Last week (October 28 to November 2), the two enterprise software giants SAP and ServiceNow successively released their Q3 2026 results. Putting the two earnings reports side by side, I found an intriguing contrast: on one hand, Agent consumption in customers' production environments surged 13-fold year over year; on the other hand, neither company broke out (or was too embarrassed to break out) "standalone AI revenue" separately.
Global Tiger Finance put this contrast bluntly in an analysis in early November — "AI snatches away the operator console, SAP holds its core base." What AI is easiest to snatch away is the operating interface that employees click every day in enterprise software; what is hardest to snatch away are the rules behind every transaction: whether the budget is sufficient, who has approval authority, how invoices are verified, and who is responsible when something goes wrong. And these rules are precisely the business ERP has been doing for decades.
In this issue, we’ll use the newly disclosed financial report data to break down the “scissors gap” of “AI is hot, the books are steady, and profits are still far away,” and see what it really means for enterprise CIOs.
1. "The Console Was Seized, but the Ledger Didn't Move": The Real Division of Labor Between AI and ERP
First, let us clarify a commonly confused piece of logic. AI agents are indeed taking over "operations"—price comparison, generating contract drafts, summarizing the pipeline, writing emails, and scheduling work orders; these actions can be completed in seconds. But when it actually comes to spending money, recording an accounting entry, or issuing a payroll payment, things immediately become complicated: it needs to pass through six gates—budget, qualifications, approval, bookkeeping, reconciliation, and accountability.
The figures SAP itself provides are very telling. Based on the 2026 Q3 earnings report disclosed in late October and a November 2 research report from CSC Financial: SAP already has more than 110,000 Joule Work Desktop users internally, productivity in core functions has increased by more than 20%, and year to date, the number of pull requests merged into the codebase per developer has risen by about 70%. Note that these are all efficiency gains at the 'console' level—a faster interface, a smoother assistant, and easier coding.
What AI is taking away is the "front office" of enterprise software, while what ERP safeguards are the "back-office rules" behind every enterprise transaction. Whoever takes the operating entry point in the AI era, and whoever safeguards the enterprise ledger—this is the question most worth studying in financial reports.
And on the ledger side, SAP’s third-quarter total revenue reached €10.270 billion (+13% year-over-year), cloud revenue reached €6.668 billion (+26% year-over-year), and current cloud backlog reached €24.144 billion (+28% year-over-year). The cloud business almost filled all the gaps left by the shrinking traditional license and maintenance business, and also underpinned all of the growth. In other words: growth came from "moving the old business to the cloud," not from new profits created by AI alone.
II. The “Scale–Profitability Scissors Gap” in the Two Financial Reports
Putting SAP and ServiceNow's key metrics side by side makes the scissors gap even clearer.
| Indicator | SAP 2026 Q3 | ServiceNow 26Q3 (disclosed on 10/28) |
|---|---|---|
| Total revenue | EUR 10.27 billion (+13%) | USD 4.106 billion (+20.5%) |
| Cloud / Subscription Revenue | EUR 6.668 billion (+26%) | US$3.988 billion (+21%) |
| Contracts confirmable in the next 12 months | 24.144 billion euros (+28%) | cRPO 13.65 billion (+20.5%) |
| AI signal | 110,000+ Joule users / productivity +20% | Production environment Agentic AI consumption +13x year-over-year |
| Independent AI revenue | Not separately disclosed; cloud gross margin -0.8pct YoY | AI ACV target raised to $1.7 billion (from $1 billion at the start of the year) |
| Earnings quality | Cloud gross margin slightly under pressure | GAAP net profit 277 million (-45%), after-hours -18% |
ServiceNow's story is even more extreme. It raised its 2026 AI annual contract value (ACV) target from $1 billion at the beginning of the year all the way to $1.7 billion (+70%, the second increase this year), and disclosed that since the beginning of this year, customers' usage of agentic AI in production environments has grown by more than 13x year over year; in a single quarter, it closed 19 new deals with ACV exceeding $5 million (+135% YoY), and the number of customers with annual ACV exceeding $5 million reached 680 (+23% YoY).
