, drive, and task list, touches every channel, and works a night shift on a vetted library of playbooks and skills — PRs, blogs, call agendas, and campaign launches built overnight, approved by a human in the morning. Five clients, one operator, two shifts.

5 clients · 1 operator10 playbooks + 8 skills~$4/overnight taskCassi Claw
→ read