Everything About AI
NextFin News - SAP is selling investors a detailed vision of the autonomous enterprise while offering a much thinner set of numbers with which to judge whether artificial intelligence is producing a financial return. The tension is visible in the company’s latest disclosures: SAP says 45,000 cloud customers use its AI solutions and that its current cloud backlog reached €24.1 billion, yet the reviewed Q3 results and AI-release materials do not separately present AI revenue, paid usage, retention, AI-linked bookings or customer-level return on investment. That is not evidence that the strategy is failing. It is evidence that investors still cannot isolate what is working.
SAP’s third-quarter results, released Oct. 21, showed the broader business moving in the right direction. Current cloud backlog rose 28% to €24.1 billion, or 26% at constant currencies. Cloud revenue increased 26% on a reported basis and 25% at constant currencies. Cloud ERP Suite revenue grew 29% on a reported basis and 28% at constant currencies, while total revenue rose 13% to €29.703 billion. Chief Financial Officer Dominik Asam said SAP crossed €10 billion in total quarterly revenue and €3 billion in non-IFRS operating profit.
Those figures matter because SAP’s AI is not being sold as a separate consumer application. Joule, Joule Work, specialized agents and the Business AI Platform are embedded in the cloud suite that produces the company’s reported growth. That makes the accounting story understandable: AI may help SAP win migrations, raise the value of its applications and improve customer retention without appearing as a standalone revenue line.
But the same structure creates an attribution problem. If cloud growth accelerates, investors cannot tell how much came from AI functionality, how much came from the broader migration cycle and how much came from SAP’s existing enterprise-software franchise. If AI customers rise, investors cannot tell whether they are paying for the tools, using them repeatedly or merely receiving features bundled into contracts they already signed.
The central question is therefore not whether SAP has launched AI products. It clearly has. The question is whether the company has built a measurement system capable of distinguishing a product catalogue from an economic engine.
The Numbers Show Cloud Momentum, Not AI Payback
The strongest evidence for SAP’s AI strategy is indirect. In its Q3 results, the company tied its Autonomous Enterprise strategy to cloud and backlog growth. Christian Klein, SAP’s chief executive, said customers were investing “from the cloud and data foundation to our Business AI.” That statement connects AI with commercial momentum, but it does not quantify the connection.
“Our Autonomous Enterprise strategy is not just resonating, it is being adopted. Our customers are investing in the technology behind it – from the cloud and data foundation to our Business AI.” — Christian Klein, SAP chief executive
The distinction between correlation and contribution is essential. A €24.1 billion current cloud backlog is a measure of contracted future cloud business. It is not an AI backlog. The 26% constant-currency increase demonstrates demand for SAP’s cloud transition, but it does not establish that an AI feature caused a customer to sign, expand or renew.
SAP’s Oct. 26 AI-release update supplies a second set of numbers. The company says 45,000 cloud customers are using its AI solutions. It also cites estimated benefits of a 35% reduction in the time required to build and deploy custom Joule Agents and a 10% reduction in the time to deploy custom Joule skills. These figures are useful product claims. They are not substitutes for recurring revenue, gross margin, usage frequency or customer retention.
The gap becomes clearest when the figures are placed side by side. SAP reports €29.703 billion of quarterly revenue and €24.1 billion of current cloud backlog with financial precision. It reports 45,000 AI customers with a reach metric. It reports 35% and 10% reductions with estimated implementation benefits. The reviewed materials do not present average AI revenue per customer, the percentage of cloud contracts that include paid AI, active users per customer or the cost of serving AI workloads.
That is more than a presentation choice. It determines what investors can test. A conventional software business can be evaluated through annual recurring revenue, net retention, seat growth, bookings and margins. An AI business also needs evidence of inference costs, attach rates, active use and expansion. Without those measures, the investor is left with a portfolio of demonstrations and a consolidated cloud number.
SAP’s product breadth makes the problem harder. Joule is positioned as a portfolio-wide AI engagement layer. Joule Work is designed to connect users with assistants and agents across business functions. The Business AI Platform grounds those systems in SAP and non-SAP data. The commercial value may show up in a larger ERP deal, faster migration or lower churn rather than in a separate AI invoice.
