Salesforce's AI Growth Needs a Stable Denominator

Executive summary

Salesforce's fiscal second quarter delivered the strongest evidence yet that enterprise AI is becoming a real software workload. Agentforce annual recurring revenue exceeded $1.5 billion, quarterly agentic work units nearly doubled from the prior quarter, and bookings for premium editions that embed agent capabilities more than doubled sequentially. Current remaining performance obligations grew 14% in constant currency, the fastest pace Salesforce has reported in several years.

The mainstream reading is straightforward: enterprise agents have moved from pilot to production and are reaccelerating Salesforce. The evidence supports part of that view, but not the whole of it. The company's AI and data metrics have expanded through acquisition, product bundling and changing definitions at the same time that usage has accelerated. Reported second-quarter revenue grew 10.8%; subtracting the disclosed $456 million Informatica contribution produces an estimated 6.4% increase for the rest of the business on a reported-currency basis. Agentforce ARR also changed scope in the quarter to include additional AI offerings, Slackbot and Headless 360.

Our contrarian thesis is not that Salesforce's AI growth is artificial. It is that the commercial category is widening faster than the public evidence needed to separate new workload, acquired revenue, bundled value and relabeled installed-base spend. That is strategically rational for Salesforce: the company wants to make data, governance, workflow and agent execution one platform. It is less useful as a customer ROI measure.

Operators should therefore use vendor momentum as evidence of ecosystem formation, not as a substitute for their own denominator. The relevant questions are how much net-new spend an agent requires, how much production work it completes, which existing costs it changes, and whether those results persist after pilots, credits and bundle migrations.

Market context

Salesforce reported second-quarter fiscal 2027 results on August 26. Revenue reached $11.345 billion, up 10.8% from $10.236 billion a year earlier. Subscription and support revenue rose 11.7% to $10.820 billion. Current RPO was $33.5 billion, up 14% in constant currency, while total RPO reached $66.3 billion.

The quarter also contained a material acquisition effect. Salesforce disclosed $456 million of Informatica contribution to total revenue and $440 million to subscription and support revenue. For the full year, its 11% constant-currency revenue outlook includes slightly more than three percentage points from Informatica. For the third quarter, the acquisition contribution is expected to exceed four points.

At the same time, the operating signals strengthened. Agentforce and Data 360 ARR reached nearly $3.9 billion, up more than 210% year over year. Agentforce ARR exceeded $1.5 billion, up more than 240%. Salesforce delivered 3.2 billion agentic work units in the quarter, up 97% sequentially, and Data 360 ingested 104 trillion records, up 355% year over year. Reuters reported that Salesforce raised its annual forecasts and that Agentforce, Data 360 and Slack were offsetting continued volatility in overall license revenue.

Salesforce also announced Claudeforce, an expanded Anthropic partnership that makes Salesforce data, workflows, actions and governance accessible from Claude. The initial plugin includes 37 prebuilt sales skills. This is an important strategic clue: the value proposition is shifting from a proprietary model or chat interface toward the enterprise context and permission layer that lets multiple models act inside a system of record.

Findings

Finding 1

The reported acceleration mixes organic demand with acquired revenue.

The quarter's 10.8% revenue growth is accurate, but it is not an organic growth rate. Subtract the disclosed Informatica contribution and the remaining Salesforce revenue grew by an estimated 6.4% from the prior-year base. Subscription and support revenue excluding the disclosed contribution grew by an estimated 7.1%.

The same issue is visible inside the new revenue categories. Data 360, Headless Platform and Other rose 20.3% to $3.618 billion. Salesforce says this category includes Informatica. Removing the disclosed $440 million Informatica subscription contribution leaves $3.178 billion, about 5.7% above the prior-year category. This is an approximation because the company discloses the acquisition contribution at the total subscription level rather than a formal organic category reconciliation.

