From the Editor’s Desk:
“I enjoyed co-authoring this piece with Christian Klein, CEO of SAP, and appreciated his candid admission that he had overestimated what AI could do on its own. But it’s the argument that follows that has stayed with me. Every business depends on judgment people develop through years of doing the work, much of which never makes it into a document or a system. Once AI starts learning from that experience, ownership becomes a practical concern. Companies need to know who will control the intelligence built from the knowledge they spent years developing. What follows is his perspective. I hope you enjoy this one!” – Victor Dey
Over the past year, my conversations with CEOs about AI have become more focused on business results. A year ago, many wanted to know which large language model to choose and whether they were investing enough. Today, they ask why those investments still haven’t shown up on the profit and loss (P&L) statement.
That pressure for results also raises the stakes for trust. As companies give AI a greater role in decisions about customers and suppliers, leaders need confidence that it understands the consequences of its recommendations. They need to know where it can act independently and where human judgment remains essential. Meanwhile, frontier models continue to improve, and companies are already using open-weight models for production workloads. Access to capable AI is becoming something competitors can readily buy. Building an advantage depends on how effectively a company connects that technology to its own knowledge and experience. The value comes from using that context to make better decisions and deliver results the company can measure and trust.
The companies that win won’t necessarily have better models. They’ll be the ones whose AI delivers outcomes they can trust. Early on, I overestimated the impact AI alone would have on a business. A chatbot is great, but it doesn’t translate into real value for a company on its own. The differentiator is context.
AI needs to understand the business it operates in: its processes, the regulations it must comply with, the decisions people make and why they make them, and the consequences that follow. When money moves, when a supplier gets onboarded, when payroll runs, a good answer is not enough. You need the right answer every single time, compliant and traceable. For decades, companies built their advantage through deep expertise and institutional knowledge accumulated over years of running the business. What catches a lot of executives off guard today is that these are the assets AI has not disrupted. AI has made those assets more valuable. Business context becomes the scarce asset when the underlying technology is widely available.
Connecting AI Decisions to Business Outcomes
A business doesn’t win because someone writes a cleaner email or summarizes a meeting more quickly. A business wins or loses through thousands of connected decisions. A shipment from India misses its window by ten days, throwing inventory off across three regional warehouses and forcing a pricing change on a line that was supposed to carry margin this quarter. That pricing change breaks a volume commitment made to a retail partner six weeks earlier. None of it appears as one dramatic error. The consequences accumulate.
A model can understand language. It doesn’t automatically understand how one supplier delay creates inventory pressure, a broken customer commitment and a margin problem that affects financial performance. For AI to participate meaningfully, some representation of the business must become explicit. Technologists call it an ontology: the relationships between customers, suppliers, products, assets, obligations and policies that shape decisions every day. Translating that structure into something AI can reason about is where the hard work sits.
The companies that recover fastest are often those where people closest to the work recognize a change before anyone else does. AI creates value by understanding the chain reaction and helping the organization act while there is still time to change the outcome.
The most successful AI deployments create value by knowing enough about how a business works to act on its behalf.
Aeropuertos Argentina offers an example. It connected weather, runway, maintenance and operational data to coordinate winter operations across its Patagonian airports. The reported results included improved runway safety, a 16% reduction in direct costs and a 90% reduction in administrative effort. The value came from connecting signals across the operation and helping people act earlier and more effectively.
Those reported improvements must also be considered alongside the ongoing cost of running the system. AI should meet the same test as any strategic investment: fewer disruptions, faster decisions, lower operating costs, stronger customer outcomes and better financial performance. If the gains don’t exceed the operating costs, the value hasn’t reached the P&L. That’s a higher bar than most AI pilots and agents are currently being held to. Our industry needs to be more honest about that.
Bringing Human Judgment Into AI Decisions
Much ordinary business judgment never makes it into a policy document. An experienced planner knows when to add inventory before a disruption appears in the dashboards. A procurement lead knows which suppliers usually recover quickly and which tend to sound more reliable than they are.
That judgment becomes more valuable as AI enters the picture. Planners don’t know more facts than a model does. Experienced operators recognize when a disruption is routine and when it signals that something structural is changing underneath the business. No report fully captures that distinction. The organizations getting this right put human instinct in conversation with AI, so each catches what the other misses.
