Why the leaders best placed to navigate AI will be those most willing to navigate paradoxes.
Companies are investing heavily in artificial intelligence, hoping to cut costs, raise productivity and gain a competitive edge. Yet the technology is advancing faster than many organisations can work out how best to use it.
A study by KDVI, The Human Multiplier, draws on interviews with 61 senior leaders across more than 20 sectors and 14 countries. The report identified five core findings. In this series, we return to the interviews behind the report to unpack the human dynamics unfolding inside organisations and what they mean for leaders.
Our first finding is that leaders face paradoxical choices, which can sharpen or distort judgement. The hardest part of adopting AI is often not the technology itself, but how leaders respond to choices that pull them in opposite directions.
In 51 of the 61 interviews, leaders expressed conflicting feelings about AI. They saw its potential, but also its risks. Some were unsure how much control they could retain; others questioned what adopting it would mean for their people and businesses.
These doubts don’t necessarily signal weak leadership. The greater risk may be the opposite: becoming convinced too soon.
The danger of deciding too early
A semiconductor sales chief described AI as “a powerful engine with unclear brakes”. A telecoms HR head on his company’s AI steering committee admitted, “I feel more incompetent!”, while expressing a strong desire to learn.
Both were grappling with a technology that presents genuine trade-offs. Automating routine work may free employees to do more valuable tasks, but it may also weaken their skills or sense of purpose. AI may help people learn faster, yet over-reliance can deprive them of the experience needed to become experts.
Ambivalence, in this context, need not mean a lack of information or an inability to decide. It can reflect two valid concerns that cannot easily be reconciled.
Yet senior leaders are expected to provide direction, and admitting doubt can feel uncomfortable. Some respond by championing AI; others become sceptical. Both approaches offer a simpler story than the reality.
The danger is that a firm view can narrow thinking. Once leaders decide that AI is either the answer or the problem, they may stop revisiting their assumptions, asking questions and being open to emerging issues. This certainty trap is particularly costly when a decision affects jobs, investment or the organisation’s future capabilities.
Certainty is attractive for a reason. Holding two valid concerns at once is uncomfortable, and a settled view brings relief. Under pressure, people tend to reach for the simplest available response. The champion and the sceptic escape the same discomfort, only in opposite directions.
The trap works in both directions. A senior executive at an industrial group with deep engineering roots described the most change-averse organisation he had ever seen. Teams reluctant to use AI cited data protection, even though the company had its own secure internal tool, and then pointed to its mistakes. “If it does make some initial errors, it is easy to kill it,” he said. The board and executive committee remained unconvinced.
At a French company, a senior leader described how the chief executive’s closest advisers had settled on the view that the company’s data was not clean enough to move forward. The result was no sense of urgency. He suspected it would take a dramatic move by a competitor before the subject was taken seriously.
In both cases, a firm sceptical view closed the question as effectively as enthusiasm closes it elsewhere.
Doubt is not the same as paralysis. Productive ambivalence keeps experiments running and questions open. Paralysis only postpones. The test is whether doubt is producing learning or merely delay.
Three recurring dilemmas
Our research identified three dilemmas that recur across sectors. Read in sequence, they follow the path of many AI decisions: what leaders believe is happening, what they choose to prioritise, and what those choices leave behind.
The first concerns what gets scrutinised: leaders check AI outputs for hallucinations, yet may accept optimistic accounts for their own progress without the same rigour. The second sets values against efficiency: the pressure to cut costs can crowd out questions of fairness, trust and what the organisation stands for. The third sets short-term gains against long-term consequences: savings made today can erode the skills and talent the business will need tomorrow.
We call these dilemmas, but they behave more like paradoxes. A dilemma asks leaders to choose between options. A paradox asks them to hold both sides at once, because each is valid and neither goes away. None of these tensions can be resolved once and for all. Weighed openly, they sharpen judgement, because they force leaders to consider what each option costs. Settled too quickly, they distort it.
| The Dilemma | CAUGHT IN THE CERTAINTY TRAP Judgement is distorted | WORKING WITH THE TENSION Judgement is sharpened |
|---|---|---|
| 01 What We believe Organisational vs AI hallucination | Confident AI outputs and overly optimistic adoption reports are taken at face value. The gap with frontline reality widens and is unseen. | Claims are tested against what happens in practice. AI outputs and management narratives get the same scrutiny. |
| 02 What we prioritise Values vs efficiency | Savings targets are set before the work is understood. Ethics becomes one function's concern while people disengage. | Purpose is agreed before targets. Work is redesigned first, and efficiency and values are weighed together. |
| 03 What we leave behind Short-term gains vs long-term consquences | Entry-level roles and the work through which people learn are out. Expertise and the talent pipeline erode. | Gains are banked while protecting the experience through which people build judgement and expertise. |
The examples below, drawn from the interviews, illustrate the different outcomes to each dilemma.
