Olena Bilan · Independent Executive Adviser
Enterprise AI rarely fails on the model. It fails on economics, ownership and execution.
The problem is almost never that the technology will not work. It is that the decision has been deferred more than once, or the mandate has landed on someone without the authority to carry it, and nobody has established what the commitment actually costs or who owns the result. My work is an independent read on exactly that, before more capital or reputation goes further in.
For owners, boards, and the CFOs, COOs, CIOs, CTOs and functional leaders accountable for an AI decision, including the decision about what to automate inside their own teams.
Why it does not get resolved internally
Every answer carries an economic motive, and it is rarely stated.
Finance measures what is already measurable.
Technology proves feasibility.
Risk describes controls.
The vendor defends its own solution.
The programme office reports activity.
Each is correct inside its own frame. None of it comes from carelessness. It comes from incentives: the integrator's margin, the vendor's licence logic, the internal preference for an architecture someone has already committed to. Reading those motives is not cynicism. It is how you understand the advice you are being given.
And AI never arrives on its own. It comes attached to data, contracts, processes, incentives and decision rights that were built for something else. That is where the full cost sits, and it is almost never in the paper you were shown.
The result is that the person who owns the decision has no one to ask a direct question without that question becoming a signal about their own confidence. That cost sits outside any project budget.
Why this view is different
Five sides of the same decision.
Seeing all of it at once takes having stood in more than one of those positions.
Group CFO
I decided whether the money moved and carried the consequence when it did. A different relationship to a business case than advising on one.
Programme sponsor
Automation and ERP across nine businesses at different levels of process and team maturity. The obstacle was rarely the technology. It was competing priorities, undocumented process, people who had run the work one way for twenty years, and a generational split in what staff believed the system was even for.
Delivery
Accountable for making it work after the decision was signed, with the timeline, the resistance and the residual risk that came with it.
Inside the vendor
Enterprise AI and analytics into regulated industries for SAS, the American AI and analytics software company. I know how a proposal is constructed, what it is designed to leave out, and where the commercial interest sits inside it.
Architect
I build agentic systems myself: problem framing, architecture, orchestration, evaluation, failure modes and the cost of supervision. It is why I can tell you what an agent programme costs at a hundred rather than at one, and what a proposal is actually asking you to buy.
No vendor, no licence, no implementation revenue. Nothing in my income depends on which way your decision goes.
The work
Decision Frame
Four to five weeks · six to ten conversations · one written position · one working session
The organisations that get a return are rarely the ones with the most AI activity. They are the ones that settled what the money was for before it moved. So the review starts where that gets decided: what choice are you actually making, and what are you deliberately not doing?
-
Follow the money. What has been committed, what it will actually cost, and which costs never reach the technology budget.
-
Map the system. The data, contracts, processes and incentives this attaches to, all built for something else.
-
Test the assumptions. What the case rests on that nobody has checked, including whether this is the problem worth the capital at all.
-
Locate the decision rights. Who decides, who owns the result, and which decisions have already been made by default.
-
Name what has to be true. The conditions under which this works, stated plainly enough to act on.
It ends in a written position you can defend, with a number, an owner and a route. Or a documented reason to stop, pause or do less, which is also a result and usually a cheaper one. We take it apart together with the people who will have to act on it.
Work contracts and extends from either end. What you are buying is judgement about consequences, not information about AI.
Questions
What I am usually asked first.
What does it cost?
A fixed fee, set by scope and duration, quoted in writing before any work begins. There is no hourly rate and no variable element.
Is this a sales conversation?
No. If further work does not make sense, I will say so on the call.
We already have a consulting firm on this.
Then you have a strategy and a roadmap. What you may not have is someone whose only interest is whether the number holds and whether anyone owns the result.
About
I built the financial system before I started reading the AI ones.
I came into finance leadership through business analysis and a three-year ERP implementation, which meant building the financial system rather than reporting through it: 350+ legal entities, the controls, the data, and the dependencies running between every part of the business.
Group CFO for a diversified family-office portfolio. Eighteen years across finance, operations and enterprise transformation, which is where the five views above come from.
Based in London. I work in English, Ukrainian and Russian, across the UK, Europe, Central and Eastern Europe and the Middle East.
INSEAD — Chief Operating Officer Programme, Leadership through Boards of Directors
LSE (London School of Economics and Political Science) — AI Leadership Accelerator
Aspen Institute Fellow, UK and Ukraine
How to start
A short form, then a thirty-minute call. The form covers the decision, what has been committed to it and where the mandate sits. I read it in advance, so the call starts at the substance. Confidentiality applies throughout.