AI enablement
AI enablement, run as a P&L.
A pilot stalls when the spend has no owner and no number. I run enablement as a small P&L instead: find the expensive repetitive work, gate every project on a number, tie every dollar to the initiative that owns it, and build standards instead of buying seats. Hands-on, from the engineering to the adoption.
01 / The operating model
Four moves that fund the next one
The whole thing follows from one decision: the team that puts AI to work carries a budget and has to show a return against it. Run right, it is a loop, where each shipped win pays for the next one.
Find the work worth automating
Find the expensive, repetitive work the way a founder does discovery, and prove the value with the cheapest experiment before building a platform.
Read the write-up 02Take only projects with a number
A project does not start until someone can name what it will move: hours saved, cost removed, pipeline added, or risk reduced. The gate kills toys early.
Read the write-up 03Attribute every dollar, from day one
Each initiative gets its own metered credential up front, so the provider rolls spend up by initiative and every dollar carries the label of the thing that spent it.
Read the write-up 04Build standards, do not ration seats
Encode how the team builds so each project reuses the last, and control model cost through routing and caching instead of cancelling licenses.
Read the write-up02 / Worked examples
What this looks like on real work
Details changed enough to protect the client, true to the shape. Each one names a number and ties the engineering to it.
A marketing ops cleanup and migration
A marketing team with 200+ ungoverned email templates across two tools, dreading a manual migration. Run as an AI-driven pipeline it goes 80 to 90% automated, with a human keeping the judgment. The cleanup is the first win, before the migration even starts.
A support agent into production
A vendor support agent for a customer-facing workload in a regulated industry. The work was the evals, a calibrated judge, and adversarial tests before turning it on, then measuring it after. That is where the value sat.
Cutting a coding agent’s token bill
A code-knowledge graph so an agent stops rediscovering the codebase, proven with a one-day test under $5 before a $63-a-month build, plus model routing and prompt caching. The bill comes down through engineering.
03 / Put a number on it
The estimate I start every project from
Every project I take on has to name its number. Move the inputs and the P&L reads out live: what a repetitive task costs when done manually, and what enablement returns.
A back-of-the-envelope, not a quote. How I attribute the spend →
04 / The first 90 days
How an engagement runs
Buying the setup and the first measured wins, so the P&L is real before anyone talks about a permanent hire.
Find and instrument
- Sit with the teams whose work is expensive and repetitive
- Set up a metered credential per initiative, so spend is attributed from the first token
- Name the number each candidate project would move
Ship the first win
- Run the cheapest proof on the top candidate, and keep a value journal
- Build the first automation for real against live data
- Book the saving to the department that felt the pain
Prove and standardise
- Report cost per outcome, per initiative, in one view finance can read
- Put the standards into the tooling so the next project starts from the last
- Queue the next projects on the same number gate
05 / The playbook
Read the thinking behind all of it
A field guide to adopting AI in engineering
How a 200-person org adopts AI: baseline the work, pilot on real code, measure outcomes, guardrail the agents, and route every model call through one gateway. Written three weeks into a live rollout.
Run AI enablement as a P&L
The full operating model: why the committee-plus-licenses setup cannot pay off, and the four moves that do.
How to tell which AI projects are worth building
Find the pain, gate on a number, and prove value with a one-day experiment before you spend on a platform.
How to attribute AI spend to the initiative that owns it
Give each initiative its own metered credential up front, and the spend attributes itself.
Why standards beat handing out AI licenses
Rationing seats caps your upside to save a fraction of your downside. Control cost through standards and engineering.
An AI enablement case in marketing ops
The marketing ops cleanup and migration, stage by stage, with an interactive walkthrough of the pipeline.
Let’s put a number on your most expensive workflow
If you are standing up an AI enablement function, or you have one and it feels like a backlog of ideas rather than a set of shipped, measured wins, I am happy to get into the specifics of your setup.