AI has become a strategic priority for organisations across the region, with many racing to be early adopters. Yet being first does not always equate to success. Where do organisations most commonly struggle when moving from proof of concept to tangible business value?
Enterprise AI adoption today is largely defined by the rush to be an early adopter. Many organisations want to demonstrate progress — often to their boards — by showing that they are “doing AI,” and that instinct is entirely understandable.
The challenge is that proof-of-concept initiatives are built to demonstrate possibility, not to support production-grade systems that organisations can depend on over time. Once the focus shifts from experimentation to deployment, an entirely different set of challenges emerges.
Production systems require deep, reliable connections to enterprise data rather than limited extracts. They must be closely aligned with how the business actually operates. Once deployed, these systems need continuous monitoring at scale, with transparency and traceability built in. When issues arise, organisations must be able to understand exactly how the system behaved in order to address root causes, rather than applying superficial fixes.
Many organisations struggle because production deployment demands a fundamentally different mindset, approach, and toolset compared with pilots or POCs.
Transparency is often cited as a critical success factor for enterprise AI. How important is it to design trust into AI systems from the outset?
It is essential. Trust at the level required to deploy AI at scale must be designed into systems from the beginning. Governance and oversight cannot be treated as an afterthought.
Trust depends on how data is selected, how models are chosen, and how those decisions are reviewed and validated. When governance is embedded into the design process, organisations can trust not only individual outputs, but the entire system used to design, develop, and deploy AI.
At Dataiku, our focus is on ensuring that the system itself — effectively the factory that produces AI solutions — is trustworthy, so the outcomes it generates can be trusted as well.
As governance and compliance requirements intensify globally, how does Dataiku help organisations manage this growing complexity without slowing innovation?
Organisations face significant pressure to move quickly while also dealing with uncertainty around evolving regulatory frameworks that are often fragmented across regions.
Dataiku provides built-in governance across the full AI lifecycle, while allowing organisations to tailor those processes to their specific needs. For example, organisations subject to the EU AI Act must classify applications based on risk. Dataiku enables customers to apply that framework in a way that reflects their region, use case, and industry.
Rather than enforcing a rigid, one-size-fits-all approach, we provide the tools organisations need to build governance systems that scale efficiently and work within their operational reality.
How should organisations think about balancing automation with human oversight as AI systems become more capable?
There is no single answer. It is best viewed as a sliding scale between fully manual and fully automated processes. Where that balance sits depends on the organisation, the application, and even the point in time.
Most organisations begin with a high degree of human oversight. As trust builds through monitoring and governance, systems are gradually allowed to operate more autonomously. This is how enterprises move towards more autonomous AI — trust must be earned over time.
It mirrors how autonomy develops with junior employees: responsibility increases as consistency, reliability, and competence are demonstrated.
Is there an aspect of AI that you believe is overhyped — and something that is not receiving enough attention?
There is a widespread misconception about the capabilities of AI models, often driven by messaging from AI labs themselves. While these models are undeniably powerful, there is an overemphasis on the idea of imminent superintelligence.
That narrative can be dangerous. It encourages organisations to wait for the next model rather than addressing foundational challenges around data access, governance, and system design today. Organisations that adopt a “wait and see” approach risk falling behind competitors who are doing the hard work now.
What is discussed far less is how different forms of intelligence — including agentic AI, predictive models, statistical intelligence, and human expertise — must work together. An excessive focus on large language models alone risks leaving significant value untapped.
Looking ahead, what should enterprises expect from the AI landscape over the next year — and where will Dataiku focus its efforts?
I expect 2026 to be the year in which AI agents mature and begin delivering real business value across complex, mission-critical processes.
That is where Dataiku is focused — working closely with customers and partners to build agent-based systems that can operate reliably at scale. Our product roadmap is aligned with providing organisations the capabilities they need to deploy agents across their most important business processes.
Discussion about this post