AI in production is the only finish line that matters

It has been roughly a year since most large enterprise organizations formulated their first AI strategy. Pilots have been run. Demos have been shown to leadership teams. But when we asked Risto what is different in the market right now compared to six months ago, his answer cut straight to it.

"The question has changed. Before, it was: what can we do with AI? Now it's: what have we actually built, and what value is it delivering to the business?"

The shift in tone is evident. What has not shifted is the underlying problem.

The gap between ambition and reality

Risto describes it as a "vending machine problem". Enterprise organizations have AI ideas in abundance. What they lack is a standardized function for turning those ideas into solutions that can be maintained and scaled.

"They are doing things today, but in completely different ways, by different teams, with no common standard. That creates value on paper but not in production."

The result, when that foundation is missing, is a portfolio of AI solutions that nobody truly owns and that never reach the business impact that was the whole point.

It is something Risto sees play out across industries, but particularly in financial services, where regulatory requirements add a layer of complexity that makes the lack of standardization even more costly.  

“We are working with an FSI company on exactly this —building a self-service AI platform that gives their teams governed access to pre-approved AI services, with a clear path from experimentation all the way to isolated, production-ready environments.”

‍

‍

The cloud pattern is repeating itself

Risto keeps coming back to one parallel: cloud. Back then, people talked about shadow cloud. It was happening, but nobody had full visibility. Today, he sees shadow AI playing out the same way. Solutions built in silos. Data not centralized. AI policies that exist on paper but only cover tooling, not the organization as a whole.

"Nobody had visibility into the cloud bill until it arrived. With AI, the risk is the same, and it is even harder to see coming, because the cost is embedded in how you build, not just how much you run."

FinOps (Financial Operations) for AI needs to be owned by the business, not delegated to IT and forgotten. Someone has to be accountable for it before the costs make the conversation unavoidable.

The data control question is surfacing in other ways too. Sovereign cloud is coming up in more enterprise conversations, with both AWS and Microsoft now offering solutions aimed at keeping data within EU borders. Risto sees genuine interest, but also a reality check: moving existing infrastructure is enormously costly, and sovereign strategies realistically only work for new builds. Some organizations are even going back to on-premise.

What Risto would prioritize as CIO

We asked him directly: if you were sitting in the CIO or Head of AI chair today, what would you tackle first?

1. Appoint AI owner(s). Who owns AI, business or IT? A dedicated function or a virtual one? Without that ownership map, everything else is suboptimal.
2. Get the data foundation in order. Not necessarily perfect, but in order.
3. Establish a clear tooling policy. Claude, Copilot or Gemini? The choice changes over time, therefore the policy is even more important.
4. Identify 2–3 AI use cases and build them into production. Get solutions into production fast but organized, start with smaller scope and then scale in phases.  
5. Establish an AI competence baseline across the organization. Expect everyone to use AI in their daily work, develop, learn and share internally.
6. Give AI cost control an owner from day one. Nobody had visibility into the cloud bill until it arrived. Do not repeat the same mistake with AI.

Agents: the vision exists, reality takes time

Every conversation with Risto eventually lands on agentic AI. He takes the vision seriously; AI agents that are managing operations, presenting reports, and making decisions, but he is equally clear about what works right now.

“At a manufacturing customer we work closely with, agentic workflows are being run in phases, one slice at a time, with clear checkpoints along the way. That is deliberate. You still need a human in the loop, and the trust in these systems has to be earned before autonomy can expand.”

Risto expects AI validation to become a major focus area in the near term, making sure outputs are correct, that systems do not hallucinate, that accountability is built in from the start.  

The pace at which things are being built right now is extreme. Risto's point is not to slow down. It is to build with clear ownership, a solid foundation, and production as the only finish line.