The framework. 3 layers. In order.
Every engagement moves through Control, Flow, and Production in order, because AI belongs on Layer 3, and Layer 3 only holds once Layers 1 and 2 are real.
Control
This is the non-negotiable base: who owns which outcome, which numbers are real, and where decisions get made. Without it, automation only amplifies the chaos it runs on.
At a European energy utility, I joined a stalling enterprise program as PMO cover and then took the lead through a critical phase: I rebuilt the plan, consolidated the cross-country replanning, and made ownership clear across functions, so that escalations came down and decisions started landing again.
Flow
A cadence that fits the work, 1 leading tool per purpose, and the right people in the right meetings.
On the global consumer-goods rollout across 40+ countries, once the flow was right, cycle time per country collapsed from 12 weeks to 4. This layer is often where 80% of the compounding value actually shows up.
Production
AI and Python inside real workflows, picked for payback: cloud LLM APIs for document processing, decision support, status roll-ups, and the glue between systems.
At a European retail group, I'm currently steering an AI marketing platform from PoC toward governed production: brand-compliant content generation inside real guardrails, with quality gates and 1 owner per outcome, across marketing, tech, and AI. The mechanics only work once Layers 1 and 2 are real.
PoC vs. production: where scale-ups get stuck.
If most of the left column describes your AI program, then what you have is not a model problem but a delivery problem.
| Dimension | What most AI programs do | What production needs |
|---|---|---|
| Ownership | Shared between product, ops, data, so nobody owns | One named owner per workflow, accountable to a metric |
| Scope | Cool demo on a clean dataset | Production workflow with real edge cases |
| Governance | Steering slides every 6 weeks | Weekly operating review with decisions logged |
| Integration | Chat UI in a tab | Called from the tools the team already uses |
| Measurement | Qualitative, vibes-based | Baseline, target, actual, tracked weekly |
| Handover | "The vendor keeps it running" | Internal product owner, documented, reusable |
scroll →
1 engagement. 3 phases.
Scope and exit criteria are agreed before each phase starts, so there's no open-ended retainer creep.
Diagnose
Establish the real status quo: where the program is, where AI is stuck, what has to change before anything else matters.
Deliverables
- Status quo report, board-ready
- AI portfolio scorecard
- Backbone gap list
- Phase 2 scope proposal
Exit criteria
- Exec sponsor signs off on the picture
- 1-3 AI workflows selected for Phase 2
- Go / no-go within one week
Install
Put the backbone in and ship the first production AI workflow end-to-end, proving the model on a single case before anything scales.
Deliverables
- Operating backbone: governance, cadence, decision RACI
- Production AI workflow with measurable outcome
- Measurement dashboard with first 4-8 weeks of data
- Phase 3 rollout plan
Exit criteria
- One AI workflow live in production for 4+ weeks
- Weekly operating review running without me
- Internal owner runs the workflow solo
Ship
Scale from that first production workflow to the next 2 or 3, and then hand over cleanly.
Deliverables
- 2-3 additional AI workflows in production
- Internal "how we put AI in a workflow here" playbook
- Handover to a named internal leader
Exit criteria
- Internal team owns and extends the work
- P&L-relevant metric visible to the exec team
- Engagement closed cleanly, no dependency tail
Is this a fit?
An easy no for the wrong buyer. An easy yes for the right one.
Green flag: call me
- Board or CEO is asking where AI leverage actually is, and the honest answer is "in pilots"
- You have more than 2 AI initiatives running and none is reliably in production
- Ops grew fast and the backbone (ownership, cadence, tooling) hasn't caught up
- You need someone who can sit with engineering and brief the board in the same day
- Budget exists, sponsor is named, scope is shapeable
Red flag: not a fit
- You want a PoC partner to demo something flashy
- You don't have at least one workflow worth putting AI inside
- You already have a crisp operating model and a working PMO
- You want a deck writer or a pure strategy deliverable
- No budget, no sponsor, or scope is politically frozen
Common questions.
"We don't have budget for outside help right now."
That's a fair question, but the real one isn't my fee; it's what another quarter of AI-stuck-in-pilot costs you. If your board is asking where the leverage is and you don't have a 90-day answer, that's the budget conversation worth having, and I'm happy to send a one-pager that makes exactly that case to your CFO.
"We already have a PMO / Chief of Staff function."
Good, because I don't replace them; I install the missing piece. Usually that's the AI-specific operating cadence and the workflow integration that an internal PMO isn't set up to do, and Phase 1 tells us within three weeks whether I'm adding value or not.
"Why should we trust you on AI when your case studies are SAP and ITSM?"
Straight answer: the AI layer is the differentiator, and the enterprise delivery track record is where the discipline behind it was built. Most AI programs at your scale die on the Layer 1 and 2 problems I've been solving for years, and while the AI work on Layer 3 is genuinely hard, it's rarely where these programs actually die. So if you want a pure AI research partner, I'm the wrong call; if you want AI to actually reach production, I'm the right one.
"We want to try this in-house first."
That makes sense. If it's working in 90 days you won't need me, and if it isn't, you'll know exactly what's missing, which is a conversation I'd far rather have than sell you something premature. Want me to send the self-diagnostic checklist we'd use in Phase 1, so you can run it yourselves?
"We'd want to start with a smaller pilot."
Phase 1 is the smaller pilot. 3 weeks, fixed fee, clear deliverable, no commitment to Phase 2 until you've read the output. That's the smallest useful slice I can carve.
"Can we talk to a reference first?"
Yes, after the intake call, because I want us both to know what you actually need before I spend a reference's time on it.
"The sponsor wants to see a proposal."
I'll send a one-page scope note within 48 hours of our call, and that is the proposal: short on purpose, because executives don't read 40-page decks, and I don't write them.
Recognized your situation?
If you read all of this and saw your own program in it, the intake call is the next step.
Book the intake call →