How AI Transforms Enterprise Planning & Decision-Making

AI has entered a phase of practical adoption, influencing how organizations make decisions, structure workflows, and run their operations. To bring customers and partners together around the practical realities of this AI transformation, we hosted our Let’s Talk AI event, an in-person knowledge-sharing gathering at our headquarters in Zoetermeer.

The event focused on how AI is being applied in planning and decision-making, where automation and AI-enabled workflows are delivering measurable value, and how enterprises can move from experimentation to large-scale impact. With contributions from ORTEC, Microsoft, and Stolt Tankers, the event highlighted both strategic perspectives and real-world use cases now emerging across industries.

Erica D’Acunto (VP Innovation, ORTEC) opened the event by outlining the current state of AI adoption. Drawing on recent market research, she highlighted the growing scale and maturity of experimentation across industries.

Generative AI: Moving Beyond Assistance to Workflow Transformation

“80% of organizations are exploring or piloting AI technologies,” she said, noting that much of this activity focuses on generative models used for productivity tasks. According to the same report, 64% of organizations already see these tools accelerating innovation, primarily because they enable faster iterations and make new ideas feasible.

She added that experimentation is no longer limited to individual use. “62% of organizations are automating parts of entire workflows using generative AI. That shift marks the step from assistance to true process transformation.”

At the same time, she noted that organizations increasingly debate the return on AI investment. “The companies that achieve returns are not only looking at efficiency, they are integrating AI into products and services to differentiate, not merely optimize.” 

The first keynote came from Sander Vlot (VP Center of Excellence, ORTEC Data Science & Consulting), who discussed AI’s role in planning and decision-making processes across industries.

He began by acknowledging the simultaneous enthusiasm and frustration surrounding AI. “There is enormous investment and interest, but 95% of organizations still see no return on their AI projects.” The task, he argued, is to understand what the remaining 5% are doing differently.

For Vlot, the starting point is clear: value emerges when organizations automate the structured, repeatable work that planners perform every day. These tasks include extracting information from emails, interpreting supplier updates, digitizing documents, forecasting demand, running scenario analyses, and shaping recurring reports.

“These activities are rational, predictable, and carried out continuously. They are ideal candidates for automation.”

He encouraged organizations to think in terms of cognitive capabilities rather than technologies. Large language models excel at interpreting unstructured inputs such as emails, documents, and PDFs, while forecasting and optimization often require traditional machine learning or mathematical optimization tools.

Vlot also addressed the widely discussed notion of AI agents. While multi-agent systems can autonomously pursue goals, they are complex, difficult to control, and rarely necessary in early adoption stages. “For most organizations, the first ‘AI agent’ should be an AI-enabled workflow, not a fully autonomous system.”

He concluded with a reminder that adoption is not just technical. “AI requires a mindset shift. As more planning activities become automated, planners will increasingly supervise, adjust, and improve automated policies rather than manually create them. That evolution takes time.”

The next speaker, Francesco Tumiatti (Cloud and AI Specialist, Microsoft), explored how organizations progress from experimenting with generative AI to deploying AI agents that handle substantial portions of operational workflows.

Tumiatti introduced the concept of a “frontier firm”, organizations that combine human judgment with AI-driven execution. In these firms, employees interact daily with AI systems and delegate tasks to specialized agents.

He outlined three phases of AI adoption:

• Individual productivity, through tools such as Copilot or ChatGPT
• Task delegation, where AI agents perform defined activities such as research, drafting, or validation
• Process automation, where teams of AI agents coordinate end-to-end workflows

To illustrate, he shared a case study of a consumer goods company automating its order-to-cash process. The organization mapped its workflow, identified the required capabilities, and assigned responsibilities to different agents, each powered by an appropriate model.

“Agents used small, fine-tuned models for customer interaction, and reasoning-focused models for complex validation steps. Enterprise systems imposed the business rules, preventing invalid actions.”

The key barrier, he stressed, is not technology. “Models are commodity. The real challenge is selecting the right process, breaking it into jobs agents can perform, and ensuring enterprise readiness such as telemetry, monitoring, and governance.” Without that maturity, proof-of-concepts rarely reach production. 

How to Identify “Agent-Ready” Work

Francesco Tumiatti highlighted that organizations can begin working with AI agents by selecting small, well-defined tasks that are suitable for automation. A task is “agent-ready” when:

• It follows a clear, repeatable sequence of steps
• Decisions can be made using existing data or business rules
• Exceptions are easy to detect and escalate to humans

His recommendation: start with one contained activity, automate it with proper monitoring and guardrails, and only expand once the first slice performs reliably. This step-by-step approach helps organizations build confidence while safely scaling agent-driven workflows. 

The most operational case study came from Mohit Kumar (Global Director Business Technology and Innovation, Stolt Tankers), who explained how Stolt is using historical data to predict and reduce safety incidents across its global fleet.

Stolt operates roughly 160 vessels worldwide, many transporting hazardous chemical cargo. The core question: can data reveal which voyages carry higher risk and can those insights improve safety?

“70% of maritime accidents relate to human factors,” Kumar said. Stolt therefore sought to analyze patterns that consistently correlate with safety outcomes.

Several strong predictors emerged:

• Audit frequency: fewer audits correlated with significantly higher incident rates
• Cargo complexity: voyages with many cargo types had higher incident probability
• Vessel age: older ships experienced more operational events
• Crew composition: diversity and familiarity were decisive

Kumar elaborated: “Our data seems to suggest that if the captain is not part of the majority nationality, or if crews have worked together only a short time, incident rates seem to rise measurably.”

Stolt trained a machine learning model using seven years of data and validated it on recent voyages. The model proved strongly predictive. “Roughly 90% of high-risk voyages had indeed experienced events. And around 65% of accidents fell into categories the model flagged beforehand.”

The company is now working to integrate these insights operationally. While some risk factors such as ship class or cargo complexity cannot be changed, others can be adjusted.

After the keynotes, the speakers joined a panel moderated by Chuck Ng, Principal Consultant at ORTEC Data Science & Consulting. The conversation revolved around three central questions: how to separate hype from real value, where AI is already delivering measurable impact, and which barriers organizations encounter most often.

The panel agreed that the distinction between hype and value starts with a simple filter: does the technology meaningfully improve a process, a decision or a customer outcome? If not, it is hype.
Sander Vlot added that many challenges can be solved without sophisticated AI at all. “If better UX solves the problem, use better UX. Not every interaction requires a chatbot.”

Francesco Tumiatti emphasized that alignment with business goals is essential. “If you cannot connect technology to a measurable business result, it is not value, no matter how advanced it looks.” 

The panel highlighted several examples where organizations are already seeing tangible gains: productivity improvements through Copilot, faster design sprints using generative AI, stronger tender responses and utilization improvements at Stolt, and everyday efficiency benefits such as automated meeting summaries and documentation.

The final part of the discussion focused on barriers to adoption. Mohit Kumar pointed to culture and literacy: “If you automate tasks close to what people already understand, they adopt quickly. When you transform the process itself, change becomes harder.”

Vlot noted that many organizations struggle simply to identify routine work. “People only start seeing automation opportunities once they work with AI tools themselves. Experimentation builds awareness.”
Tumiatti added that experimentation alone is not enough. “Bottom-up experimentation is valuable, but without a unified AI strategy, prototypes rarely move to production.”

The conversations during Let’s Talk AI made one point unmistakably clear: value doesn’t emerge from bigger models or more features, but from reimagining how decisions are made and how work flows through the enterprise. The organizations that take this seriously now will define the next decade of operational performance.