Hero imageMobile Hero image

November 20, 2026

Why do so many AI projects fail to move from pilot to production? Kim Berg, Sogeti’s Global CTO for Data and AI, works with organizations tackling this exact transition: moving AI from experimentation into scalable, production-ready solutions. According to him the biggest barrier to scaling AI is rarely the AI model itself. More often, organizations lack the data readiness, governance and operational integration required to deploy AI at scale. The result is what many organizations experience as AI pilot purgatory: promising proof-of-concepts that never translate into measurable business impact.

Why do AI projects get stuck in pilot purgatory?

This “pilot purgatory” isn’t unique to your organization. Various industry surveys suggest that more than 80% of AI initiatives never progress beyond the pilot stage. It’s not usually due to a shortage of good ideas or smart data scientists – they can build impressive models. Instead, pilots stall because the business environment and data landscape are not ready for production. In other words, the roadblock to scaling AI is rarely the algorithm itself; it’s everything around it. The data might be siloed or of uncertain quality. Governance might be an afterthought, making leaders uneasy about risk. Workflows might not exist to integrate the AI into day-to-day operations, leaving even a good pilot isolated in a lab. The opportunity, then, lies in tackling these underlying issues head-on to unlock real progress.

How to scale AI from pilot to production

For me, the turning point often comes when organizations stop treating AI pilots as isolated experiments and re-tool their data foundations for scale. There’s usually an “aha” moment: the team realizes that to move faster later, they have to lay some groundwork now. Here are three shifts I focus on:

  • Data made accessible and consistent: Break down data silos so everyone works from one reliable source of truth. When each AI project draws from the same, well-governed data, there’s less rework and fewer nasty surprises later. Up-front investments in data integration mean future AI initiatives start on solid ground instead of shaky, ad-hoc datasets.
  • Governance and security baked in: Build clear data ownership, privacy safeguards, and quality checks into your data pipelines from the start. By embedding compliance and security, you create a controlled environment that executives can trust. When a CFO knows that every algorithm is fed with verified, compliant data, they’re far more comfortable green-lighting a broader deployment.
  • Integration with real workflows: Design each pilot as if it’s already part of the business process – not a tech experiment in a bubble. That means automating data flows and linking into existing systems sooner than later, and involving actual users early on. If the AI pilot is developed within real operational conditions, then when it succeeds, it can seamlessly transition into day-to-day use without massive rework.

These changes might not be as glamorous as a new AI model, but they are game-changers. For example, one manufacturing team I worked with decided to consolidate decades of scattered production data into a single modern data platform. We coupled this with data quality checks and access controls. The result? Within months, we deployed an AI-powered assistant that drastically reduced troubleshooting time – potentially saving thousands of downtime hours. More importantly, that same data environment is now powering multiple AI solutions. Once we got the first one to production, the next few use cases ramped up much faster because the hard groundwork was already done.

This story reflects how we at Sogeti approach AI transformations in our AI in Action program. We emphasize “shifting left” – solving data and governance challenges early – so that AI solutions can plug into a robust core. Our Plug & Deploy AI Starter Packs encapsulate that philosophy: they offer pre-built, modular solutions equipped with integrated data pipelines and governance frameworks (think security, compliance, data management all preconfigured). That way, organizations can move from stuck pilots to scaled AI quickly and safely. Likewise, our AI Collaboration-as-a-Service model teams our experts directly with the client’s team, helping to align stakeholders and embed AI into actual workflows from day one. In short, the focus is always on building the right environment first, so pilots can seamlessly convert to full-scale deployments.

Why data readiness is critical for scaling AI

When data readiness becomes a priority, the changes are dramatic. Those AI pilots that once stalled out in isolation start to gain momentum. I’ve seen organizations go from one or two half-abandoned experiments to a growing portfolio of AI solutions delivering real benefits. Deployment cycles shrink from months to weeks. Models produce results that are more reliable, because they’re drawing on high-quality, up-to-date data. And trust increases – both human trust in the AI (since it’s governed and transparent) and the AI’s “trust” in its data (since the inputs are consistent and well-managed).

Crucially, alignment unlocks repeatable success. Instead of a one-hit wonder project, companies build an AI assembly line for innovation. Each new use case can reuse data pipelines, compliance processes, and integration patterns. The cumulative effect is bigger than any single pilot – it’s a faster learning curve for the organization and a compounding ROI on your AI investments. According to Capgemini Research Institute, companies that successfully scale AI often see significantly higher returns (on average around 1.7 times the ROI of those still dabbling in pilots). For CFOs and transformation leaders, that’s a sign of real bottom-line impact, not just tech experimentation.

How to move AI pilots into production

So how do you finally escape pilot purgatory? In my experience, the difference between endless experiments and our Plug & Deploy AI Starter Packs and AI Collaboration-as-a-Service often comes down to making that behind-the-scenes investment in data and operational readiness. It may not be as exciting as a new algorithm, but it’s the crucial step in turning promising pilots into tangible business outcomes. If you focus on the foundations now – accessible data, built-in governance, integrated processes – you’ll dramatically increase your chances of seeing your AI projects take flight and deliver real value. In short: don’t scale your AI initiatives until you’ve scaled your data and your processes. Do that, and you’ll marvel at how quickly those “stuck” pilots become success stories.

Kim Berg

Kim Berg

Global CTO for Data & AI Engineering (DAE)

Read more articles

Building governed Agentic squads for enterprise delivery

Discover how Agentic Squads help organizations balance AI autonomy, governance, and human oversight to accelerate software delivery while maintaining quality, security, and control.

From AI-assisted engineering to context-driven delivery

AI coding assistants are only the first step. Learn how Agentic Squads, AI-native engineering, and context-driven delivery are enabling governed, scalable software engineering at enterprise scale.

The model changed while you were congratulating yourself

One contradiction sits at the center of assessing anything intelligent: you judge its future behavior on the only evidence you have, how it behaved on the prompts you happened to try. With ordinary software that gap is bearable, because the code sits still. An AI system refuses to sit still, and that changes what a passing assessment is worth.