Salesforce's $1.5B AI Business: Check the Boundary Before the Growth

The important news sits beside the growth rate

On August 26, Salesforce reported results for its second fiscal quarter of 2027, ended July 31. Agentforce ARR exceeded $1.5 billion, up over 240% year over year. The same disclosure says the metric now includes AI offerings such as Slackbot and Headless 360. Source: company earnings release and metric definitions.

A broader product portfolio can reasonably lead to a broader metric. That does not make the growth false. But turning the number into a long-term valuation assumption requires another bridge: how much business grew within a consistent perimeter?

My view: the next dividing line for software stocks is not who can report the biggest AI-labelled revenue pool. It is who can demonstrate that customers' wallets are expanding. Investors may pay for the future, but they should not compound a change in reporting scope.

From $1.2 billion to $1.5 billion is not a new-sales receipt

Salesforce's May 27 first-quarter release reported $1.2 billion in Agentforce ARR, up 205% year over year. Subtracting that from the latest number and calling the difference net new sales conflates both scope and timing.

ARR measures the annualized recurring value of active subscription agreements at period-end. It is neither revenue recognized during the quarter nor cash already collected. The second-quarter announcement reviewed for this article does not provide a dollar bridge between the old perimeter, newly included products and comparable growth, or explicitly explain how historical comparisons were handled under the new scope.

We therefore cannot calculate a supposedly 'true organic growth rate.' Nor can we conclude that reclassification accounts for most of the reported 240%-plus growth. The missing item is a reconciliation—not permission to select whichever narrative suits a position.

The counterevidence investors must not ignore: refills

On pages 18–19 of the earnings-call transcript, management addressed the move from Agentforce pilots to paid use. It said roughly half of the relevant bookings came from customers buying more after consuming Flex Credits. That is closer to commercial demand than cumulative activity alone: someone used the service and chose to pay again.

But half of bookings coming from refills does not mean half of customers renewed. It is not net revenue retention either. Concentrated large-customer purchases, promotions and different contract lengths can all affect booking mix. That statement does not supply churn, downgrades or annual net spending for a fixed customer cohort.

The strongest bullish argument is not a bigger AI label. It is that existing workflows can repeatedly generate additional purchases. Skeptics must also explain why customers would keep paying if refills persist. Reporting comparability is worth investigating, but it cannot substitute for studying actual demand.

Even without additional spending, AI might protect existing revenue by reducing churn, or lower service costs. That defensive value should not be written down to zero. It is, however, a different valuation path from incremental revenue and needs different evidence. Whether customers would have left without AI remains a counterfactual to test.

Revenue guidance already shows how a bridge should work

The company explains its full-year revenue guidance increase clearly: $100 million from organic growth, $200 million from the pending Contentful and Fin acquisitions, offset by a $100 million currency headwind, producing a net $200 million increase. Acquisition contributions are conditional on closing. Source: second-quarter guidance explanation.

This is not evidence of a weak business. It is a reminder that the entire increase cannot be attributed to existing operations, much less to AI alone. An AI metric would similarly benefit from a bridge separating comparable growth and perimeter changes.

Another trap is dividing annualized ARR by quarterly task volume to produce a 'price per task.' The numerator annualizes a period-end stock; the denominator measures activity during a period and may cover different products. The resulting decimal is not a reliable unit-economics model.

Buffett asks for explanation; Marks asks about expectations

The next useful table: the same customers' wallets

I would rather see net spending by a fixed cohort before and after AI adoption: additional AI purchases, less displaced subscriptions, discounts and churn, followed by the incremental cost of delivery. The analysis must retain customers who did not renew, not just successful upgraders.

If net payments and contribution from that cohort rise persistently, economic value could improve even as the AI metric's growth slows. If the label grows rapidly while total spending stays flat, expectations for incremental monetization should come down. Both scenarios must remain open to subsequent evidence.

Use a reverse DCF to test the growth demanded by price, and record scope, renewal and cost assumptions in an investment risk checklist. What deserves a premium is not a bigger reporting perimeter, but customer value that grows—or is defended more securely—while that perimeter stays fixed.