AI Revenue Disclosure: 7 SaaS Metrics, 7 Definitions, and Which Ones to Trust

Microsoft says its AI business runs at more than $37 billion a year. Salesforce says Agentforce and Data 360 carry nearly $3.9 billion of annual recurring revenue. ServiceNow says its AI crossed $1 billion of annual contract value. That is AI revenue disclosure in 2026: three headline numbers, three different bases, and no honest way to line them up.

The short answer

Public SaaS companies define AI revenue in at least four incompatible ways. They are recurring revenue from AI subscriptions, signed annual contract value, an annualised run rate, and growth rates or bookings shares with no dollar base. None of these is audited revenue. None can be compared directly with another company's figure.

What an AI revenue disclosure actually counts

AI revenue disclosure is any figure a listed company publishes that claims to isolate what it earns from AI products. It is voluntary. No accounting standard requires it and no auditor signs it, so every company picks its own unit.

The unit matters more than the number. There are four in circulation among large public software companies, and each answers a different question.

Annual recurring revenue (ARR) is the yearly value of subscriptions that are active on the reporting date. It describes the installed base. Annual contract value (ACV) is the yearly value of contracts signed, and in ServiceNow's usage it is often a bookings measure. Run rate annualises a recent period's revenue, usually a month or a quarter. The fourth unit is not a unit at all: a growth rate or a percentage of bookings, published without the dollar base underneath it.

My view is simple. A metric without a written definition is a talking point, not monetization data. That rule alone removes most of the AI numbers quoted on earnings calls this year.

The AI revenue disclosure reference table

The table below compiles the AI revenue metric each company put in its most recent primary release. Where a figure appears only on an earnings call or in secondary coverage, it is marked as such or left out.

Every row is taken from the company's own earnings release or 8-K exhibit, read for this post. "Written definition" means a definition printed in the release itself, not wording from a call. Figures are the latest available on 10 October 2026.

Read the last two columns before the third. The figure column is what gets quoted. The definition and scope columns decide whether the figure means anything a quarter later.

The same comparability problem exists one layer up, among the AI labs, where gross and net reporting distort the totals. That version of it is covered in the AI revenue leaderboard built on booked numbers. This post stays with listed software companies, where the disclosures sit inside SEC filings and should therefore be held to a higher standard.

Salesforce Agentforce ARR: the best-defined AI metric, with a moving scope

Salesforce does the thing every other company on this list avoids. It prints a definition of its AI revenue metric in the earnings release, under "Other Metrics", where anyone can quote it.

What Agentforce and Data 360 ARR counts

The company defines the combined metric in one sentence. It counts "the annualized recurring value of active Data 360 and certain generative artificial intelligence" subscription agreements, "including those for Agentforce, generative AI products and features, and Headless 360". Those agreements are counted at the end of the reporting period. That is an ARR definition in the textbook sense. It describes the installed base on one date.

In the Q2 FY27 release of 26 August 2026, that combined figure reached nearly $3.9 billion, up over 210% year over year. Agentforce ARR on its own exceeded $1.5 billion, up over 240%.

The combined number has a problem the definition admits openly. Data 360 is a data platform. Much of what customers pay for it is storage, ingestion and unification, not generative AI. Salesforce does not break Data 360 out, so you cannot tell how much of the $3.9 billion is AI in any meaningful sense.

The Q2 FY27 scope change

The same release carries one sentence that changes how to read the Agentforce growth rate: "Beginning in Q2 FY27, Agentforce ARR includes our AI offerings Slackbot and Headless 360". The definition section adds that the company "may update this metric to incorporate new products, feature offerings, or acquired technologies".

A metric whose numerator gains products while its year-ago comparison does not will report faster growth than the underlying business. The release does not say whether the prior-year figure was restated on the new basis. Until it does, I treat the 240% as an upper bound, not a like-for-like rate.

