In this episode of On the Subject of Leadership, I speak with Clare Kitching, founder of Cambiq Consulting—formerly of McKinsey, QuantumBlack, and Treasury Wine Estates where she served as General Manager of Data, Insights and Analytics. Clare has spent the last fifteen months building an independent practice helping Australian businesses make sense of artificial intelligence—its uses, its limits, and the leadership reckoning it forces on the executives who adopt it.
This is not a conversation about which model to choose, or about agentic architectures, or about productivity benchmarks—though all three surface in the course of it. It is a conversation about what AI cannot do for you, about the leaders who are quietly hoping no one notices they have never logged in, and about the discipline of distinguishing what a client says they want from what they are actually asking.
Clare's argument, developed across more than a decade in global consulting and now refined in the unsentimental environment of small and medium business advisory, is unromantic and worth taking seriously: the AI question is a leadership question; the genuinely strategic work is precisely the work the technology cannot do for you; and the organisations that are getting this right are not the ones with the most sophisticated tooling, but the ones whose leaders are still willing to admit what they do not know.
The Question You Cannot Ask the Model
The most decisive moment in the conversation is also the simplest. AI, Clare observes, can do a great deal. It can draft your slides, summarise your reports, automate your processes, and—increasingly—sit between you and a great many of the tasks that previously consumed your day. What it cannot do is tell you how to grow your business. You cannot, she says, go to chat and ask make my business grow, how do I do this?—because the answer, if it exists, has not been written down yet.
This is a more substantive observation than it first appears. The economic literature on innovation distinguishes, following James G. March's 1991 article in Organization Science, between exploitation—the refinement of existing capabilities—and exploration—the discovery of new ones. The first lends itself to optimisation; the second does not. AI, in its current form, is overwhelmingly an exploitation technology. It accelerates the work of producing outputs that resemble outputs it has already seen. It is not, and cannot be, a substitute for the imaginative work of producing outputs no one has yet produced.
The implication is uncomfortable for the prevailing AI discourse, which has tended to elide the distinction. Productivity tools and growth strategies are routinely conflated, as though saving the marketing team three hours a week were on the same ontological footing as deciding what business one is in. The first is a question AI can help with. The second is a question only a leader can answer, and the more capable the technology becomes at the first, the more visible the failure to attend to the second becomes.
Clare is alert to this asymmetry. Her observation that the size of the prize in productivity is "a lot less than what you can do in terms of growth and taking some big bets" is not a contrarian provocation; it is a quietly stated empirical claim. McKinsey's State of AI in 2025 survey supports it: of the 88% of organisations now using AI in at least one function, only 39% report any EBIT impact—"most of those respondents say that less than 5 percent of their organization's EBIT is attributable to AI use". Firms who have gone beyond the 5% are those that have moved past the productivity framing and begun to pursue revenue-generating use cases.
When the Brand Walks in Before You Do
Clare's account of leaving McKinsey opens the conversation, and it deserves attention not as a founder anecdote but as a leadership observation. She describes a coaching exercise inside McKinsey in which she was asked to introduce herself and got as far as I'm Clare, I'm from McKinsey before stalling. While Clare was quick to assert that McKinsey was always willing to provide media training for people representing the firm, there is a deeper question of brand names carrying individuals.
This is a recognisable pattern. Edgar Schein's foundational work on professional identity describes the way institutional affiliation furnishes its members with what he calls a career anchor—a set of values, competencies, and identity claims that travel with the role. When the role is removed, the anchor is removed with it, and the individual is confronted with the harder question of what, exactly, they still are. Most consultants leaving Tier 1 firms underestimate the cognitive labour this requires; many never quite recover from it.
The leadership generalisation is broader. In any large organisation, senior people accumulate a great deal of borrowed credibility—from the brand, from the title, from the institutional context in which their judgement is exercised. The question of what remains when those scaffolds are removed is one most leaders prefer not to ask. Independent practice forces the question. The independent's value proposition cannot be inferred from a logo; it must be articulated, repeatedly, in environments where the listener has no prior reason to grant trust.
Clare's answer, developed by trial and error, is that the only path to fluency is repetition. There is no shortcut, and the discomfort is the work. This is consistent with the deliberate-practice literature—Anders Ericsson's research on expert performance has consistently shown that fluency in any complex articulation task emerges only through extensive, feedback-rich repetition. The corollary is that leaders who have spent decades inside institutions that did the articulation for them are likely to be much less fluent than they suppose.
