01 Triggering Event

On November 11, 2026, Bloomberg Technology reported that SAP was intensifying its promotion of its AI tools while still withholding key metrics for determining whether those investments are generating returns.

Only two pillars of information are available: SAP is selling AI tools, and the company has not provided sufficient metrics to demonstrate the results. The original report gives no figures for ARR, AI product revenue, renewal rates, productivity gains, or cost savings. Those omissions cannot responsibly be filled with assumed growth numbers.

This is better understood as a disclosure-quality event than a product-launch event.

02 What It Really Means

SAP’s challenge is not whether to pursue AI, but whether it can turn AI from a software feature into a priceable, attributable economic outcome.

Enterprise software has historically relied on a clear monetization path built around seat counts, contract values, renewals, and implementation cycles. Once AI agents reshape workflows, value may instead come from completing tasks, resolving tickets, shortening cycle times, or reducing human intervention rather than from adding another license.

That directly challenges SAP’s legacy business model.

If AI remains an add-on feature within an ERP module, customers will continue paying by seat or subscription, but the incremental value generated by AI cannot be isolated. SAP will fail to capture an AI premium, or customers will retain the savings themselves.

If SAP intends to charge by usage, resolution, or outcome, it must build a credible measurement layer: which agent completed which task, how many labor hours it saved, whether throughput improved, and how quality and cost changed. SAP’s internal billing design cannot be confirmed from the available information. However, the disclosure gap shows that at least the externally visible measurement system remains immature.

What ultimately gets priced may not be the number of AI tools SAP has released, but who can help the CFO verify how much productivity each dollar of AI spend returns.

SAP’s distinctive advantage is that it already holds critical enterprise data: orders, inventory, supply chains, financials, human resources, and customer relationships. In theory, it is uniquely positioned to connect AI effects back to business outcomes through distribution and data advantages.

But data location alone is not a moat. Only when enterprises are willing to connect, govern, and audit that data can SAP convert its data advantage into a workflow control point. Otherwise, SAP is merely an interface layer containing AI.

03 Historical Analogies / Structural Comparisons

There is one similarity between SAP’s AI narrative and the business story surrounding the 2007 iPhone launch: demonstrating functionality is easy, but attributing benefits is difficult.

Apple showed that a phone could combine music, a browser, email, and a touch interface, but it did not prove on launch day how much value mobile computing would create. The real inflection point came later through the App Store, carrier subsidies, mobile advertising, and mobile workflows.

SAP’s AI narrative is currently at the first stage: demonstrating that AI tools can enter enterprise processes.

Apple ultimately built distribution and platform economics. If SAP wants a comparable outcome, it cannot rely on promoting the number of agents it offers. It must make AI outcomes renewable, scalable, and attributable product units.

An even closer comparison is AWS in 2014. Rather than relying on one impressive aggregate revenue figure to prove its cloud advantage, Amazon decomposed that advantage through instance types, object storage, managed services, and usage-based billing, turning infrastructure advantages into products developers could consume continuously.

SAP needs a similar decomposition. It needs to explain what customers are actually purchasing: tokens, agent actions, workflow runs, resolved tickets, or final business outcomes. Without such product units, AI economics remain bundled inside traditional software contracts.

Historically, Microsoft has often accepted the difficulty of separately attributing feature value, first bundling Copilot into Microsoft 365 and then expanding through its established distribution. SAP could follow the same route. Yet that would also mean AI cross-selling might succeed while AI pricing power remains unproven.

This judgment could be mistaken. SAP may already possess customer-level ROI data but refrain from disclosing it because of sensitivity or because the benefits are still early-stage. Without access to internal data, that possibility cannot be ruled out.

04 What It Means for AI Builders

The immediate adjustment for AI builders is not which models to select, but how to frame the language of customer purchasing.

When selling to large enterprises, AI builders should not treat accuracy, latency, and token cost as the core metrics alone. They should map those measures to business metrics: ticket-resolution time, the cost of handling invoice exceptions, inventory turnover, revenue per employee, and agent intervention rate.

Without a baseline, ROI is merely narrative. At minimum, builders should design three sets of data:

  • Pre-deployment processing time, labor cost, and error rates;
  • Post-deployment changes measured on the same basis;
  • Incremental effects after controlling for model volatility, seasonality, and sample selection.

For API consumers, this also suggests avoiding premature promises about how many employees will be replaced. Mature buyers will demand an audit trail, permission scope, human escalation, and outcome attribution. An agent’s ability to execute actions does not mean an enterprise will permit it to do so.

For independent AI entrepreneurs, however, this creates an opportunity. SAP controls access to systems but may remain constrained by legacy contract structures, complicated implementations, and attribution challenges. A startup that can deliver verifiable outcome pricing around a high-value workflow could take budget from alongside SAP’s feature catalog.

That window will not open by being “smarter than SAP.” It will open through a narrower scope, faster deployment, and more credible measurement.

Teams building token gateways should also reassess usage logging. Recording only models, tokens, and latency is not enough to answer whether an AI invocation created value. By preserving tool calls, workflow state, and outcome linkage, a gateway could evolve from a cost-routing layer into an AI spend observability layer.

My initial view is that enterprise procurement will increasingly resemble infrastructure procurement: unit economics first, capabilities second. Vendors such as SAP must first prove delivery, while AI startups must first prove measurement.

05 Counterarguments / Risks

SAP’s decision not to disclose figures may not reflect strategic blindness. Regulatory disclosures can themselves create misleading signals.

AI benefits vary with deployment cycles, customer industries, and workflow maturity. A unified ROI metric is not suited to both an agent deployment for a manufacturing customer and a scenario for a software sales team. Publishing aggregate benefits too early could collapse all those differences into one misleading number.

Enterprise software also follows a common pattern: critical products monetize first through bundling and seat expansion, then develop independent pricing only after they mature. Microsoft’s Teams, Power Platform, and Copilot all suggest that unclear disclosure boundaries do not necessarily mean economic value is absent. SAP may be following that path, using AI to increase retention and deal size across the SAP Business Technology Platform.

Accordingly, the earlier argument that a measurement layer will determine pricing power may be too strong. Pricing power may also depend on the switching costs, permission boundaries, and procurement relationships surrounding an existing ERP. Even if SAP cannot separately prove an individual agent’s ROI, control of the system of record may still give it a moat.

For startups, the greatest risk is misreading this signal and treating “the giant does not disclose ROI” as evidence that the market does not need ROI. Large enterprises not publishing averages does not mean budgets do not exist. SAP’s problem may be attribution, not demand.

The more conservative conclusion is that SAP has not yet demonstrated to the market that AI has become an independent, scalable profit pool. It has demonstrated product supply, but not economics. Whether this represents temporary silence or a structural limitation of the legacy software business model remains uncertain without access to SAP’s contracts, usage data, and customer renewal evidence.