01 When belief runs ahead of reality
At a global IT services firm, a senior executive responsible for solution incubation described a gap between what leaders believed about AI and what employees were doing with it. Management thought adoption was widespread. On the frontline, he found, “there is almost no AI in practice”.
He called this “organisational hallucination”, where leaders can be as prone to wishful thinking about AI as the technology is to produce false information. He argued that AI’s errors can be examined, while mistaken management assumptions may be treated as fact. This is the first dilemma in its sharpest form.
The same dynamic can play out within a team. At a wellness company, the founders connected AI to the business’s data and, according to the chief operating officer, “just believe everything that comes out of it”. Employees were reluctant to challenge either the founders or the technology. One early chatbot was so unreliable that staff nicknamed it “the drunk uncle”. The company introduced clearer accountability only after its shortcomings became apparent, appointing AI monitors in each function.
In both cases, certainty at the top made doubt hard to voice. When leaders are visibly convinced, employees may stop raising concerns, and leaders lose the information they most need.
Where experimentation is visible and safe, the gap can close. A head of business analysis and change at a media company described teams “borrowing people from other teams to come and show them how they did something with AI”, so that “knowledge is flowing more freely across them”. It helped that people could “literally see your CEO using these tools”. When leaders use AI openly, their picture of adoption is more likely to match what is actually happening.
TAKEAWAY
Test the story against the frontline. Treat confident accounts of AI adoption with the same scrutiny as confident AI outputs, and make it safe for people to say when the two diverge.
02 When efficiency crowds out values
A senior executive at a global beverages company commented on what he is seeing when savings are decided first: leaders assuming AI would let them cut the workforce by 20% without examining how jobs and processes should change. The promised productivity gains did not follow. Instead, employees faced longer hours, frustration and burnout, while the technology was not being used effectively. The trap was to decide how much money AI should save before establishing how it could improve the business.
There is also a psychological cost. Returning to the wellness company, AI had taken over much of the software testing work, and the quality-assurance team had shrunk from 15 people to three. When the company distributed AI licences, the COO heard a question that exposed the tension: “Why should we bother learning AI if you’re just going to fire us anyway?” Management saw an opportunity to innovate and compete more effectively. Employees saw a possible threat to their jobs.
Telling people to adopt a technology is unlikely to work if they believe doing so will make them redundant. Apparent resistance may reflect concerns about job security, professional identity or the loss of skills. Apparent enthusiasm may also conceal similar worries, and neither should be taken at face value.
The split can also be built into the structure of leadership. At one telecoms company, the supervisory board focused on the financial case for AI, while the HR head became the main advocate for ethics on the executive board. Both concerns are legitimate. But if financial returns are treated as a board-level issue and ethical questions are left to HR, important decisions may be made without considering the full consequences. Someone needs to connect the two.
Some leaders are already working the other way round. A federal AI officer in the public sector pointed to “a few genuinely innovative people who see that we should not just put AI on the existing processes but rethink the whole thing”. Starting from the work rather than from a savings target is what allows AI to build capability instead of simply removing cost.
Who sits around the table matters too. The executive chairman of an Asian real estate group described a deliberately mixed structure: a steering committee of “the more senior members, the C-suite, looking at the more operational issues”, alongside “a very intelligent group of youngsters, asking them to go bottom up and look at where the bottlenecks are in the business, what’s not working”. His reason was candid: “not many of us, including myself, have enough detailed knowledge to be able to know where you can really make a difference.” Pairing seniority with first-hand knowledge of the work lets an organisation see where AI can improve things before deciding what it should cut.
TAKEAWAY
Decide what AI is for before deciding what it should save. Redesign the work first, keep efficiency and ethics on the same agenda, and explain how roles will change and what support people will receive.
03 When today’s gain becomes tomorrow’s gap
A chief financial officer in the travel sector was keen to explore AI’s potential. During a long-haul flight, he used it to develop a four-stage implementation plan for every workstream in his function. The resulting 22-page document went to his leadership team as a prompt for discussion.
The CFO intended to open possibilities, but his colleagues saw something rather different. One interpreted the plan as a warning that the CFO would proceed “with them or without them”. Another saw it as a proposal to cut his team from 14 people to seven. The CFO hadn’t proofread the document, and the proposed cuts were an AI-generated conclusion, not a decision he had made. When a senior executive circulates a polished plan, people may assume it carries the authority of a settled decision, whatever the author intended.