Q2 FY27 measure	Reported result	Disclosed Informatica contribution	Estimated result excluding contribution	Estimated growth excluding contribution
Total revenue	$11.345B	$0.456B	$10.889B	6.4%
Subscription and support revenue	$10.820B	$0.440B	$10.380B	7.1%
Data 360, Headless Platform and Other subscription revenue	$3.618B	$0.440B*	$3.178B*	5.7%*

Source: Salesforce Q2 FY27 earnings release and supplemental revenue analysis. Period: three months ended July 31, 2026 versus July 31, 2025. Units: U.S. dollars in billions and year-over-year percentage change. Method: subtract disclosed Informatica contribution from current-period reported revenue, then compare with the published prior-year base of $10.236 billion total revenue, $9.690 billion subscription revenue and $3.008 billion for the data/platform category. *The category estimate assumes the disclosed subscription contribution sits in the category that Salesforce says includes Informatica; it is not a company-reported organic measure. Currency effects are not removed, and acquired costs are not analyzed.

This distinction does not invalidate the strategy. Informatica gives Salesforce integration, data quality, governance and metadata capabilities that agents need. Acquisition can be the fastest path to a coherent platform. But it changes what the growth rate proves. The headline demonstrates that the combined company is larger and selling faster; it does not show that pre-acquisition products alone are growing at the same pace.

Finding 2

AI ARR is becoming a portfolio label, not a clean product cohort.

Salesforce's Agentforce ARR rose from $800 million in the fourth quarter of fiscal 2026 to $1.2 billion in the first quarter of fiscal 2027 and more than $1.5 billion in the second quarter. That sequence is impressive. It is not perfectly comparable.

In Q2, Salesforce expanded Agentforce ARR to include its AI offerings, Slackbot and Headless 360. Its definition also says the company may update the metric to incorporate new products, features or acquired technologies. In the combined Agentforce and Data 360 figure, previous quarters explicitly included $1.1 billion of Informatica Cloud ARR. The Q2 headline did not provide the same component bridge.

Fiscal quarter	Agentforce ARR	Agentforce and Data 360 ARR	Scope note disclosed by Salesforce
Q4 FY26	$0.8B	More than $2.9B	Combined figure explicitly included $1.1B Informatica Cloud ARR
Q1 FY27	$1.2B	Nearly $3.4B	Combined figure explicitly included $1.1B Informatica Cloud ARR
Q2 FY27	More than $1.5B	Nearly $3.9B	Agentforce scope expanded to AI offerings, Slackbot and Headless 360; no component bridge in headline

Source: Salesforce Q4 FY26, Q1 FY27 and Q2 FY27 earnings releases dated February 25, May 27 and August 26, 2026. Period: quarter ends January 31 through July 31, 2026. Units: annual recurring revenue in U.S. dollars. Method: direct transcription of company-reported thresholds and scope notes. Limitations: values are rounded, ARR is annualized contract value rather than recognized revenue, and changing inclusions prevent a strict like-for-like growth calculation.

The strategic logic is clear. Once agent functions are embedded across Sales, Service, Slack and a headless platform, drawing a hard boundary around one SKU becomes artificial. Customers may buy a premium edition, flex credits, data capacity and integration together. Salesforce is organizing around the job to be done.

For measurement, however, a wider label can turn ordinary portfolio expansion into apparent product acceleration. Bundled AI can be commercially successful even if customers would have renewed the core software without it. ARR can rise because a contract is active, while production utilization remains low. A metric is most informative when its population and denominator remain stable.

Finding 3

Usage is the stronger evidence, but it still stops short of ROI.

Agentic work units are Salesforce's attempt to move beyond token counts. The company defines an AWU as a discrete production task executed by an AI agent, such as resolving a case, updating a record or triggering a workflow. Q2 delivered 3.2 billion AWUs, 97% more than Q1, taking the cumulative total to 7.0 billion. Data 360 record ingestion and zero-copy activity also rose sharply.

