According to SAP, enterprise AI governance ensures profit margins by replacing statistical guesswork with deterministic controls.
When you ask a consumer model to count words in a document, it can often be off by as much as 10%. Manos Raptopoulos, global president of Customer Success Europe, APAC, Middle East and Africa at SAP, says there is an absolute operational gap between near-perfect and perfect.
“The distance between 90% and 100% accuracy is not incremental. In our world, it’s existential,” Raptopoulos said.
Raptopoulos emphasizes that as organizations bring large-scale language models into production, metrics are formally shifting toward accuracy, governance, scalability, and tangible business impact.
The pressing challenges facing corporate boards center on their evolution from passive tools to active digital actors, a transition Raptopoulos identifies as a key governance moment and will be one of the topics SAP will focus on at this year’s AI & Big Data Expo North America.
Agentic AI systems have the ability to plan, reason, coordinate with other agents, and autonomously execute workflows. Because these systems interact directly with sensitive data and influence decision-making at scale, Raptopoulos argues that failing to manage these systems exactly the same way you manage your human workforce exposes your organization to serious operational risks. He warns that agent sprawl will mirror the shadow IT crisis of the past decade, but the stakes are much higher.
His framework calls for establishing agent lifecycle management, defining autonomy boundaries, enforcing policies, and implementing continuous performance monitoring.
Integrating modern vector databases (that map the semantic relationships of enterprise languages) with traditional relational architectures requires significant engineering capital. Teams must proactively limit agent reasoning loops to prevent hallucinations from disrupting financial or supply chain execution paths. Setting these strict parameters increases the computational latency and computational cost of the hyperscaler, changing the initial profit and loss forecast.
If an autonomous model requires constant and frequent database queries to maintain a deterministic output, the associated token cost increases rapidly. Governance becomes a hard engineering constraint, not a compliance checklist.
Raptopoulos argues that corporate boards need to solve three fundamental problems before deploying agent models: determining who is responsible for agent errors, establishing an audit trail of machine decisions, and defining precise thresholds for human escalation. Geopolitical fragmentation makes these questions difficult to answer.
Sovereign cloud infrastructure, AI models, and data localization mandates are regulatory realities in key markets spanning New York, Frankfurt, Riyadh, and Singapore. Enterprises need to embed deterministic control directly into probabilistic intelligence. Raptopoulos views this requirement as an executive remit, not an IT project.
Building relational intelligence for commercial operations
AI systems are still completely dependent on the quality of data and the processes by which they operate, representing what Raptopoulos calls a data-founded moment.
Fragmented master data, siled business systems, and overly customized ERP environments create dangerous unpredictability at the worst moments. Raptopoulos explains that when autonomous agents rely on a fragmented infrastructure to provide recommendations that impact cash flow, customer relationships, and compliance positions, the resulting operational damage quickly mounts.
To unlock tangible enterprise value, we need to evolve beyond common large-scale language models trained on Internet-scale text. True enterprise intelligence, outlined by Raptopoulos, must be based on unique enterprise data such as orders, invoices, supply chain records, and financial postings that are embedded directly into business processes. He argues that relational-based models specifically optimized for structured business data consistently outperform general-purpose models in prediction, anomaly detection, and operational optimization.
Many implementations stall due to significant operational friction in making an overly customized ERP environment understandable to the underlying model. Data engineering teams spend excessive cycles sanitizing fragmented master data just to create a baseline for AI to ingest.
If your relational model needs to accurately interpret complex, unique supply chain records along with raw invoice data, the underlying data pipeline must operate with zero latency. When data ingestion fails, the model’s predictive ability immediately degrades, making the agent’s functionality dangerous to your business.
Integrating legacy architectures with modern relational AI requires an overhaul of deeply ingrained data pipelines. Engineering teams are faced with indexing decades of non-sensitive planning data so that embedded models can generate accurate vector representations. Following Raptopoulos’ logic, boards should assess whether their current data assets are truly ready, rather than simply layering probabilistic intelligence on top of a disparate foundation.
Designing an intent-based interface
Enterprise application interactions are moving from static interfaces to generative user experiences, and Raptopoulos developers flag this as the moment for employee interaction.
Instead of manually navigating a complex software ecosystem, employees express their intentions to the system. Raptopoulos gives an example of a user instructing the software to prepare an information session for the week’s most profitable customer visit. The AI agent then coordinates the required workflow, assembles the surrounding context, and uncovers recommended actions.
But Raptopoulos emphasizes that adoption among employees is still contingent on trust. Employees will embrace these digital teammates only if they can be confident that the system’s output respects established governance boundaries, reflects authentic business rules, and provides measurable productivity gains.
Designing these systems requires role-specific AI personas tailored to roles such as CFOs, CHROs, and heads of supply chains. Raptopoulos observes that to successfully close the adoption gap, these personas need to be built on trusted data and embedded into familiar corporate workflows.
Achieving this level of integration is a design decision with significant consequences. Organizations that want to invest capital in AI-native architectures see an accelerated return on investment, but companies that look to bolt probabilistic models into legacy interfaces struggle significantly with reliability, ease of use, and scale.
Technology leaders seeking to force modern AI orchestration into monolithic software applications often encounter significant integration delays. Routing probabilistic API calls through older enterprise middleware introduces lag in the user interface and breaks intent-based workflows. Designing role-specific personas requires more than quick engineering. Complex access controls, permissions, and business logic must be mapped into the model’s active memory.
Engineering competitive defenses
The financial benefits of AI will be seen earliest during customer interactions. Raptopoulos points out that training models based on proprietary records, internal rules, and historical logs creates a layer of customer-specific intelligence that competitors can’t easily copy. This setting is most effective for workflows with many exceptions, such as dispute resolution, claims, returns, and service routing.
Deploying autonomous agents that can classify cases, uncover relevant documents, and recommend policy-aligned solutions turns these costly processes into clear competitive differentiation.
These models adapt based on the outcome of each interaction. Raptopoulos points out that corporate buyers prioritize reliable, relevant, and responsive service over technological gimmicks. Companies deploying AI to handle heavy workloads while maintaining strict oversight of the final output build barriers to entry that general-purpose tools cannot penetrate.
Deploying enterprise intelligence requires executives to coordinate three different layers in parallel, which Raptopoulos defines as a strategic moment.
The first layer includes built-in functionality that integrates persona-driven productivity directly into your core applications for quick revenue. The second layer requires agent orchestration, which facilitates multi-agent coordination across workflows between systems. The final layer focuses on industry-specific intelligence and features highly specialized applications co-developed to address the most high-value challenges specific to specific sectors.
A trap awaits leaders who fall victim to incorrect ordering. Focusing solely on built-in tools leaves huge financial value untapped, while increasing enterprise risk by aggressively jumping into deep industry applications without first achieving proper governance and data maturity.
Raptopoulos advises that scaling these models requires matching a company’s ambitions with its actual technical readiness. Leadership teams must fund a clean core architecture, update data pipelines, and enforce cross-functional ownership to get through the pilot phase. The most profitable deployments treat AI as a central operating layer that requires the same governance as human staff.
The financial gap between 90% accuracy and complete certainty determines where the true value of a company lies. Governance decisions made in the coming months will determine whether certain AI deployments become powerful sources of lasting benefits or costly lessons learned.
See: Governance of AI agents comes into focus as regulators flag control gaps
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expos in Amsterdam, California, and London. This comprehensive event is part of TechEx and co-located with other major technology events such as Cyber Security & Cloud Expo. Click here for more information.
AI News is brought to you by TechForge Media. Learn about other upcoming enterprise technology events and webinars.