That is a credible model. It is also a model in which management must disclose the bridge from product capability to economic outcome. Otherwise, “AI is included in the deal” can mean anything from a decisive sales advantage to a costly feature that customers barely use.
There is a practical reason to demand that bridge even when AI is bundled. The same €24.1 billion backlog can contain contracts with very different economic profiles. One customer may buy a broad cloud migration and activate AI in several production workflows. Another may sign a similar contract for infrastructure and core ERP functionality while leaving AI dormant. Treating both contracts as equivalent AI evidence would confuse availability with consumption.
The distinction also affects how investors interpret growth quality. Cloud revenue and backlog are forward-looking indicators of SAP’s commercial momentum, but they aggregate products with different pricing, implementation and support requirements. If AI raises conversion rates, it could improve the value of backlog even before it produces a separate invoice. If AI raises service intensity or model-delivery costs, it could weaken margins even as backlog expands. The same headline can therefore conceal opposite unit economics.
The Missing KPI Is a Structural Problem
The disclosure gap is structural, not merely a temporary lack of quarterly detail. SAP has chosen to make AI an embedded layer across applications, data and workflows. That may be strategically sensible, but it dissolves the boundaries needed to measure the investment.
Under a standalone model, the chain is easier to follow: a customer buys an AI product, the vendor records recurring revenue, usage generates variable infrastructure cost, and retention or expansion shows whether the product earned its place. Under SAP’s model, the chain is longer. AI functionality can help a customer migrate to the cloud; the cloud migration can increase the value of the ERP suite; the larger suite can improve the economics of SAP’s overall relationship with that customer. The AI contribution is real if it changes the chain, but it is not visible in any one reported line.
This is the transmission mechanism investors should focus on. The first-order effect is product utility: an assistant automates a task or an agent coordinates a workflow. The second-order effect is commercial: the customer accepts a broader SAP cloud footprint, renews more software or reduces the internal cost of adopting SAP. The third-order effect is financial: backlog, cloud revenue, margins and lifetime value improve. SAP currently discloses the first and third layers in fragments, but not the bridge between them.
The issue is not that every AI feature needs its own revenue line. That would create artificial precision. The issue is that SAP could disclose a small set of operating indicators without changing its accounting structure: AI attach rate in cloud orders, paid versus bundled deployments, monthly active users, usage intensity, AI-related expansion, gross retention among AI customers and the incremental infrastructure cost associated with those workloads.
A useful disclosure framework would separate four stages. The first is availability: how many customers have access to Joule, agents or other AI tools. The second is activation: how many customers have moved from testing to production. The third is monetization: how many pay for additional capacity, premium features or AI-linked services. The fourth is economics: whether those accounts expand, renew and produce acceptable gross margins. SAP currently offers a clear signal on the first stage and selected productivity estimates, but the later stages are not separately visible in the reviewed materials.
That sequencing matters because enterprise software adoption is rarely instantaneous. A customer can announce a pilot, spend months governing data and permissions, and only later put the tool into a production workflow. A reach metric may therefore lead financial results by several reporting periods. But the lag must eventually resolve into a conversion rate or an expansion pattern. Otherwise, the metric remains a permanent promise rather than a leading indicator.
Until then, the company’s figures answer a narrower question: Is SAP’s cloud strategy gaining traction? The answer appears to be yes. They do not answer the harder question: Is AI improving the economics of that strategy faster than it adds costs and complexity?
The distinction matters because AI can increase revenue and reduce margins at the same time. Every automated workflow may make an application more valuable, but every query can also carry model, hosting, data-governance and support costs. A 35% reduction in the time to build and deploy a custom agent could lower implementation friction, yet it could also encourage more usage that SAP must deliver at a profitable price. The productivity metric is therefore an input, not a return.
There is another structural complication. SAP’s 45,000-customer figure counts customers using AI solutions, but the release does not define the threshold for “using” in the retrieved material. It does not distinguish a customer testing one assistant from a customer running agents across finance, procurement or supply-chain operations. Without a usage definition, the number is a funnel metric, not a monetization metric.
