Q1. Could you start by giving us a brief overview of your professional background, particularly focusing on your expertise in the industry?

I am an independent AI adviser and transformation program leader with over 20 years across government and commercial transformation, including senior roles at Big Four consultancies. I build and deploy agentic AI in live operational settings - recently across the UK public sector - and have contributed to frontier LLM development. 

I also teach digital transformation and strategy at a UK business school and have authored a book on structured problem-solving.

Q2. How are enterprise AI buying decisions changing? Who is becoming more influential in approving AI investments today?

Enterprise approval is increasingly a ratification activity for what's already underway. Shadow AI means teams are using tools whether sanctioned or not, so the decision in front of a leadership team is rarely whether to adopt. It's whether to direct something already happening. The sponsorship seat has moved away from technology leadership. These decisions used to sit with the CIO or IT director on largely technical grounds. Increasingly, the transformation director owns the agenda, because models and platforms enable activities that generate no value on their own. Value comes from changing how work is organized and how people operate. This is a cross-functional mandate and so shouldn’t sit solely with technology or a Head of AI.
Beyond value creation, the veto on what goes into production also spans multiple areas: security, data governance, and legal. Deloitte found 74% of companies plan to use agentic AI within 2 years, while only 21% report mature governance for autonomous agents. 

So, transformation leaders think in operating models and enterprise value, while risk, security, and data assurance functions think in permissions and auditability. 

Most organizations aren't managing either conversation well. The Microsoft Work Trend Index found only 26% of AI users say leadership is clearly and consistently aligned on AI. Where things have stalled, it's usually because one of those two conversations never happened or leaders aren’t ensuring the conversations are aligned.

Q3. Where are you seeing the biggest disconnect between how AI vendors position their solutions and what enterprise customers actually need?

Vendors sell access. Value comes from adoption. NBER research across nearly 6,000 firms found that 69% actively use AI, while 89% reported no productivity impact over the last three years. Simply giving people access to the tools won’t drive ROI.

Fragmented Solutions vs. End-to-End Value

Value accrues across an end-to-end value stream, and almost every AI product, when it’s baked into an existing enterprise tool, addresses a single slice of one. AI can help customer service agents respond faster, which is a real gain, but that sits inside a much longer chain of intake, triage, resolution, and escalation. Enterprises are often sold a component and asked to imagine a system, with the integration work left to them, unpriced and usually unplanned.

Pilots Reveal Underlying Structural Issues

Pilots - which are a good way to demonstrate use cases - also consistently expose underlying structural problems: SharePoint permissions never properly set, sensitivity labeling missing, no retention schedule, no clear data ownership. That gets characterized as AI causing the problem, when AI is just shining a light on architecture that was already weak. None of it appears in the pre-sales conversation, and it all lands on the customer's budget.

Impact of Organizational Guardrails

Furthermore, disconnects also happen when a product meets organizations' guardrails designed in the pre-AI world. I've seen genuinely valuable agents, triaging contact center queries and passing them to a human for validation, wrapped in so many protective controls that the agent effectively gives up and returns a generic response. 

Vendors optimize for what they can package and demonstrate. Enterprises need something that still works after security, legal and data governance have finished with it. They don’t need a generic product; they need an end-to-end solution. 

Q4. Where do you believe the next wave of enterprise AI spending will be concentrated over the next two to three years?

On adoption

In work I've done, structured adoption returned £3.35 for every £1 spent and saved 21,319 hours, with 94% reporting better collaboration and 88% seeing time savings. Access created the opportunity. Adoption activity created the value.

What produced that result is specific and buyable: confidence-building sessions, peer-led community support, and targeted pilot support. That's the shopping list, and almost none of it is a license.

Testing Organizational Readiness

The same logic changes what a pilot is for. The signal that a pilot is ready to scale isn't that the technology worked. It's that the pilot has stopped proving the technology works and started proving the organization can support it at scale, covering data, permissions, the support model and ownership. That shift, from testing tools to testing organizational readiness, is where a large amount of money is about to go.

Underneath sits the strategic question of where to concentrate the effort. Commodity processes can be bought off the shelf, at the same price, from the same vendors as your competitors. These give genuine efficiency, but no advantage. Differentiating processes can't be bought, which is precisely why they need capability built around them: operating model redesign, hybrid human and agent teams, and people trained to work that way.

The precedent is cloud. Migrating an on-premises tool delivered no performance improvement on its own. The benefit arrived when organizations re-architected how they used it and introduced DevOps practices to match. AI is the same. 

Q5. As a final takeaway, what do you believe is the most important consideration for organizations as they develop their AI strategies over the next few years?

Understand that AI moves the cognitive load rather than removing it. These tools get you to a first draft quickly, so the thinking shifts from ideation to validation. That's a different skill, and almost nobody has been trained in it.

The practical model I'd offer is to keep the machine in the middle. Human ideas first. Machine critique to challenge your thinking. Machine work to build the draft. Then human refinement to finish it. Most people stop at machine work and treat the draft as the output.

The strategic consequence is that the domain specialist doesn't leave. Someone has to be accountable for whether the output is right, and that takes enough expertise to know when it isn't. This is why cost cases built on headcount reduction keep disappointing. Last year, a large financial services business got a lot of press for replacing contact center staff with AI; they've since hired people back because the human in the loop did more than anyone had accounted for.

So build the strategy on the unit economics of hybrid work rather than a technology roadmap. What does this task cost today? What does it cost with a person and an agent working on it together, counting both license and consumption? How do we get more from that person given those costs?