The Leader Who Has Never Logged In
The most uncomfortable observation in the conversation, and the one most likely to produce productive friction in the audience Clare reaches, concerns the executive who has signed off on an AI strategy but has never personally used the tools their organisation is being asked to adopt.
The pattern is, in her experience, common. The executive endorses the initiative. The team is told to get on with it. The board is reassured. And nobody has yet asked the executive what they personally use the technology for, or what they have learned by using it.
This is a leadership failure rather than a technological one. Henry Mintzberg's work on managerial behaviour, going back to The Nature of Managerial Work in 1973, has long argued that senior managers who lose contact with the actual content of the work over which they preside become increasingly dependent on filtered information—and increasingly poor judges of which filters are reliable. The AI case is a near-perfect example of this pathology. A leader who has never used the tools cannot evaluate whether their team's account of those tools is accurate, ambitious, or self-serving. They are reduced to refereeing claims they cannot independently verify.
Clare's diagnostic question—the one she wishes more boards would ask their executive teams—is therefore not rhetorical. Are you using AI day to day, personally? is, in fact, the single most discriminating question a board can ask about the seriousness of its organisation's AI posture. The leader who has logged in and used the tools, however unevenly, has acquired a small but irreplaceable form of operational knowledge. The leader who has not is operating on borrowed conviction.
There is a secondary consequence worth naming. In six months, Clare observes, the team members who have been told to adopt AI but whose leaders have not adopted it themselves will begin to ask, reasonably, why should I? The cultural cost of an unmodelled expectation is real, and it compounds. The literature on what Schein termed primary embedding mechanisms—what leaders pay attention to, model, and reward—is unambiguous on this point: people learn the operating reality of an organisation from what its leaders actually do, not from what they declare.
The Buzzword Substitution
Clare describes a recurrent client conversation. An executive team announces that it wants to "get started with AI." A scoping conversation begins. Within minutes it becomes clear that the executive team does not, in fact, want to know about AI. They want to know how to use Copilot, or Claude, or ChatGPT. The substantive question has been displaced by an unrelated one, and the displacement has gone unnoticed because the vocabulary was sufficiently fashionable to cover the gap.
This is a familiar pattern in management practice. Buzzwords function, in part, as social signals of currency. They allow participants to demonstrate that they are orientated toward the right things without having to commit to any particular interpretation of what those things are. The cost is that genuinely substantive questions—what business problem are we solving; what change in operating reality are we attempting to produce—are quietly displaced by questions whose answers happen to use the same vocabulary.
The remedy, in Clare's account, is patient interrogation. The consultant's first job is not to answer the question the client has asked, but to discover what question the client is actually asking. This is closer to the diagnostic work of clinical medicine than to the prescriptive posture of much commercial consulting. The danger of the latter is well documented: William F. Gellermann's classic critique of consulting practice—that the consultant who answers the question the client poses, rather than the question the client needs to pose, is functionally complicit in the client's failure—applies with renewed force in the AI context, where the question itself is often poorly formed.
Clare's discipline here is not glamorous. It looks like asking, of every confidently deployed buzzword, what are you actually trying to know? But it is the work that distinguishes a useful adviser from an expensive echo.
Curiosity Is Not a Personality Trait
The conversation arrives, by way of a discussion about which employees adopt AI most readily, at one of its sharpest observations. The differentiating variable is not age, Clare argues, and it is not seniority, and it is not technical background. It is curiosity—and specifically, the willingness to lean into ambiguity rather than away from it.
The empirical literature broadly supports this. Carol Dweck's work on growth and fixed mindsets, developed across more than three decades at Stanford, demonstrates a robust association between the disposition to treat one's own capabilities as expandable and the actual rate at which those capabilities expand. The mechanism is unsurprising: those who believe they can learn are more willing to expose themselves to the temporary incompetence that learning requires. Those who believe their abilities are fixed avoid the same exposure, and the avoidance is self-confirming.
Todd Kashdan's research on curiosity, more recent but consistent with Dweck's, has identified what he calls stretching—the active pursuit of novelty and challenge—as a stable predictor of professional adaptability across a range of domains. The implication for AI adoption is direct: the leader who treats the technology as something to be approached with curiosity, rather than as a threat to existing expertise, will, on average, get more out of it.
