From Token Binges to Tangible Returns: The New Discipline of AI in Enterprise ERP

Hiroshi Ozaki connects SAP’s operating signals to executive decisions

I have been through enough technology cycles to recognize a pivotal moment. In the late 1990s, ERP projects were justified by the mere promise of integration. Then we learned to ask: integration for what? We asked again during the first wave of cloud migration, and we are asking now as enterprise AI moves from playgrounds to the core of ERP operations. The bill has arrived. CFOs are no longer funding AI based on token volume or adoption dashboards; they are demanding measurable improvement in actual workflows. That shift will reverberate through SAP’s customer base, its partner ecosystem, and the company’s own AI monetization strategy.

The Business Signal

For the past eighteen months, many enterprise AI initiatives inside SAP landscapes followed a familiar pattern: deploy an AI copilot, measure success by tokens consumed or queries answered, and hope that productivity would magically appear. That phase is ending. Finance leaders are tightening budgets and requiring that AI deployments be tied to outcome-based KPIs—reduction in manual effort, shorter order-to-cash cycles, fewer data errors that trigger costly corrections. In other words, the unit of value is no longer “tokens processed” but “cost taken out of a business process.”

This is not merely a customer-side discipline; it is a direct challenge to SAP’s AI monetization model. SAP currently delivers AI capabilities—from Joule, its copilot, to various embedded AI services—through a combination of consumption-based (token) pricing and premium subscription tiers. While token pricing can work for variable workloads, it incentivizes volume, not value. If customers start demanding that SAP’s AI services be linked to contractually measurable outcomes, the sales motion becomes far more complex. The company must then demonstrate not just that its AI models work, but that they consistently deliver business results inside messy, customized ERP environments. That is a fundamentally different bar.

The signal for investors is this: SAP’s ability to transition from volume-based AI incentives to transparent outcome documentation will materially influence AI revenue growth and renewal confidence. If the company continues to report only token consumption or “user adoption” without connecting AI to key process metrics, savvy CFOs will delay or defund projects. Conversely, if SAP embraces outcome-based success stories—backed by auditable metrics—it can justify premium pricing and deepen S/4HANA migration incentives. The caveat is that outcome measurement remains difficult. Every customer’s definition of “improved order processing time” varies, and ERP environments are notoriously heterogenous. Attempting to standardize outcome metrics could overpromise and underdeliver, creating contractual disputes.

What It Means for SAP Customers

I have often reminded clients that technology adoption is never just a technical act. It is an organizational change. The same holds true for AI in ERP. In my decades advising Japanese manufacturers, we learned that chasing the latest machine without redefining the work itself leads to what my colleagues called “clean, shiny waste.” For AI projects, the risk is similar: many organizations are running token binging without any Kaizen—no continuous improvement of the underlying process.

To escape this trap, I recommend several practical steps that align with the new fiscal reality:

• Replace volume KPIs with outcome KPIs. Instead of measuring “queries resolved by Joule,” measure “reduction in manual order-entry hours” or “percentage of invoices matched without human intervention.” These are the metrics a CFO will fund.
• Implement model routing strategies. Not all AI tasks justify the most expensive, most capable model. Routine tasks like classification of purchase requisitions can use smaller, cost-effective models; reserve high-cost models for complex, high-value analytics. This alone can cut token spend by 30–50% in ERP workflows.
• Audit AI token consumption across ERP workflows. Map every API call, identify redundancies (such as multiple models called for the same data extraction), and refine prompt engineering. One client discovered that 40% of its token usage came from context windows stuffed with unnecessary historical data—simply trimming them reduced cost without affecting accuracy.
• Build a financial business case before scaling. Link AI initiatives to measurable operational improvements—faster delivery confirmation, higher first-pass yield in quality inspections, fewer

References

• AI’s Token Binge Is Over—Now Enterprise Budgets Want Proof
• AI’s Token Binge Is Over—Now Enterprise Budgets Want Proof
• SAP AI Core Documentation

References

• SAP AI Core Documentation
• AI’s Token Binge Is Over—Now Enterprise Budgets Want Proof
• AI’s Token Binge Is Over—Now Enterprise Budgets Want Proof