Salesforce also reports usage. It counts 3.2 billion Agentic Work Units in Q2, defined as "discrete tasks executed by AI agents in production". That is a genuinely useful volume signal and it is not revenue. Whether those tasks earn money depends on the pricing model, which is the argument made in the analysis of Salesforce as platform or roadkill.

ServiceNow AI ACV: a bookings figure dressed as a revenue milestone

ServiceNow's headline bullet for Q2 2026 reads: "ServiceNow AI crossed $1 billion in annual contract value in Q2 2026". It is the first line of the release, filed as an 8-K exhibit on 22 July 2026.

The release introduces the abbreviation ACV, as "net new annual contract value", and never defines it further. It does not say whether the $1 billion is cumulative contract value, value currently under contract, or something closer to ARR. Those are different numbers and a reader cannot reconstruct one from another.

ACV measures signed contracts, not revenue

Contract value is booked when a deal is signed. Revenue arrives as the service is delivered. For a subscription business the gap is usually small, but for AI add-ons sold with ramps, credits or consumption terms it can be large. A customer can sign for AI capacity that it does not use for a year.

ServiceNow also folded Now Assist into a new product called ServiceNow Otto, which "combines the intelligence of Now Assist, Moveworks, and AI Experience". The release does not say whether Moveworks contracts count toward the AI ACV figure. If they do, part of the $1 billion was acquired rather than grown.

The company has set a long-range goal of 30% of ACV coming from AI by 2030. That target is expressed in the same undefined unit. I would not underwrite a valuation on it until the unit has a printed definition. The wider investor reaction to ServiceNow's AI story is dissected in the ServiceNow bear case.

Adobe AI-first ARR: a growth rate with no published base

Adobe is the clearest example of a disclosure getting less informative over time. It is worth tracing release by release, because each step looks reasonable on its own.

From dollar figures to growth rates

In its Q3 FY25 release of 11 September 2025, Adobe gave two numbers. Chief executive Shantanu Narayen said AI-influenced ARR was "surpassing $5 billion" and AI-first ARR was "already exceeding our $250 million year-end target".

The Q4 FY25 release in December 2025 defined AI-influenced ARR and gave no figure for either metric. The Q1 FY26 release in March 2026 said AI-first ARR "more than triples year over year", again without dollars. The Q3 FY26 release of 10 September 2026 said AI-first ARR "grew more than 150% year over year", and nothing else about it.

Arithmetic tempts here. If AI-first ARR exceeded $250 million at the end of Q3 FY25 and grew more than 150% over the following year, it would exceed $625 million now. That is my calculation, not Adobe's, and it only holds if the product set inside "AI-first" stayed constant. Adobe has never published that product set in a release, so the calculation cannot be checked.

AI-influenced is a product tier, not AI spend

The one AI metric Adobe did define is the broader one. "Total AI-Influenced ARR reflects ARR from the product offerings and tiers that the customers select that are enhanced by AI features and functionalities embedded in the product". The same release says the included tiers are updated periodically.

Read that closely. It counts the full ARR of any tier that contains AI features. A customer who buys a Creative Cloud plan for the photo editor and never touches generative fill still counts. It measures how much of the catalogue has AI in it, which is a product decision, not customer demand. On this basis almost any software company could report most of its ARR as AI-influenced within two years.

Microsoft and Workday: thinner AI monetization data

Two other companies show what happens at the low end of the disclosure range. One publishes an enormous number with no definition. The other publishes a ratio with no denominator.

Microsoft: the largest number, the thinnest definition

Microsoft's FY26 Q3 release of 29 April 2026 quoted Satya Nadella: "Our AI business surpassed an annual revenue run rate of $37 billion, up 123% year-over-year". On the call, chief financial officer Amy Hood repeated it. Neither the release nor the prepared remarks say what the AI business includes.

That omission is not a small one. The figure could include Azure capacity sold to model developers, Azure AI services, Microsoft 365 Copilot seats, GitHub Copilot and more. Some of those are infrastructure resale and some are software. They carry very different margins, which is the point made in the contribution margin test for AI features.