Clare's account is alert to the failure modes. Curiosity, she notes, can also become its own pathology—the late-night rabbit hole, the cognitive overload, the sense that one is being productive when one is in fact merely busy. There is a recent body of research, including studies cited by the Microsoft Work Trend Index and various academic reviews, suggesting that heavy AI users report higher rates of cognitive fatigue and burnout than light users, despite producing more output. The discipline, Clare suggests, is the same one good leaders have always exercised: to know when to stop, to know what to filter, and to recognise that the abundance of possibility is itself a leadership burden.
The Maintenance Problem
The set-and-forget assumption that characterised enterprise software for two decades does not apply to artificial intelligence. The models change. The outputs drift. The usage patterns of one's own team change, often without anyone noticing. The technology Clare's clients deploy in March is, in non-trivial respects, not the same technology in September.
This has uncomfortable implications for procurement and governance. Most organisations are equipped to make capital expenditure decisions on a multi-year horizon. They are not equipped—culturally, structurally, or financially—to treat the implementation as the beginning of a relationship rather than the end of a project. The board sees the demonstration, approves the spend, and is then bewildered, six months later, that the operational metrics have moved in ways no one predicted. The metrics moved because the technology moved.
Clare's framing is that AI implementation is now an ongoing partnership rather than a discrete deployment. This is consistent with what Boston Consulting Group identified in The Widening AI Value Gap—the small minority of organisations they classify as "future-built" are distinguished by their treatment of AI as an enduring operational capability rather than a programme. The majority, who treat it as a programme, find that the programme finishes but the technology does not.
There is a leadership discipline here that is easily overlooked. The leader who treats AI deployment as a discrete decision can return to the rest of their work once the decision is made. The leader who treats it as an ongoing relationship must build, into their own cadence, the recurring attention the technology requires. This is the maintenance no one budgeted for, and the absence of it is the most reliable predictor of the disappointed twelve-month review.
The Conversation That Has Not Happened
Late in the conversation, Clare names something that few organisations have yet articulated. Most teams have not had an explicit conversation about how they will use AI together—what quality standard applies, what kinds of work are appropriate, when the use of AI must be disclosed, and what the team's collective expectation is of its members.
The result is predictable. One member produces three pages where three bullet points were requested, because the tool made it easy. Another member produces three bullet points where three pages would have been more useful, because they have already pre-summarised. A third produces a draft that no one is quite sure was written by the person whose name appears at the top. The work is being done, but the basis on which it is being done has not been agreed.
This is a coordination problem of a familiar kind. Elinor Ostrom's Nobel-winning research on the governance of common-pool resources demonstrated, repeatedly, that communities sustain shared resources successfully when they have explicit agreements about use, sanctions for violation, and forums in which disputes can be aired. AI in a team setting is, at its current stage, exactly such a common-pool resource—a capability whose value is degraded by uncoordinated use, and whose productive deployment requires explicit team-level agreement.
Clare's prescription is unromantic and useful: have the conversation. Not as a compliance exercise—the paper-policy approach she is consistently sceptical of—but as a genuine working agreement. What are we using AI for? What aren't we using it for? What does good look like when AI is in the loop? What is each of us willing to disclose about how we used it?
The conversation is not difficult to have. It is merely unfamiliar. And the cost of not having it, in her account, is paid in volume—endlessly expanding outputs that no one has time to read, no one has incentive to compress, and no one has authority to filter.
A Practical Account of Leadership
Clare Kitching is not selling a methodology, and this is not a conversation about AI in the abstract. It is a practical account of what it has taken, over the past fifteen months of independent practice and more than a decade inside the institutions that trained her, to help Australian businesses get genuinely useful work out of a technology whose vocabulary has run substantially ahead of its applied practice.
What she offers, across the arc of the conversation, is not a method but a set of dispositions. Use the tools yourself before you ask others to use them. Distinguish what your client says they want from what they are actually asking. Treat curiosity as the differentiating trait it is, not as a personality quirk. Do not confuse productivity with growth. Have the team conversation no one has had yet. And accept that the technology you deploy is not the technology you will be using in twelve months, and budget your attention accordingly.
If you are leading an organisation through AI adoption, or have signed off on an AI strategy you have not yet personally tested, or simply want to hear what genuinely thoughtful practice in this area sounds like, this is a conversation worth hearing in full.
Good night, and good luck.