4 Oct 2026

Companies are creating roles such as Head of AI, VP of AI Transformation and Chief AI Officer.

The job descriptions usually combine several ambitions: find use cases, define the strategy, launch pilots, drive adoption and transform the business.

There is one problem: these are two different jobs.

The first is consulting

Before introducing AI, someone needs to understand how the company actually works: where value is created, how decisions are made, which workflows are slow, where information is lost, what errors cost and which constraints are real rather than inherited from an old procedure.

For most organisations, I would timebox this review to approximately two months.

That is enough time to:

• map the important workflows and decision loops;
• establish operational and economic baselines;
• identify bottlenecks and unnecessary gates;
• separate process problems from technology problems;
• determine where conventional software is sufficient;
• identify where AI or agents can materially improve the outcome;
• define ownership, controls, escalation paths and success metrics.

The question should not be: “Where can we add AI?”

It should be: “Where can we improve revenue, cost, speed, risk or customer experience — and is AI the best mechanism?”

AI is perfectly capable of automating a badly designed process. It will simply help the organisation do the wrong thing faster, at scale and with a more impressive interface.

A useful diagnostic phase should end with a transformation portfolio, not an AI presentation: what to stop, what to redesign, what to automate conventionally and what genuinely benefits from AI or agentic execution.

But the review is the easy part

The main work begins afterwards.

Implementation means changing operating processes, systems, responsibilities, incentives, controls, budgets, data flows and management metrics. It means integrating new capabilities into real production environments and remaining accountable when adoption is slower, exceptions appear and the first assumptions prove wrong.

Consider recruiting.

AI can improve sourcing, screening and candidate scoring. But if the underlying process still treats qualified candidates as traffic approaching a procedural checkpoint, the company has only built a more efficient checkpoint.

The larger opportunity is to redesign the function around its actual outcome: making better hiring decisions and creating the best possible conditions for strong candidates to evaluate the company and accept the role.

That is not primarily an AI problem. It is an operating-model problem.

Why the mandate fails

This is why a Head or VP of AI often receives an impossible mandate. The role may be expected to transform workflows without owning those workflows, change priorities without controlling budgets, and drive adoption without authority over the teams whose behaviour must change.

The predictable result is a well-documented portfolio of pilots.

Real AI transformation requires CTO- and COO-level authority — or a combined mandate with equivalent decision rights. The person responsible must be able to connect architecture with operations, technology investment with business economics, and product delivery with organisational change.

This is not about title inflation. It is about having enough authority to be accountable for the result.

The diagnostic phase can be completed by a consultant in roughly two months. The implementation phase requires an executive operator who can remain with the consequences.

AI transformation is not ultimately about installing intelligence into a company.

It is about changing how the company works — and proving that the new way creates more value than the old one.

Where we have done this

DGTL.TECH applies this approach in its own operations, which serve more than 70,000 customers. Our implementations of agentic AI cover billing, financial control, customer support, infrastructure monitoring and the automation of marketing programs.