As artificial intelligence (AI) rapidly permeates every corner of industry and enterprise, leaders are grappling not only with what these new technologies can do, but more importantly, how they can be made purposeful, accessible, and truly valuable. In the latest episode of the Cambridge Executive Business Insights: Rethinking AI podcast, these questions are explored through a candid and deeply practical conversation with Paul Henninger, Global AI Leader and UK Head of Technology and Data at KPMG.
Drawing on decades of experience, including formative work at Fair Isaac (the company behind the world-changing credit score), the episode offers a blueprint for integrating AI into organisations in a way that creates real, measurable impact, whilst navigating the complexities inherent to transformative technological change. Below, we examine the key themes and actionable insights from this rich discussion, considering not just the promise of AI, but the human, organisational, and operational realities underpinning its adoption.
Beyond exponential hype: making AI purposeful
The conversation focused on the idea that AI’s future is not simply a matter of making models bigger, with ever-larger datasets and more parameters, but rather making them smarter, leaner, and more attuned to real-world problems. One concept discussed was the shift from sheer computational power to a more frugal, purposeful deployment: designing systems that do more with less, minimising energy usage and maximising equity and efficiency.
A key theme that emerged was the challenge of translating theoretical AI “magic tricks” into meaningful, tangible change within actual organisations. Those who have worked at the coalface of AI know that the leap from breakthrough maths to day-to-day business value is significant, and fraught with obstacles both technical and, perhaps more so, human and organisational.
The dual challenge: implementing and experimenting
Reflecting on the dual roles of leading on technology delivery and spearheading AI strategy at KPMG, the discussion explored several tensions underpinning AI adoption. The real challenge is not just keeping up with the relentless pace of new models, but embedding AI in the messy, complex environments of real business, where security, regulatory compliance, budgeting, and the reality of thousands of employees create significant additional layers of challenge.
Several points were raised, including the fact that the most profound insights around integrating AI have not always emerged in traditional technical settings, but in the day-to-day operational work of running a large professional services business. As AI is incrementally introduced into everyday processes, be it deploying a Microsoft solution or transforming board-level decision making—the difficulties encountered often revolve less around algorithms and more around navigating the complex “landscape of friction” inherent to large organisations.
The uneven landscape of AI transformation
The discussion explored three broad patterns observed across sectors and geographies in AI’s adoption:
- Uneven, yet fundamental, progress – fundamental changes are underway, particularly in back-office and operational processes across finance, risk, and business operations. However, these changes typically account for less than 10% of the potential value AI could unlock in the coming years. While some organisations are making headway, most are still at an early transformative stage.
- Universal barriers: fear, focus, and friction – regardless of industry, every organisation encounters “fear, focus and friction” as they push past initial AI wins and strive for broader change. These barriers manifest as:
- Fear: concerns about automation, job security, loss of identity and purpose as roles evolve.
- Focus: the challenge of concentrating the right resources and making bold decisions, especially when transformation requires betting “against business as usual”.
- Friction: organisational immune systems and risk management architectures kick in, slowing the pace of change, and making every step forward require negotiation with entrenched processes.
- Regulated sectors: paradoxical early movers – one surprising insight is that highly regulated sectors such as financial services, healthcare, and life sciences, are among the furthest ahead in rebuilding their businesses with AI. This head start is largely owed to pre-existing investments in data infrastructure and modernised systems, necessitated by regulatory pressures. As a result, these industries are paradoxically better positioned to leverage AI at scale, even as they operate under stricter scrutiny.
Blueprint for organisational AI success: lessons from real-world transformation
The episode delved deeply into learnings from KPMG’s own journey with AI transformation, distilling these into several actionable strategies relevant to any organisation seeking meaningful change.
Embedded engineering: people at the heart
One concept discussed was the power of embedded engineering, that is, not simply having AI experts in one corner and business domain experts in another, but physically and operationally intertwining these groups. By embedding top AI engineers directly within domain teams (e.g. finance, audit and forensics), the cycle time from idea to result is dramatically shortened. Crucially, this fosters rapid, iterative feedback loops, allowing both sides to understand what’s possible and what is truly needed.
Aim high: ambitious automation targets
Rather than dabbling with a handful of small use-cases, the most impactful strategy involved targeting transformative, end-to-end automation. Teams were challenged to take at least 30% (and up to 70%) of the effort out of entire processes, be it an audit, a payroll run, or a forensic investigation. The logic is compelling: whilst early wins at 10-15% automation are celebrated, such incremental gains rarely move the dial on financial performance or strategic advantage. Only by pushing for much higher rates does AI meaningfully impact the bottom line, yielding genuine cost and time savings.
Redesign beyond technology: rewiring the organisational immune system
A key realisation was that, beyond a point, further AI improvements do little unless paired with organisational redesign. As automation rates climb, the conversion of technical wins into value plateaus if teams, processes, and governance structures remain unchanged. Critical complementary changes include:
- Redefining roles and workflows
- Revamping risk management and control processes
- Rethinking how value is tracked and measured
Thus, successful AI change demands parallel transformation in human, structural, and cultural domains, not just technology.
Navigating the human element: fear, focus, and the meaning of work
The friction points described above are far from theoretical. They directly affect people as AI automates not just manual tasks but deeply ingrained professional activities.