ServiceNow production environment agentic AI consumption year-over-year growth (disclosed in 2026 Q3; source: Zhitong Finance / Tiger Brokers)
But in the same earnings report, GAAP net income was only $277 million, down about 45% from $502 million in the same period last year, and the stock fell more than 18% after hours. Burning cash at scale, AI deployment services being split out from subscription revenue into professional services, and capital expenditure growing faster than operating cash flow—all of this shows that: Agent consumption is exploding, but turning it into independent, sustainable profit still has a financial hurdle to clear.
This pours cold water on all managers who shout "AI-native leapfrogging": AI's "scale metrics" (user count, call volume, number of agents) and "profitability metrics" (standalone revenue, gross margin, net profit) are increasingly diverging into a scissors gap. When purchasing AI, if a vendor can only show you the former and cannot give you the latter, that budget is most likely a "demo fee" rather than a "productivity fee".
3. Agent consumption is up 13x, and the security bill has arrived too
The other side of explosive scale is that risks are amplified in lockstep. Just two weeks before earnings season, on October 16, India's CERT-In reported: agents from multiple AI Agent vendors had reached the production systems of Indian enterprises, read personal data and confidential business records, and gone beyond the scope of task authorization; the intrusion path was credentials left in application configurations, public code repositories, and overly broad cloud roles, and "not a single affected organization noticed," with the AI vendors ultimately proactively reporting it. Three days later, Microsoft publicly recommended that enterprises "treat AI agents as insider threats."
This corroborates the security retrospectives of Odoo 20 conducted by multiple implementation partners at the end of October: in Odoo 20's MCP write operations, "read-only" is only a hint rather than a lock, and ultimate control rests entirely on the "connected user's access permissions." The larger the scale and the more autonomous the Agent, the less this gate can be guarded by "trusting the model."
Scale and risk are rising in tandem. When Agentic AI consumption has increased 13x year over year, have your governance, audit, and permission revocation capabilities also increased 13x in tandem? If the answer is "no," then every "consumption surge" is a potential liability exposure. ServiceNow itself also deployed Control Tower, AI Passport, and new security capabilities across agent discovery, permission verification, continuous monitoring, and anomaly handling in Q3—which precisely shows that vendors are also catching up on the lessons "after scale."
IV. Practical CIO Advice: Don't Count Agents—Check Whether AI Actually Touches the Ledger
Having discussed data, let's get practical. If you remember just one sentence, I suggest it be this: To judge AI's value, don't count the number of Agents—check whether it has safely touched your core ledgers and compliance processes.
First, tie AI budgets to "core process transformation," not to "tool procurement." SAP's cloud growth comes from moving legacy ERP to the cloud, while ServiceNow's AI consumption comes from penetrating the business processes of its existing customers. First list your three most painful core processes (reconciliation, closing, procurement approval), then decide where to invest in AI—rather than buying a bunch of "general-purpose assistants" first.
Second, establish an Agent ledger and a least-privilege baseline. Learn from the CERT-In lesson: clean up residual credentials in configurations, tighten cloud roles, give each Agent an independent least-privilege user, and retain human confirmation for write operations. For native MCP integrations like Odoo 20 in particular, you must "lock first, then open."
Third, accept AI using "verifiable metrics." Do not accept vague promises like "a 20% improvement"; require vendors to provide specific baselines in your core processes (such as days to close the books, manual hours for reconciliation) with before-and-after comparisons. Otherwise, just like financial reports that "do not separately disclose AI revenue," the accounts will never add up.
Finally: it is certain that AI will seize the control console, and it is equally certain that ERP will hold the ledger. What is truly uncertain is the middle part—"automation ≠ autonomy." Whoever first fills the gaps in governance, auditing, and accountability can truly turn +13x consumption into +13x profit. The biggest takeaway from this earnings season is not how capable AI is, but how solid the ledger is.
This article is based on analysis of public information: SAP 2026 Q3 financial report (disclosed in late October), ServiceNow 26Q3 financial report (disclosed on 10/28, compiled by Zhitong Finance / Tiger Brokers / China Securities), Global Tiger Finance's "AI Seizes the Control Console, SAP Holds the Base" (early 2026-11), China Securities research report "AI Computing Power Investment and Commercialization Form a Linkage" (2026-11-02), India's CERT-In security bulletin (2026-10-16), and TechUltra Solutions' "Odoo 20 AI Agents & MCP 8-Point Security Checklist" (2026-10). Data as of 2026-11-02.
Zhongdi Information · Focused on Odoo and Enterprise Management Digitalization Solutions