Yet the same CFO was far from uncritical about AI’s longer-term effects. He was challenging his chief executive over plans to reduce junior recruitment. If companies stopped hiring people at the start of their careers, he asked, where would their future experts come from? “We are potentially cannibalising our own future,” he warned. Here his doubts sharpened his judgement rather than clouding it. Reducing headcount may improve costs in the short term, but cutting off the entry-level roles through which people develop expertise could leave the business short of talent later.
The same tension runs through how leaders see their own roles. The wellness COO described the opportunity: “I’m just starting to see what focusing purely on the strategic could actually look like without all the burden of analysing data, without all the administrative grunt work.” The semiconductor sales chief quoted earlier named the cost: “When AI can generate analyses, recommendations, or even strategic narratives within seconds, it naturally raises the question: Where do I add value?”
TAKEAWAY
Be explicit about whether AI-generated work is an experiment, a proposal or an agreed plan, and weigh every short-term saving against the expertise and talent pipeline it may erode. Leaders who can acknowledge both faces of AI are better placed to spend the time it frees on the judgement only they can provide.
Make decisions & keep questioning
Across all three dilemmas, the pattern is similar. Judgement suffered not because leaders lacked conviction, but because conviction arrived before the right questions had been asked; where leaders kept the tension in view, the same dilemmas led to better decisions.
The answer is not to delay decisions until every uncertainty has been resolved. Businesses need to test AI, invest in it, and expand its use where the evidence supports it. But there is a difference between making a decision and assuming that the decision must be right.
I think sometimes leaders feel they have to have all the answers. With AI, the leaders who are doing well are the ones who are comfortable saying they don’t. Not yet, anyway.
KDVI describes this as the leader’s role as a judgement steward: deciding where AI adds value, where human judgement remains necessary, and how the two should work together.
In practice, this means making room for challenge, checking whether expectations match results and being prepared to change course. It also means looking beyond immediate efficiency gains to the skills and experience the organisation will need in future.
Senior leaders can start with three questions:
| 01 YOURSELF Look in the mirror ASK Where do I have doubts about AI? Have I settled on a firm view because the evidence supports it, or because uncertainty is uncomfortable? | 02 YOUR TEAM Listen to the room ASK Can people challenge my assumptions without fear of the consequences? What might their enthusiasm or resistance be telling me? | 03 YOUR ORGANISATION Check the reality ASK Are our claims about AI adoption borne out by what happens in practice? Are we weighing short-term savings against ethics, skills and long-term capability? |
That does not mean a leader should avoid taking responsibility. It means being clear about what is known, what is assumed and what still needs to be tested.
KDVI describes this as the leader’s role as a judgement steward: deciding where AI adds value, where human judgement remains necessary, and how the two should work together.
In practice, this means making room for challenge, checking whether expectations match results and being prepared to change course. It also means looking beyond immediate efficiency gains to the skills and experience the organisation will need in future.
Senior leaders can start with three questions:
The aim is to avoid making expensive decisions based on untested assumptions.
The best AI leaders may not be those with the strongest convictions. They may be those who know when to question them.
Working through it with KDVI
The leaders we interviewed asked less for new tools than for support with the human side of AI. KDVI’s three AI specific services are designed to help leaders identify their assumptions and work through ambivalence rather than escape it:
| 01 UNDERSTAND AI Value & Adoption Diagnostic See where you really stand | 02 DECIDE The AI Decision Room Make the decisions that matter | 03 EMBED AI Acceleration Circles Turn decisions into practice |
|---|---|---|
| HOW IT HELPS WITH THE CERTAINTY TRAP Replaces assumption with evidence: how AI is really used and experienced, from the boardroom to the frontline. | HOW IT HELPS WITH THE CERTAINTY TRAP Pulls competing priorities on one table, such as efficiency and ethics or savings and capabilities, so hard choices are made together. | HOW IT HELPS WITH THE CERTAINTY TRAP Groups or four to six leaders, each with a coach, work on real AI problems where it is acceptable not to know. |
| WHAT YOU GET A clear picture from each level Where views diverge and what goes unsaid Three priorities agreed with your leadership team | WHAT YOU GET Difficult conversations brought into the open Decisions made, owned and reviewed An AI Leadership Charter | WHAT YOU GET Better decisions on real challenges A greater tolerance for ambivalence A view of where adoption is stuck, kept anonymous |
To find out more, and receive your copy of The Human Multiplier please visit our AI Leadership Study page.