Fear: more than job loss
While initial anxieties revolve around job security (“Will robots replace me?”), the deeper, more lasting challenge turns out to be one of professional identity and purpose. At high automation rates, individuals, and particularly leaders, must grapple with fundamental questions: What is my unique value? What remains for me to do? Adapting to this new reality is uncomfortable, yet those involved in the change process report higher engagement and agency as they help shape new ways of working.
Focus: betting against business as usual
Perhaps the most profound human challenge is the act of betting on the new, i.e. committing resources, trust, and operational cycles to approaches that break radically from tried-and-tested methods. This is no longer akin to redecorating a familiar room; at 70%+ automation, organisations are effectively tearing down the house and rebuilding from the ground up, all while trying to keep the business running.
Friction: the organisational immune response
Change triggers an immune reaction, as “latent processes” and informal controls (which are often undocumented) spring up to slow, interrogate, or block new ways of working. The more significant the attempted change, the more meetings, reviews, and queries proliferate. For example, a two-month sprint in one project spawned over 240 hours of additional meetings as the old and new collided. This friction is not just a nuisance; it is a signal of the scale of transformation underway, and the distance still to travel.
Born digital, built-for-AI: will the incumbents survive?
The question of whether “AI-native” organisations will ultimately supplant established incumbents is complex. While start-ups and digital natives can move faster, legacy enterprises possess immense advantages: robust commercial relationships, supply chain scale, and brand trust, to name a few. The real differentiator, then, is not size but willingness and ability to transform boldly, to rebuild systems, processes, and teams at a fundamental level.
Drawing a parallel with the first Industrial Revolution, the discussion explored how true transformation does not come from bolting new engines onto old workshops, but from inventing entirely new models, of supply chains, roles, and systems. AI’s equivalent journey is now underway, with all sectors, from government to the smallest start-up, figuring out in real time “how do you build these new models?” There may be no prewritten playbook, but the rewards for those who move decisively are substantial.
The future of work
Central to the societal debate on AI is the question: what happens to people and jobs? The discussion explored this issue in depth.
From human-in-the-loop to human-in-the-lead
A key insight is that the best deployments of AI do not reduce human input to after-the-fact checking. Instead, the optimal pattern is “human in the lead”, where people actively direct, decompose, judge, and shape the problem-solving done by machines. Not only does this safeguard against technical pitfalls (such as “hallucinations” in generative models), it aligns with how value is unlocked: through a productive interplay of machine power and human discernment.
Redeployment, not redundancy
While previous technological revolutions have always led to some job displacement, the net effect, if managed intentionally, can be positive. As in the case of the adoption of spreadsheets, the mix of jobs changes: repetitive, manual tasks decline, while roles centred on analysis, judgement, and creativity grow. The true challenge, as the discussion made clear, is ensuring that people are redeployed, reskilled, and empowered to do the new tasks that add value in the AI era.
Training for the new world: skills that matter
The question of which skills to prioritise in this rapidly shifting landscape was explored at length. Three categories emerged as especially important:
- Fundamental cognitive and interpersonal skills – as machines automate more routine and even “clever” pattern-matching, the irreplaceable value of judgement, creativity, ethical reasoning, and relationship-building comes to the fore. From product design to financial forecasting, the ability to make nuanced decisions, understand context, and exercise discernment becomes paramount.
- Deep domain and pattern recognition abilities – despite AI’s prowess, human intuition and experience in specific domains, such as knowing when a model “just doesn’t look quite right” or when a process is derailing, remain critical. The skill of unearthing new patterns, adapting blueprints, and lateral problem-solving is in even higher demand.
- Technical fluency – with demand for AI and data skills outstripping supply, there remains a pressing need for more engineers, quality assurance testers, data scientists, and process designers. Critically, because the field is evolving so rapidly, there is a relatively level playing field: anyone motivated to upskill can, within a few years, become highly valuable.
Lessons from history: the importance of constraints and operational change
In drawing from past experience, particularly the large-scale fraud detection models at Fair Isaac, the discussion underscored that the greatest value emerges when innovative maths and operational change are combined. Revolutionary as the models were, the step-change in value only occurred when workflow systems and analyst roles were designed around the model output, creating an ecosystem in which human and machine could work seamlessly together.
Another key lesson: sometimes, more constrained AI applications, that is, those tightly focused on a single task, with limited degrees of freedom, can be both safer and more valuable in practice than sprawling general-purpose models. The organisational challenge, then, is designing systems where maximum power is directed, not diffused.
Defining outcomes: the heart of effective AI transformation
A final theme was the importance of clarity of outcome. AI is most effective when the objectives it is asked to deliver are precisely and meaningfully defined, not just at the technical level, but in terms that resonate with overarching business goals and stakeholder needs. This clarity enables both effective AI performance and programme success, anchoring the transformation amidst inevitable ambiguity and challenge.
The road ahead: boldness, particularity, and purpose
As the digital revolution gathers pace, the lesson for organisations is clear: success depends less on off-the-shelf technology, and more on the willingness to redesign, take calculated risks, and foster deep collaboration across technical and domain teams. There is no shortcut around the hard work of addressing fear, focus, and friction, but for those willing to meet these challenges, the potential rewards are transformative.
The conversation concluded with a call to action: define what extraordinary success looks like, engage your most talented people, be prepared to rewrite your playbook, and never underestimate the power, or the necessity, of humans in the lead.
For a full exploration of these themes, and a wealth of practical stories and insights, listen to the full episode, available wherever you get your podcasts.
