What CEOs Get Wrong About AI Capability

If you are the CEO of a business investing seriously in AI, the signs can be misleading.

You have bought new tools. Teams are experimenting with generative AI. You may have hired data scientists, engineers or AI specialists. Different parts of the business are bringing forward use cases and vendors are showing you increasingly sophisticated technology.

From the boardroom, it can look as though capability is growing.

But you can have a great deal of AI activity without becoming much better at putting AI to work across the business.

Lucent Search's research with AI and data leaders shows how common that problem is. 91% reported challenges moving AI pilots into production and more than 89% struggled to integrate AI into existing systems.

If you want to understand how capable your organisation really is, look at what happens after the prototype works.

Can you connect it to existing systems? Can you get the data you need? Can someone make the decisions required across functions? Will the business change the way it works? Is there a clear route through risk and governance? Can you secure the investment to scale it?

Those are usually harder questions than buying the technology or hiring another technical specialist.

Start with what you actually need AI to change

One of the first problems appears when AI has become a strategic priority without being translated into a sufficiently specific business agenda.

You know AI matters. Your board knows it matters. Different executives can see opportunities.

Very quickly, you can end up with dozens of use cases competing for money and attention.

Marketing is experimenting with generative AI. Operations has automation opportunities. Technology is looking at platforms and infrastructure. Data teams are developing models. Individual business units may already be buying tools themselves.

There may be a lot happening, but it can still be difficult to tell which activity is actually worth backing.

You need a clearer view of where AI could materially change the performance of the business.

That might mean reducing the cost of a core process, improving asset reliability, making better pricing or investment decisions, changing customer service economics or managing risk more effectively.

Once you know which outcomes you are trying to change, many of the organisational questions become easier.

You can see which data is required, which systems are involved, where processes need to change and which executive already owns the underlying business result. You can also see where the gaps really are.

Without that clarity, it is easy to mistake a large portfolio of AI activity for progress.

Buying the technology is often the easy part

AI tools have become much easier to access. That has lowered the barrier to experimentation, but it can also give you a false sense of progress.

Your team may be able to get a platform running and produce an impressive prototype surprisingly quickly.

Then it has to operate inside your company.

The data it relies on may sit across several functions. The integration work may depend on a technology team with a full programme of existing commitments. Risk or legal may need to approve the application. The process itself may have to change. Finance may want stronger evidence before releasing further investment.

None of that is unusual in a large organisation. It is where many apparently straightforward AI projects become difficult.

Lucent's research found that integration remains one of the most significant barriers to scaling AI, with 39.1% of AI leaders describing it as very challenging and 50% moderately challenging.

So before you approve another major platform or tool, establish what will have to change around it.

If the answer depends on five other functions doing something differently, you need to know who has the authority to make that happen.

Pay attention to the decisions that sit between your executives

This is where many AI programmes become much more difficult than they looked at the outset.

You may already have good people in the obvious roles.

Your CIO controls significant parts of the technology environment. Your CDO or CDAO may own data and analytics. Operations owns critical processes. Business unit leaders carry P&L accountability. Finance controls investment. Risk and legal set important boundaries.

Each of those arrangements can work perfectly well on its own.

The problems often sit between them.

• Who can decide that an AI initiative takes priority over another technology programme?

• Who is responsible when the data required for a valuable use case is poor?

• If a business unit needs to redesign a process, who makes sure that actually happens?

• If risk appetite, commercial benefit and technical feasibility point in different directions, who makes the trade-off?

Lucent's research found that 51.1% of AI leaders considered cross-departmental collaboration very challenging and a further 35.9% moderately challenging. Respondents described competing priorities, fragmented data agendas and difficulty coordinating work across functions.

You can hire excellent AI people into that environment and still make very little progress.

Someone needs enough authority to resolve those issues without rebuilding the coalition for every project.

That person could be your CIO, CTO, CDO, CDAO, Chief AI Officer or another senior leader. The title is secondary. You need to be clear about which decisions they can actually make, what they control and where they depend on their peers.

If every important AI decision requires another round of negotiation between functions, you do not yet have a reliable way to scale it.

Do not let the AI team inherit the business result

As your AI capability becomes more established, there is another trap to avoid.

The specialist team starts to become responsible for almost anything involving AI.

That can look sensible because the AI or data team has the expertise. But the business outcome still belongs to the executive who owns that part of the company.

If you are using AI to change customer service, the executive responsible for customer performance still owns the result.

If you are changing an operational process, the operations leader still owns its performance.

If the investment case depends on improving margin within a business unit, the P&L owner still owns the economics.

Your technology, data or AI function may provide the platform, engineering, models, governance and specialist expertise. It cannot carry accountability for every commercial or operational result produced using those capabilities.

You need business leaders involved early enough to shape priorities, make process changes and take responsibility for adoption. Otherwise, your specialist team can deliver technically good work that never becomes part of how the business operates.

You need a repeatable route from idea to deployment

As your investment grows, you should be able to explain how a promising AI opportunity moves through the business.

Who assesses whether it is worth pursuing? Who decides whether it has priority? Where does the technical capability sit? Who deals with the data and integration requirements? How does the business unit participate? Who decides when there is enough evidence to scale?

There is no standard organisational model you need to follow.

A central AI team may make sense for you. You may want more capability inside individual business units. In many large organisations, some combination of central expertise and distributed delivery will be more practical.

The design will depend on your existing technology environment, risk profile, organisational maturity and the types of AI you are trying to deploy.

But you do need enough structure that progress does not depend on a small number of people knowing who to call.

