Your CTO
Is keeping the platform up, the roadmap moving, and the team hired. AI is a tenth priority competing with nine things that break loudly if ignored.
Michael Kilty · Head of AI
AI needs one owner.
I take the seat.
I work as Head of AI — the executive between the business and the technology. I own the strategy, the process redesign, the governance, and the vendor calls, and I answer for the number they all have to move. Typically two days a week, on a three-month initial term.
87 days
from fragmented pharmacy operations to a live, regulated AI platform — in the Head of AI seat at an Athens compounding pharmacy.
~95%of GenAI pilots never reach productionMIT
76%of organisations now have a Chief AI Officer — up from 26%IBM 2026
78%of leaders could not pass an AI governance auditGrant Thornton 2026
~95%of GenAI pilots never reach productionMIT
76%of organisations now have a Chief AI Officer — up from 26%IBM 2026
78%of leaders could not pass an AI governance auditGrant Thornton 2026
The questions that keep CEOs up
If any of these sound like you, you don’t have an AI problem. You have an ownership problem.
These are the sentences I hear in first calls. If more than two of them land, the conversation is worth having.
“We’re paying for eleven AI tools. I couldn’t tell the board what a single one of them returns.”
My jobI connect every tool to a P&L line — and cancel the ones that don’t earn their seat.
“Everyone on my leadership team has an AI pilot. Nobody has a number.”
My jobPilots get an owner, a budget, and a metric. The ones that can’t carry all three get stopped.
“My engineers say it works. My lawyers ask who approved it. I don’t have an answer.”
My jobThe approval and audit trail gets built into delivery, so the answer is a document rather than a shrug.
“The board asks about our AI strategy every quarter. I change the subject every quarter.”
My jobYou get a strategy you can defend in one slide — because every line on it is already moving.
“Our competitors announce AI agents. I don’t even know what we’d let one touch.”
My jobWe draw the autonomy lines together: what agents may do alone, what needs a human, what never ships.
“I know AI matters. I just don’t have anyone who can hold both the tech and the business end of it.”
My jobThat is the entire job description. Both rooms, one person, accountable.
01
01Why the seat is empty
AI ownership is homeless.
Not because nobody is capable. Because in most companies the job falls between four people who each have a good reason not to take it.
So it goes to whoever is most enthusiastic, which is rarely the same as whoever is most accountable. That is the actual failure — not the technology, not the budget.
Your COO
Knows exactly which processes leak money but cannot tell which of them AI can actually fix — and gets a different answer from every vendor.
Your consultants
Produce a good deck, run a workshop, and leave. Nobody who wrote the recommendation is there in month four when it turns out to be harder than the slide said.
Your AI working group
Meets fortnightly, has no budget authority, and cannot cancel anything. Committees can advise on AI. They cannot own it.
02
02What I own
The whole job, not half of it.
Most companies split this role in two — a strategist who can’t build, or an engineer who can’t hold the board conversation. The gap between them is where AI programmes die. I hold both ends.
01
Strategy tied to the P&L
Not a vision deck. A ranked portfolio of AI bets across the business — each with an owner, a cost, and a number it has to move. Anything that cannot be tied to a line in your accounts does not make the list.
02
Process redesign, function by function
AI applied to sales, operations, finance, compliance, and support by redesigning the workflow around it — not by bolting a chatbot onto the old one. This is the part most programmes skip, and it is where the returns actually live.
03
Governance that survives an audit
Risk ownership out of the IT basement and into the business. Approval models, human-review boundaries, documentation, and EU AI Act readiness designed into delivery — so compliance is a by-product of how you build, not a project you run afterwards.
04
Agentic AI without the incidents
Deciding what agents may do alone, what needs a human in the path, and what happens when one gets it wrong. These are leadership calls with real liability attached, and they should not be made by whoever happened to write the integration.
05
Build, buy, and vendor calls
The architecture and vendor decisions that quietly burn six months when they go wrong. I have made them with my own hands on the keyboard, which is mostly useful because it means a good demo does not impress me.
06
Adoption and capability
A system nobody uses returns nothing. Training, incentives, and a working rhythm that makes the new way the easy way — and an internal team that gets more capable each quarter rather than more dependent.
03
03Ways to work together
Start small. Scale the seat.
Most engagements start fractional and grow with the results. Some companies want the sprint first; some want the seat filled from day one.
2 weeks · fixed scope
Roadmap Sprint
A short, blunt engagement to find where AI genuinely moves a number in your business, what to leave alone, and what has to be true before anything ships. Useful when you need to decide something before you commit budget.
- A ranked shortlist of AI bets across your business functions
- Data, tooling, and workflow gaps mapped honestly
- Governance and risk constraints on the table from day one
- A specific recommendation: build, buy, or wait
Fixed fee. No commitment beyond the two weeks.
