The second night of AI-focused roundtables from Tech Monitor and AMD this March took the team to Copenhagen. After an evening of discussion in Stockholm, it was time to talk ‘From AI promise to business impact: building future-ready enterprise AI’ with a group of senior IT professionals representing some of Denmark’s biggest enterprise organisations. As before the objective of the night was to understand the challenges of rolling out meaningful and measurable generative AI (GenAI) applications and to share best practice.
Cultural resistance and the need for active leadership
As on previous occasions, attendees attempted to diagnose the issues responsible for the sometimes slow rollout of GenAI applications. Pilots and proofs of concepts are one thing, but deploying applications that drive business value is a trickier proposition altogether. Why? Siloed systems and siloed teams featured near the top of the list of reasons. Similarly, attendees identified a tendency to stick with legacy processes and thinking. If AI allows organisations to rethink the way they approach problems and opportunities – and by definition, rewire the processes that underscore delivery – organisations seem strangely reluctant to embrace the opportunity.
Cultural resistance is at the root of this tendency and it demands active leadership to overcome its impact. In practice this requires business leaders to champion change by explaining its consequences with clarity and transparency. Too often, noted one voice around the table, organisations implement change without “answering the why – why are we doing it.”
Good leadership also demands that people within the organisation understand their roles as the impact of AI begins to be felt. As one attendee noted, that means different conversations with different cohorts. The “ops people who will be impacted most” require different incentives to the developers who should be encouraged to use these new tools as a master carpenter would apply his or her craft using the best available hand tools.
How to choose the right use cases
During the evening it was suggested that one of the biggest problems with GenAI is deciding which use cases to pursue. Coming up with ideas is not the issue. The issue is selection. His fellow attendees offered some advice in response. One said simply that to understand where to focus attention “first look at what is core to your business”. Align effort to the missions and objectives of the organisation. If your mission is to create the most effective and easy to use diagnostic tools, for example, make sure every application of GenAI enhances that ambition. “It really is that straightforward,” he insisted.
Another, faced with 8,000 internal ideas of how to apply GenAI, used artificial intelligence to develop a weighted scorecard to assess suitability. “We created an AI tool to review AI use cases,” she said fully aware of the irony of that statement. The scorecard assesses hard factors (such as projected cost savings and outlay), soft factors (including customer and employee satisfaction), feasibility (by technology, data, and governance), and time to value.
Offering a different view on use case selection, another voice suggested the real test is having the bravery to “say no”. He said there was commercial benefit in “not choosing” either because technology’s advance might render an initial idea redundant, or that by the time you have clearance to build the use case, an off-the-shelf version of the same thing becomes commercially available.
Governance: not the enemy of progress
Issues of data privacy, sovereignty, and security continue to occupy the minds of senior IT leaders as they seek to advance GenAI projects within the guardrails that their industry and organisation demands. Most are trying to find practical ways of balancing governance obligations with business needs. One security specialist around the table said of her role, “My job ends when the ‘no’ turns to ‘yes’.” By way of practical example, another attendee noted that his company’s price model could never feature in a public large language model (LLM) for fear of losing control of commercially-sensitive information. The solution is to build a private model and accept the trade-offs in effort and cost.
This is a dilemma of innovation. “If we govern too much, we move too slowly. If we move too fast, we risk not governing properly,” said one IT leader, capturing the essence of the dilemma. Another observed that the more governance hinders delivery, the more individuals and teams will find workarounds. Shadow AI is born out of this tendency.
Despite the potential friction between governance and progress, there was an emerging consensus around the table that the former was not necessarily the enemy of the latter. Indeed, governance provides the structure in which smart organisations will find solutions. In effect, governance sets the boundaries and the rules of engagement.
Best practice dos and don’ts
The evening ended with a call for advice. Attendees were asked to reflect on some of the important lessons they have learnt in overseeing AI projects. A list of dos and don’ts emerged, including:
- Don’t jump straight into AI technology. Spend time considering how you will harness the skills and capabilities of your people – and how you will take them with you on the journey.
- Do be brave enough to refine your business process. It is a mistake to simply apply AI to existing ways of working. Use AI to reinvent.
- Don’t worry if the use case you are building is not 100% compliant at the outset. Get building and address the remaining compliance issues as you go.
- Do try and solve the problems of AI with AI. AI agents make useful sounding boards to test the merits of your projects. Be prepared to be “grilled” by your AI.
- Don’t forget the people in the process. Be cognisant of their needs and behaviours to inform how you approach a project.
- Do “chase the ‘no’”. In other words, be open enough to make governance part of your solution, not a blocker.
- Don’t automatically trust the quality of your data. Spend time shaping it before applying it to AI models. Remember: garbage in, garbage out.
‘From AI promise to business impact: building future-ready enterprise AI’ – a Tech Monitor / AMD executive roundtable discussion – took place on Thursday 19 March 2026 at the Nimb Hotel, Copenhagen, Denmark.