These measures are closer to operational reality than deals, pilots or token volume. They show that agents are doing work inside production systems. They also reveal why the data layer is valuable: an agent that cannot access governed records, business logic and permissions is a demonstration, not an operating capability.

But AWUs do not have a public economic denominator. A case resolution, record update and complex multistep action can each count as a unit despite different value and compute cost. Salesforce does not disclose unique active customers, AWUs per paying customer, success rate, human review, exception rate, credit consumption or gross margin for the workload. Rapid unit growth could represent deeper adoption, more customers, smaller task granularity or a mix of all three.

This is where the mainstream interpretation outruns the evidence. Production work is real. Broad, repeatable customer ROI is not yet observable from the reported aggregate. The distinction matters because usage-priced AI can expand vendor revenue even when customer savings are delayed by data cleanup, integration, governance and supervision.

Implications for operators

Treat enterprise AI as a workload portfolio, not one software line. Track each use case from contract value to production volume, successful outcomes and financial effect. A useful internal bridge is: contracted capacity, activated capacity, attempted work, completed work, accepted work and realized value. Attrition at any stage identifies a different operating problem.

Demand a stable commercial bridge from vendors. Separate core renewal, acquired products, premium-edition uplift, consumption credits and services. When definitions change, preserve the old cohort internally for at least four quarters. Procurement should know whether spend growth reflects more production work or a broader bundle.

Price the data foundation as part of the AI case. Informatica's contribution is not a distraction from Salesforce's agent strategy; it is evidence that governed data and integration are prerequisites being monetized. Include data quality, identity resolution, access control and workflow redesign in the business case before promising labor savings.

Use an outcome-weighted unit. Vendor work units can help reconcile invoices, but internal management needs value classes. A resolved service case, qualified sales opportunity and administrative record update should not receive the same economic weight. Attach completion quality, human handling time, downstream reversal and customer effect.

Preserve model optionality. Claudeforce suggests the system of record can become more valuable when it serves multiple AI interfaces. Contracts should clarify data portability, action permissions, audit logs, model substitution and the treatment of derived data. The aim is to avoid making a workflow portable at the model layer but captive at the context layer.

Risks & open questions

The thesis would weaken if Salesforce publishes a consistent scope bridge showing that most Agentforce ARR growth is like-for-like, net-new spend and if AWU growth is accompanied by stable or improving success rates, customer retention and realized savings. It would also weaken if organic revenue visibly reaccelerates in the second half as management expects, without relying on additional acquisition contribution.

The analysis may understate strategic benefit from Informatica. Acquired capabilities can raise growth in the rest of the platform through cross-sell, retention and faster deployment. Subtracting current acquisition revenue does not capture those interactions. Reported currency and constant currency measures also differ.

Three questions remain open. How concentrated are AWUs among a small number of large deployments? What portion of premium-edition bookings is incremental rather than a packaging migration? And as agents automate seat-based work, will consumption revenue grow faster than any seat contraction it causes?

Appendix / methodology notes

This report reviewed Salesforce's Q4 FY26, Q1 FY27 and Q2 FY27 earnings releases, supplemental financial tables, metric definitions and the August 26 Claudeforce announcement. Reuters reporting was used to corroborate the market interpretation and management's description of license volatility.

Estimated ex-Informatica growth is a Blackrock Research calculation, not a Salesforce non-GAAP measure. Current-period disclosed Informatica revenue was subtracted from reported revenue and compared with the published prior-year total. The calculation does not remove currency effects, acquisition-related costs or indirect cross-sell. The data-category estimate is marked because Salesforce discloses Informatica subscription contribution in aggregate rather than as a category reconciliation.

No customer ROI, labor reduction or deployment success rate is inferred from ARR, record ingestion, token or AWU totals. A stronger future dataset would include quarterly active production customers, like-for-like ARR by stable product definition, AWUs per customer, completion and human-escalation rates, credit consumption, agent gross margin and customer-reported financial outcomes.