The same problem affects customer case studies. A reported reduction in development effort or information-search time can demonstrate that a workflow works. It cannot establish the company-wide payback after implementation, training, governance and recurring infrastructure expenses. The relevant measurement is not whether an AI task became faster; it is whether the customer’s total cost of operating the process fell enough to justify the software bill.
SAP’s strategy has therefore created a reporting asymmetry. The benefits are described at the workflow level, while the financial results are disclosed at the company or cloud-segment level. The costs of AI sit somewhere in research, infrastructure, support and acquisitions. The upside sits somewhere in bookings, retention and pricing. The market needs a bridge.
The Bull Case: Embedded AI May Be More Valuable Than Standalone AI
The strongest case against this criticism is that standalone AI revenue could be the wrong test. SAP does not need to become a foundation-model vendor to benefit from AI. Its advantage is the context inside enterprise workflows: invoices, orders, suppliers, employees, inventory and financial controls. If AI makes that context easier to act on, the economic value may accrue to SAP’s core suite.
That argument has force. SAP’s Q3 results show a business with enough scale to monetize an embedded feature through a large installed base. Cloud ERP Suite revenue grew 29% on a reported basis, faster than total revenue growth of 13%. If AI helps persuade customers to move from older installations to cloud ERP, it can generate value without being priced as a separate product. The company’s current cloud backlog of €24.1 billion also gives SAP a large pool of contracted business in which AI capabilities can influence adoption and expansion.
Embedding AI may also reduce customer friction. Buyers do not want to assemble a separate model, data layer, workflow engine and compliance system for every business process. They may pay for a trusted system that already understands authorization, audit trails and enterprise data. SAP’s argument is that the closer AI sits to the transaction system, the more defensible its output becomes.
There is a further strategic benefit to integration. If Joule can coordinate actions across finance, procurement, human resources and customer operations, SAP can sell an operating environment rather than a collection of isolated copilots. That could make the suite harder to replace because the value lies in shared permissions, data context and process continuity. A standalone tool might be faster to launch, but an embedded tool could be harder for a large enterprise to remove once it becomes part of controlled workflows.
But this defense changes the burden of proof; it does not remove it. If AI is primarily a migration accelerator, SAP can report the conversion rate of AI-influenced opportunities, the win rate of deals with AI attached or the difference in cloud expansion between AI-enabled and non-AI customers. If AI is primarily a retention tool, SAP can report renewal or net-retention differences. If AI is primarily a pricing tool, it can report paid attach rates and average contract-value uplift.
None of those indicators would require SAP to pretend that AI is a standalone segment. They would show whether the embedded model is working. Without them, the bull case remains logically possible but empirically under-tested.
The second-order risk is that embedding AI can conceal weak product economics. A customer may accept AI as part of a larger cloud contract because the migration is already necessary. SAP would then receive credit for AI in the narrative while the customer’s purchase decision was driven mostly by ERP modernization. Conversely, customers could use AI heavily but receive it at a price that does not compensate SAP for compute and support costs.
That is why the next proof point should not be another agent count. It should be evidence that AI changes the economics of the customer relationship. The useful comparison is between customers with sustained, paid AI usage and the rest of the cloud base: expansion, renewal, support costs and gross margin. A company can have a rising number of active AI interactions and still fail to create attractive returns if monetization and delivery costs move in opposite directions.
What Would Prove the Thesis Wrong?
The bearish counter-thesis is straightforward: SAP’s AI story may be a marketing layer placed on top of a cloud transition that was already underway. On this view, the 26% constant-currency growth in current cloud backlog and 25% constant-currency cloud-revenue growth show migration demand, not AI success. The 45,000-customer adoption figure could be broad enough to include low-intensity or bundled usage. Product estimates of 35% and 10% faster deployment could describe development efficiency without creating SAP revenue.
That is the strongest challenge because it attacks attribution at the foundation. If cloud growth is healthy without AI, then AI may be strategically useful but financially unproven. If growth slows, management could still point to a rising number of assistants and agents, leaving investors unable to determine whether the products are cushioning the slowdown or merely generating more announcements.