When the planner and the AI disagree, neither seniority nor the machine should automatically prevail. The organization needs to understand what each sees that the other doesn’t. Operators recognize signals that haven’t been formalized in data. They know which supplier’s reassurance has held up because they’ve lived through enough cycles to judge it. AI identifies patterns across thousands of decisions that no individual could see. Neither perspective is complete.
AI systems must show the reasoning, data and context behind a recommendation. At SAP, that is a requirement. Experts should be able to challenge that reasoning with information the system doesn’t yet understand. The organization should capture what led to the decision and what outcome followed. The companies getting this right aren’t asking humans and AI to compete for the last word. Disagreement then becomes a source of learning, allowing human judgment and AI to continuously challenge and improve each other.
Turning Business Exceptions Into Governed Learning
That is why AI doesn’t become valuable simply because it is deployed. It becomes valuable when it operates inside the decisions that define how the business actually runs.
A company’s pricing strategy doesn’t live in a single document. It emerges through thousands of decisions over time. The same is true of supplier relationships, inventory tradeoffs, customer exceptions and risk management. Every forecast refined, supplier selected or exception approved adds another layer to how the business understands itself. These decisions become more valuable than a record of what happened. They become organizational learning.
That learning needs discipline. An exception may be a decision that worked once under unusual circumstances. When someone approves one, the system should document the decision, its rationale, the relevant context, the approver and the outcome. That record can flag a possible pattern for review. It cannot automatically turn one person’s judgment into future policy or autonomous behavior. An exception should never become a rule simply because it succeeded once.
Before AI incorporates that learning, the organization needs to understand why the decision was made, under what conditions it worked and whether the same reasoning applies elsewhere. That requires accountable process owners, clear oversight mechanisms and the ability to reverse course if the evidence changes.
Matching AI Autonomy to Business Risk
The foundation starts with unglamorous work: consolidating fragmented data systems, standardizing how information is structured and accessed, and connecting records that have accumulated in silos over years. These projects rarely make headlines. Yet a model operating on fragmented or inconsistent data reflects those failures regardless of its capabilities. A coherent foundation lets AI work with what the business actually knows, beyond whatever portion happens to be machine-readable.
Companies want agents doing meaningful work while their information remains scattered across systems from different vendors. Autonomy should increase as confidence increases, and confidence must be earned. Before allowing an agent to act independently, leaders should establish whether they trust the underlying data and whether business rules and decision logic are clearly defined. They must also understand what happens if the decision is wrong.
As AI participates in consequential decisions, governance becomes a business imperative. Recommending a supplier, approving a payment and analyzing a clinical trial place different demands on a system. The distinction between recommendations and commitments matters: recommending a supplier carries different consequences from committing company funds. As those consequences increase, so should the oversight. That’s not a limitation of AI. It’s good management.
Organizations should always know who authorized an action, why it was taken and who owns the outcome. That accountability matters when a decision crosses into another provider’s software. Humans oversee increasingly capable agents within well-defined boundaries.
Owning the Intelligence a Business Creates
Companies have spent years worrying about where their data lives. In boardrooms, the discussion has shifted over the past six months toward where the intelligence generated from that data lives. Most business knowledge was never written down. It accumulated through countless decisions, each teaching the organization something about how it actually works.
SAP contributes deep process knowledge accumulated across industries. Customers contribute the experience captured in every decision, transaction and exception across their business. Together, they create a foundation for turning that experience into intelligence. The more effectively the context is captured and applied, the greater the value created.
When learning remains inside the enterprise, embedded in its processes, data and decision-making, it becomes an asset competitors cannot simply license or replicate. Its value grows through the company’s own experience. Keeping that intelligence embedded in the business, governed by its rules and within infrastructure it controls, makes ownership a strategic consideration.
Generic intelligence stops being a differentiator once anyone can buy it. What stays scarce, and hard to copy, is enterprise intelligence grounded in how a business actually runs.
The Control Layer Verdict
As powerful AI becomes widely available, competitive advantage will increasingly come from how effectively companies turn their own experience into intelligence. The enduring value comes from ensuring that this learning compounds inside the enterprise, embedded in its processes and governed by its rules. Every decision becomes an opportunity to deepen institutional knowledge and strengthen the company’s ability to act. Companies already own their data. The next opportunity is to build and own the intelligence generated from using it: an asset shaped by their experience, developed through their operations, and governed rigorously enough to trust.