Personal relationships and energetic sponsors can get you through the experimentation stage. They are a poor substitute for an operating model once you are trying to deploy AI across a large or highly governed organisation.

Governance should help you make decisions, not simply add control

For many CEOs, AI governance initially appears to be primarily a risk question.

It is also a delivery question.

You need clear answers on which applications require additional scrutiny, who owns model risk, what evidence is required before something can go live, how third-party AI is assessed and who has the authority to stop a system being used.

Lucent's research found that 88% of AI leaders experienced difficulty establishing effective AI governance.

When those decisions are unclear, teams lose time finding out who can approve what. Each new use case develops its own route through the organisation and senior executives become more cautious because nobody is quite sure where the boundaries sit.

That becomes especially problematic if you operate in a regulated, asset-intensive or heavily governed environment.

You need sufficient control to understand and manage the risk, while still giving good projects a practical route into operation.

If every use case has to renegotiate that route from the beginning, governance has become part of the scaling problem.

Look at what happens when a pilot succeeds

Funding is another useful test of whether AI has genuinely become part of the business.

Getting money for experimentation is often relatively straightforward. Scaling the work is different.

Once a pilot succeeds, you may need production infrastructure, systems integration, more engineering capability, ongoing monitoring and significant changes to existing processes.

Lucent's research found that 71% of AI leaders experienced difficulty securing adequate budget for AI initiatives.

You should expect AI projects to compete for capital like anything else. A promising use case still needs a credible investment case.

The problem comes when you have no established route from evidence to investment.

If a successful pilot effectively has to restart the funding conversation from zero, you will accumulate demonstrations without building much enterprise capability.

This is one reason business ownership is so important. Finance can assess the investment more effectively when it is clear which business result should change, what it will cost and which executive is accountable for delivering that result.

The same applies to return on investment.

85% of leaders in Lucent's research reported difficulty measuring AI ROI.

Some benefits are genuinely difficult to isolate. AI may affect productivity, risk, customer experience or operational capacity rather than producing a simple standalone financial return.

But measurement also becomes difficult when nobody agreed at the beginning what the work was supposed to change.

You should know why an initiative exists before you scale it. That gives you something meaningful to measure afterwards.

Before you hire another executive, work out what is actually failing

As AI becomes more important, it is natural to ask whether you have the right leadership.

Sometimes you do need a different executive.

But first examine the job you are asking your existing leaders to do.

Your CIO may already be capable of leading AI if the main challenge is technology integration and they have sufficient influence across the business.

Your CDO or CDAO may be better placed if the agenda has grown primarily out of data, analytics and machine learning.

Your CTO may be the natural leader where AI is closely tied to engineering, products or platforms.

A dedicated Chief AI Officer can make sense when the agenda has become substantial enough to require sustained enterprise leadership across several functions.

You only know which of those options is right once you understand where the current model is breaking down.

Look at the practical evidence.

• Can priorities be agreed?

• Can your AI leader get decisions from their peers?

• Can funding move when a use case has proved itself?

• Can technology and data dependencies be resolved?

• Do business executives know which results they own?

• Does governance give teams a clear route to deployment?

If the people are capable but those mechanisms are weak, adding another executive may simply put another person into the same problem.

If the existing leader lacks the judgement, influence or breadth required to operate at enterprise level, you have a different issue.

That is when the leadership question becomes real.

Lucent's research also points to that shift. Strategic thinking was among the capabilities in greatest demand. Stakeholder management was rated highly in demand by 52% of respondents and change management by 43%.

As AI moves further into the organisation, technical credibility remains important. But your senior AI leader also has to make difficult choices, influence other executives, secure investment, navigate governance and connect the work to business performance.

That is a much broader job than being your strongest technical specialist.

How to tell whether you are actually building AI capability

You are building genuine capability when useful AI stops depending on extraordinary effort from a few individuals.

You can identify the opportunities worth pursuing and stop the ones that are going nowhere. Teams can get access to the systems and data they need. Business leaders know what they are responsible for. Governance decisions follow an understood route. Successful work can attract the funding required to scale.

And when something gets stuck, you know who can make the decision.

That tells you more about your real AI capability than the number of tools, pilots or specialists you can point to.

If you are spending heavily on AI but still finding it unusually difficult to move good ideas into the business, examine the organisation around the technology before assuming you need more of it.

You may need stronger AI leadership. You may need different executive responsibilities.

Or you may need to fix the conditions in which your existing leaders are trying to deliver. You should know which problem you have before deciding who to hire.

Related analysis and executive search

We’ve invested heavily in AI but aren’t seeing enough business value. What kind of leadership do we actually need?

How to distinguish between problems of business ownership, technology, data, operating model and executive capability.

Can your CIO lead AI, or do you need a different leadership model?

How to assess whether your existing CIO has the remit, capability and authority to lead enterprise AI.

Should you hire a CDO, a CDAO or a CAIO?

How to decide whether data, analytics and AI should remain together or require separate executive leadership.

What do AI and data leaders need from you to deliver?

How authority, sponsorship, resources and organisational conditions affect senior AI and data leaders.

Chief AI Officer Executive Search

For organisations considering dedicated enterprise AI leadership.

CDO / CDAO Executive Search

For organisations appointing senior data and analytics leaders.

Building AI capability beyond the tools?

Lucent Search works with boards and executive teams to establish where AI responsibility should sit, which leadership capability you need and whether your existing operating model gives that leader a realistic route to delivery.

Where a genuine leadership gap remains, Lucent conducts retained executive search for senior AI, data and technology leaders.

Speak with Rebecca