SAP can answer this challenge only with cohort evidence. It could disclose whether customers using paid AI expand faster than customers that do not, whether AI affects renewal rates and whether AI-enabled deals carry higher or lower margins. It could define “using AI” by active users, production workloads or paid contracts rather than by any deployment.
The falsifying signal for the structural embedded-value thesis is specific: if SAP begins reporting a recurring AI KPI and that measure shows no improvement over several quarters while constant-currency cloud-revenue growth falls below 20%, the claim that AI is materially strengthening the cloud engine would be weakened. A second warning would be a rising AI customer count alongside stagnant cloud expansion and declining cloud gross margin. That combination would indicate reach without economic conversion.
The opposite signal would support management’s view: AI attach rates, paid usage and AI-linked expansion rising alongside stable or improving cloud margins. Such a pattern would show that the omission is a reporting limitation rather than a business limitation.
This is where the cyclical and structural elements must be separated. SAP’s cloud growth is partly cyclical: enterprise IT budgets, migration timing and large-deal activity can accelerate or pause. Those forces can mean-revert. The disclosure problem is structural because it follows from product architecture and reporting design. A stronger or weaker quarter will not solve it automatically.
That distinction also changes how investors should read future results. A quarter of faster cloud growth could be a cyclical upswing with no incremental AI proof. A quarter of slower growth could still contain strong AI adoption if AI improves retention or workflow productivity. The relevant test is not whether SAP repeats the word “AI” more often. It is whether the company’s operating metrics connect AI use to durable customer economics.
Three Horizons for SAP’s AI Story
In the short term, the disclosure gap is likely to keep sentiment tied to SAP’s consolidated cloud numbers. The €24.1 billion backlog and double-digit revenue growth provide a strong operating cushion, so investors may tolerate limited AI segmentation while the core business expands. Product launches such as Joule Work and new specialized agents can support the sales narrative, but each launch will raise the question of adoption quality.
In the medium term, the decisive issue is monetization. SAP must show whether AI is included in more cloud orders, whether customers pay for deeper usage and whether those workloads produce acceptable margins. The company’s €29.703 billion quarterly revenue and €3 billion in non-IFRS operating profit measure SAP’s scale, not the profitability of AI. The next layer of evidence must be cohort, attach-rate or usage data.
In the long term, the structural question is whether enterprise software vendors that own transactional context can defend their position as AI agents become more capable. SAP has an advantage if its systems remain the authoritative record for finance, procurement and workforce data. It faces a threat if external agents can execute across systems without needing SAP’s interface or application layer. In that contest, data governance, workflow permissions and auditability may matter as much as model quality.
The base case is that SAP continues to grow cloud revenue and backlog while gradually adding AI disclosures, but keeps AI embedded in the broader suite. The trigger is continued adoption of Joule and agents alongside stable cloud margins and evidence that AI is present in a rising share of bookings. The upside case is that AI becomes a measurable expansion engine: paid attach rates rise, AI-enabled customers renew and expand faster, and productivity gains lower service costs. The trigger would be a recurring AI KPI that improves across at least two reporting periods.
The downside case is that AI adoption broadens but remains commercially shallow. Customers experiment, SAP absorbs infrastructure and support costs, and cloud growth slows below 20% at constant currencies without a corresponding improvement in AI monetization. The trigger would be a widening gap between AI-customer counts and cloud expansion, accompanied by pressure on cloud gross margin.
For customers, embedded AI can be valuable because it reduces the integration burden of putting agents into controlled business processes. For SAP, that same embedded model is valuable only if it produces stronger retention, higher contract values or lower delivery costs. For investors, the absence of a standalone AI revenue line is not itself a flaw. The absence of any repeatable bridge from AI activity to those outcomes is.
SAP’s AI story is therefore neither disproved nor proved by its current results. The company has demonstrated product breadth, customer reach and a powerful cloud platform. It has not yet demonstrated, in a separately testable way, how much economic value its AI layer creates.
Until SAP reports that bridge, its AI numbers will measure reach and capability more convincingly than they measure returns.
Data cutoff: Nov. 11, 2026, 12:40 UTC.
Explore more exclusive insights at nextfin.ai.