The FY26 Q4 release, filed on 29 July 2026, did not repeat the run rate. It reported instead that Microsoft 365 Copilot reached over 30 million paid seats. A seat count is useful, but it is a different metric, and it replaced the dollar figure rather than sitting beside it.

Workday: a share of bookings with no base

Workday's Q2 FY27 release of 27 August 2026 quoted chief executive Aneel Bhusri on "AI driving more than 25% of our new ACV". It also said more than 5,500 customers use at least one Workday agent, up more than 35% from the prior quarter.

Neither number comes with a dollar base. "Driving" is not defined, so a deal where AI was one line item may count in full. The release contains no AI ARR figure at all. Secondary coverage of the quarter circulated an AI ARR dollar figure that does not appear in the release. I did not use it, because a number a company will not print is a number it will not defend.

Ranking AI revenue metrics by what they tell investors

To rank the seven metrics I scored each on five criteria, 0 to 2 points each, for a maximum of 10. The weights are equal and the rubric is mine. It is published here so you can disagree with a specific score rather than with a vibe.

The first criterion is a specific dollar figure; thresholds such as "crossed" score 1. The second is a written definition in the release. The third is a unit tied to contracted recurring value rather than bookings or run rate. The fourth is a stable scope over the past year. The fifth is separability, meaning the metric isolates AI rather than bundling it with something else.

Scores reflect the primary releases listed in the references as of 10 October 2026. Adobe's AI-influenced dollar figure scores 0 because the last one was published in September 2025. Change any single score by one point and the top and bottom of the ranking do not move.

The result I did not expect is the top row. Salesforce's most informative AI metric is the one least about AI, because it bundles a data platform in. Its purer Agentforce figure loses points for the scope change. You have to read both together to learn anything.

The AI attach rate nobody publishes

An AI attach rate is the share of a company's customers, or its revenue base, that pays for an AI product. It is the metric operators most want and the one no company on this list prints. Every disclosure gives a numerator and withholds the denominator.

Workday reports 5,500 customers using an agent, but not as a share of its customer count. Microsoft reports 30 million paid Copilot seats, but not as a share of its commercial seats in the same release. Salesforce reports work units, not paying accounts. Each is a fact. None is an attach rate, and AI attach rate is the number that shows whether pricing works, as the hybrid pricing and net retention gap argues.

Where this ranking is weakest

The rubric rewards Salesforce for writing things down. A company can publish a precise definition of a metric that is still unhelpful. Salesforce's combined figure scores highest while telling you little about AI specifically. A different weighting, with separability counted double, would put Agentforce ARR on top and pull the combined metric to the middle.

I only used earnings releases and 8-K exhibits. Investor presentations, call transcripts and annual reports may carry fuller definitions, especially for ServiceNow and Adobe, whose releases point readers to supplementary materials. If ServiceNow defines AI ACV in a slide deck, its score should rise by up to 2 points. I did not check every deck, and that is a real gap.

There is also a fair case for the companies that disclose less. A voluntary metric that keeps changing invites securities litigation when it disappoints. Adobe's move to growth rates may be legal caution, not concealment. A company that prints nothing cannot be accused of moving the goalposts, and that asymmetry punishes the firms trying hardest.

Finally, seven metrics from five companies is not the whole market. Many listed software companies publish no AI revenue figure at all. They are absent from the table because there is nothing to score, which makes the sample biased toward the most forthcoming firms.

Frequently asked questions

Where to start

If you track these companies, build your own version of the reference table and update it the day each release lands. Copy the definition text verbatim, not the headline number. The moment a definition changes, or a dollar figure turns into a growth rate, is the moment the story changed, and the market rarely flags it.

If you run a SaaS company and are deciding how to report AI revenue to your board, pick ARR of active AI subscriptions, print the definition, and restate the prior year whenever scope changes. Then add the attach rate nobody else publishes. It costs you one denominator and it buys more credibility than any growth multiple, as the gap between AI spending and reported results described in the AI spending and EBIT impact analysis makes plain.

Related reading