From AI Activity to AI Accountability: What Leaders Must Get Right Now

What separates an AI-driven enterprise from one that simply uses AI tools?

And how do you know when you have moved beyond pilots to measurable business value?

This was the main crux of the conversation at our recent roundtable, where we brought commercial and business leaders together to have an open, honest discussion around AI. Thanks to the mix of job titles around the table, it was a conversation that embraced the commercial potential of AI without ignoring the technical limitations, allowing us to get a unique perspective on how enterprises today are not only approaching AI, but securing impact from it.

In this digest, we round up the core themes discussed by Fusion stakeholders at the event, focusing on five key takeaways: closing the gap between AI activity and outcomes, strengthening data and governance, driving adoption through hands-on use, managing tool sprawl, and addressing the human impact of AI. We also explore why these matter now, and what questions you should be asking to get more from AI than day-to-day efficiency.

5 key takeaways from the AI roundtable

AI tools are still easier to find in enterprises than AI outcomes

What leaders said:

Many organisations are experimenting with AI through tools, pilots and isolated use cases. The bigger challenge is not simply doing more with AI, but using it with enough discipline to build the right products (delivering on the outcomes).

Without that discipline, AI can create momentum without direction. Teams move faster, generate more ideas and ship more AI-enabled features, but still fail to prove whether they are solving the right problems or improving anything meaningful.

That is the real business impact gap: the distance between AI activity and measurable product value. Closing it means applying AI to sharper discovery, better prioritisation, clearer validation and stronger links between product decisions and business outcomes.

Why this matters now

Value definition becomes more important as AI gets easier to access. Simply having AI in your enterprise was never really a differentiator, but it’s even less so now that every organisation can access it. The small advantage gained by saving time using Copilot or ChatGPT has been eroded; if you want to differentiate with AI, you have to do something meaningful, and you need to start doing it now. Leaders need to know what problems AI is being used to solve, what customers want, what success looks like and how progress will be judged.

The useful question is no longer, “where are we using AI?” It is, “where is changing an outcome we care about?” That has to go beyond efficiency and productivity.

5 questions you should be asking about AI outcomes:

• Where is AI already being used in our enterprise?

• Do these use cases have a clear link to business value? If not, then why?

• What are we measuring AI against – its individual usefulness, or its commercial outcomes?

• What business outcomes could we focus on, and work backwards to apply AI?

• What should we stop, pause, consolidate, and what should we focus, extend, and repeat?

Data and governance influence how far AI can go

What leaders said:

Data quality, ownership and governance came up repeatedly, which is nothing new. AI may look like a new capability, but it still depends on the information, processes and decisions that are fuelling it. No matter how complex and advanced AI is, if you flood it with poor or inconsistent data, it’s going to make mistakes. What’s worse is that if you might not even spot them, as AI can fill gaps with its reasoning that previous technologies were more likely to, eventually, expose.

AI can make bad data move faster. If ownership is unclear, AI can make accountability harder. If governance is weak, AI can spread misinformation before anyone can perceive the risk.

Why this matters now:

Waiting for ‘data readiness’ shouldn’t become an excuse for doing nothing with AI; it can be broken down to become a more manageable task, not a business-wide one. If you want to use AI to, say, manage complaints processes, you don’t need every aspect of your data to be perfectly polished and well-governed. Focus on the data that matters for each use case – who owns it, what level of control is needed – and create a governance success story around it. Then rinse and repeat.

Governance should not be treated as the brake on AI. Done well, it gives people more confidence, not more fear, and helps them to use AI in ways that are safe, useful and repeatable.

5 questions you should be asking about data:

• Which data does each priority use case rely on?

• Who owns that data, and who uses it?

• Where could AI amplify poor data, weak process design or unclear decisions?

• What needs to be governed before this use case spreads?

• What story can you tell to make people understand the benefits of data governance?

People adopt AI by using it, not hearing about it

What leaders said:

One of the clearest takeaways from the roundtable was that successful adoption can only happen through use. The sentiment was that “you do AI,” not “you learn AI” (Pavandeep Wakem, Head of Strategic Operations, Investec). Experimentation is getting businesses further than theorising.

AI literacy does matter, but generic awareness can only take you so far. People need to try AI in the context of real work, with examples that make sense in their role, their tasks and the decisions they make, before they can fully understand it.

Why this matters now:

If you want to take AI from an abstract idea to a useful tool, it needs to have relevance. Leaders talked about finding a personal “hook” for AI: that moment where AI starts to make practical sense for an individual, not just as a theory.

That hook could, to start with, be taking the pain out of a repetitive task or helping to review information more quickly, but once they understand that, it quickly evolves beyond efficiency and productivity, and into more imaginative commercial uses. However, this rarely happens without basic experimentation first.

For leaders, the job is to create conditions that promote responsible use. People need enough confidence to experiment, enough guidance to avoid obvious risks, and enough context to understand the difference between saving time for themselves and creating value for the business.

4 questions you should be asking about data:

• Are people learning AI through real work?

• Where are teams already using AI informally?

• What examples would make AI feel relevant to different teams?

• How do we move from personal productivity to wider business value?

Local tool sprawl could become the next integration problem

What leaders said:

The breakfast discussion raised a clear concern around local app sprawl. As teams experiment, they may create or adopt AI tools that solve immediate problems but create complications in the long run.

This isn’t a reason to stop experimentation, but it is a reason to manage it and promote collaboration. If every local solution becomes embedded, businesses can quickly find themselves with duplicated tools, disconnected workflows and missed opportunities. Create a democratic framework that allows people to communicate and share their successes, then look at expanding them, not replicating them.

Why this matters now:

The challenge is to give teams space to explore without letting every experiment become permanent architecture. If every time a useful tool appears, it is copied, replicated, but not connected, you will eventually face the difficult task of connecting, governing or replacing them. Accept that people will use local tools, and make this part of your experimentation, instead of fighting it.

5 questions you should be asking about data:

• Where are AI tools honestly being adopted, not just ‘officially’?

• Which tools are genuinely useful, and which are duplicating effort?

• How will local tools connect into main systems and workflows?

• What guardrails would allow experimentation without creating sprawl?

• How can we improve collaboration and make it part of our testing?

It takes more than reassurance to manage

What leaders said:

Fusion stakeholders need to understand the human impact of AI, not ignore it. The discussion kept coming back to confidence, fear and trust, and the truth is, people are wary of AI as a technology that could replace or dramatically change their jobs, and there are no shortage of horror stories. This does not help seamless adoption.

Why this matters now:

AI is at a critical juncture, and you need to bring people along carefully. Yes, AI is going to have an impact on roles, skills, and processes, but that doesn’t mean it will replace or diminish. Empowerment is more powerful than reassurance, and by showing people how AI can make a difference to them, they are more likely to embrace it. The enterprises who use AI to create better outcomes with their staff, and not to create the same outcomes with less staff, are going to have more longevity. Treat training, experimentation and AI literacy as an ongoing, always-developing part of adoption.

4 questions you should be asking about data:

• Where are people excited, sceptical or anxious about AI?

• Are we making AI feel useful, or only urgent?

• Do teams know where human judgement is still needed?

• How are we helping people move from awareness to responsible use?

If you couldn’t make it to this session, but want the opportunity to talk about these issues and more, sign up for our next roundtable. What better way to improve and assess your AI strategy than over great food with other leaders facing the same challenges? Register your interest and stay up to date with all future GW events here.