Fractional · interim · full-time
The Head of AI seat
I take the seat. Embedded with the leadership team, owning AI across the business — strategy, process redesign, governance, vendors, and delivery. Start fractional; scale the commitment up or down as the work proves itself.
- Executive ownership of the AI agenda — one accountable person
- AI applied function by function: ops, sales, finance, compliance
- Governance and audit readiness owned at leadership level
- Build-vs-buy and vendor decisions made with technical depth
- A leadership rhythm: priorities, reviews, and honest reporting
Typically two days a week, three-month initial term, reviewed quarterly.
Scoped delivery
Build to Production
Hands-on work to move the workflow that matters most into production, with the controls and documentation your team needs to run it without me.
- Architecture and implementation oversight
- Workflow and system design around your real constraints
- Monitoring, rollout, and operational handover
- A system your team can maintain — and an exit date for me
Priced per scope, with a defined handover date in the contract.
The commercial shape, before you ask.
The default is two days a week on a three-month initial term, then a quarterly review. Some companies want the two-week sprint first. Some need four days a week for one quarter and one day a week after that. All of it is negotiable, and none of it is hidden — ask on the call and you get a number in the first five minutes, not a proposal three weeks later.
Default shapeTwo days a week
Initial termThree months, then quarterly review
Notice30 days, either side, no exit fee
WhereRemote-first across Europe, on-site for the work that needs a room
04The first 90 days
What the first quarter looks like.
Not a guarantee — the diagnosis in the first two weeks decides what is realistic. But this is the cadence, and if we are badly off it by week six you will hear it from me first.
Days 1–14
Diagnose
Sit with every function. Map where money leaks, where AI is already paying, where it never will. Rank the bets and kill the ones that were never going to work.
Days 15–45
Decide & build
Pick the first workflow that moves a number. Make the build-vs-buy call. Start shipping — with governance designed in, not bolted on.
Days 46–90
Production
The first system live, measured, and audit-ready. A leadership rhythm installed. The next two bets already scoped, with your team running more of the work than I am.
87Day 87 is when the pharmacy platform went live. The cadence isn’t theoretical — it has already run once, under regulation.
05
05A worked example
One case, in full depth.
A compounding pharmacy in Athens. As Head of AI I took it from spreadsheets and phone calls to a live platform — five user roles, 1,000+ medication variants, payments, and AI-assisted treatment workflows — under GDPR and pharmacy regulation, in 87 days.
I would rather take you through one engagement to the bottom than show five logos with nothing underneath them. Ask about the parts that went badly, too — those tend to be the more useful conversation.
RoleHead of AI — operating, not advisory
Outcome87 days from kickoff to production
ScopeWorkflow, platform, AI, payments, governance
EnvironmentGDPR + pharmacy regulation, five user roles
Best fit
A CEO or board that wants one accountable owner for AI across the business — not another committee.
A company spending real money on AI tools without anyone connecting them to the P&L.
A regulated or trust-sensitive business where getting AI wrong has consequences.
A leadership team that wants someone who can talk to the board at 9 and review the architecture at 10.
Not the right fit
You want a keynote speaker or a prompt-engineering workshop.
You need a 20-person outsourced engineering team on day one.
You want AI on the website but not in the operating model.
Nobody on your side can sponsor decisions or clear the path for change.
06
06The obvious objection
The obvious risk of hiring one person.
You are considering making a single external person the owner of something important. That is a real risk and you should price it in. Here is how I keep it small.
01
Decisions are written down, not carried around
Every architectural call, vendor decision, and governance boundary lands in your documentation, in your systems, under your control. If I disappeared tomorrow, the reasoning would not disappear with me.
02
Your team does more of the work each quarter
The engagement is designed to shrink. If your internal capability is not increasing by the second quarter, the model is not working and I will say so before you do.
03
A named internal counterpart from week one
Someone inside the business shadows the seat — usually a senior ops or engineering lead. They are not an assistant; they are the person who holds it when the fractional arrangement ends.
04
A stated exit, not an open-ended retainer
The goal is a company that runs AI well because of how the operating model was built. Engagements that quietly become permanent are a failure mode, not a business model.
07
07How I think
Notes on where AI work actually gets stuck.
Pharma AIMarch 7, 2026
Pharma is spending more on AI. The bottleneck still is not the model.
Budgets are up. Models are better. But adoption keeps stalling on the same thing: nobody sorted the data governance layer underneath.
Read the governance note
Regulated OperationsMarch 8, 2026
Validation is quietly becoming the most expensive part of AI in regulated companies.
The prototype is cheap. The change control, documentation, and review cycles that follow are where months disappear.
Read the validation note
Next step
If AI needs an owner in your business, let’s talk.
Fifteen minutes is enough to work out whether this is the right shape for your stage — and I will tell you if it isn’t. If it is, the next conversation is the long one, with your leadership team in the room.