Make the AI decisions that cannot be delegated
AI is already in your organisation. It is drafting client communications, screening candidates, and summarising the documents that inform decisions. Most of it arrived without a decision being made about it.
The question is no longer whether to adopt AI. It is which decisions about AI still belong to you, and whether anyone in the organisation can currently say who made the last one.
Five decisions that cannot be delegated
Where AI deserves investment, and where it does not
Every function now has a proposal. Few have a business case. The portfolio question is not which use cases are technically possible, but which justify the value they create against the control burden they impose.
Who owns an AI-assisted decision
Not who operates the tool. Who is accountable when a decision shaped by a model turns out to be wrong, and whether that person knew the model was involved.
What must be governed before it scales, and what does not
Blanket policy stalls adoption. No policy creates exposure. Proportionality means deciding, deliberately, which categories of use warrant control and which can run with light oversight. Where regulation applies, whether sector supervision or the EU AI Act for organisations placing systems on the European market, it sets a floor, not the whole answer.
What is already happening that nobody approved
Employees adopt AI faster than organisations govern it. Unmanaged use is not a productivity story. It is exposure that surfaces at the worst possible moment.
What the board actually needs to see
Which indicators belong in reporting, at what cadence, and what a board member should be able to ask and receive a straight answer to.
If two or three of these do not currently have an owner, that is the conversation to have.
If that is where you are, the first step is a short conversation about which of the five are currently unassigned. Find a time to talk.
How I work with your organisation
Decisions about AI are rarely made in one room. They are shaped vertically, from the board through management to the people who actually run the process, and horizontally across functions that each see only part of the picture.
I make sure those conversations happen. Whoever makes the decision, sits in the decision process, or lives with its consequences belongs in the work, and it is my job to bring them in rather than yours to assemble them. You do not need to prepare the organisation for this. That is part of the engagement.
That is what business first, technology second, governance always means in practice. The technology question has an answer once the organisational one is clear. Getting there requires talking to the people who know how the work is really done, not only to the people who describe it.
No pre-work. No data request list before the first conversation.
Why this works across sectors
The technology conversation is sector-specific. The decision conversation is not.
A private bank, an energy utility and a manufacturer face the same four questions: where to invest, who is accountable, what to control, and what is already running unmanaged. The context changes, a supervisory expectation here, a safety constraint there, but the decisions, and the executives who own them, are the same.
Twenty years of transformation across financial services, healthcare and pharma, energy and utilities, manufacturing, education and the public sector. From C-level to shop floor. Delivered in English, French or German.
How the work is structured
Executive session
With the executive team and the functional leaders whose areas are affected. Working through the portfolio and the ownership questions against your actual adoption, not a generic framework. You leave with the decisions named and assigned.
Advisory sprint
Scoping AI opportunities across defined processes, prioritising against operational and regulatory constraints, and defining governance guardrails, integration direction and a roadmap for follow-on work. Includes conversations across the functions involved, not only with the sponsor.
Governance review
Assessment of current adoption, decision ownership and unmanaged use across the organisation, with the gaps stated plainly and a prioritised set of actions.
Ongoing board-level advisory
Fractional engagement for organisations scaling adoption over a longer horizon. Typically a standing monthly session with the executive team plus availability between sessions, with a quarterly review at board level.
What you receive
AI strategy
Where AI fits your business strategy, which capabilities you build internally and which you buy, and what sequence the adoption follows. Set against your actual operating model, not a reference architecture.
A prioritised use-case portfolio
Which use cases enter the portfolio and in what order, with the control requirement stated alongside the expected value.
A decision-ownership map
Which AI-influenced decisions exist, who is accountable for each, and where accountability is currently unassigned.
A summary of unmanaged adoption
What is already running that nobody approved, and the exposure it creates.
A board reporting framework
What leadership should see, at what cadence, and which questions a board member should be able to ask and receive a straight answer to.
Building the capability to keep deciding
An assessment tells you where you stand today. It does not help you six months from now, when the tools have changed, three new use cases have appeared, and the person who understood the last decision has moved on.
The organisations that stay in control of AI are the ones that build a place where these decisions keep getting made.
A cross-functional structure that owns AI decisions
Not a committee that reviews proposals, and not a capability parked inside IT. A working group drawn from the functions where AI is actually used, with a defined mandate: which decisions it takes, which it escalates, and who chairs it. I advise on how it is composed, what authority it needs to be effective, and how it connects to existing governance rather than sitting beside it.
Institutional intelligence about your own AI use
Most organisations cannot answer basic questions about their AI: what is running, in which processes, at what cost, producing what result. That knowledge lives in individuals and in vendor relationships, and it leaves when they do. The structure's first job is to make it the organisation's, current, visible, and independent of who happens to be in the room.
Feedback loops that reach managers in time
What managers see, how often, and at which thresholds they act. Course correction is cheap early and expensive late. The reporting has to arrive while adjusting is still an option, which means designing it around decision points rather than around reporting periods.
The intended outcome is that you need me less over time, not more.
Selected engagements
Advisory sprint scoping AI-agent opportunities across back-office processes, prioritising use cases against regulatory and operational constraints, and defining governance guardrails and integration direction.
Company-wide AI adoption and data architecture programme. Governance established, use cases prioritised, target architecture directed across ERP, SCADA and GIS, with training and change support.
Data-platform modernisation from planning through delivery, with an agile delivery framework and cross-functional team enablement.
Industry 4.0 portfolio digitalising manufacturing processes and production machinery.
Ervane has a rare ability to connect strategic ambition with operational reality. She helped us move beyond general discussions about AI and clarify where it could create value, what needed to be governed, and which decisions required executive ownership.
Director of Strategy, Internal Control & ReportingWhat impressed me in particular was your entrepreneurial thinking and how consistently you pursue a business-driven approach. That fits our DNA exceptionally well.
Partner, Swiss private bankStart with a conversation.
Most engagements begin with a short discussion of where AI currently sits in your organisation and which decisions are unassigned. No preparation required.