Salesforce Agentforce ARR, Every Disclosed Figure: $1.5B Is 3.3% of Revenue

Salesforce says its Agentforce ARR exceeded $1.5 billion at 31 July 2026. Set against the quarter's revenue run rate of about $45.4 billion, that is roughly 3.3% of the business. Over the past 7 quarterly releases, Salesforce has published Agentforce ARR, deal counts, paid deals, tokens and agentic work units, and it has stopped publishing three of them. This post lines up every figure, with its date and source, before the next print.

Every Agentforce ARR figure Salesforce has disclosed, in order

Annual recurring revenue, or ARR, is the annualised value of subscription contracts active at the end of a period. Salesforce defines its own AI ARR metrics, and it has changed the definition and the name of the headline metric twice since early 2025. That is why a single quarter's figure tells you very little, and the sequence tells you a lot.

The table below collects every AI ARR, deal, token and work-unit figure from Salesforce's quarterly results releases and its October 2025 investor day. Each row is the company's own wording, rounded the way Salesforce rounded it. Where a cell is blank, the company did not publish that figure in that release.

All figures are from Salesforce results releases and the 15 October 2025 investor day release. "Over", "nearly" and "half" are the company's words. AWU is Agentic Work Unit. Deals and tokens are cumulative since launch.

The Data Cloud and AI era: February to September 2025

For the first three releases of 2025, Salesforce reported one blended number. The $900 million of Data Cloud and AI ARR it gave in February 2025 combined its customer data product with every generative AI subscription. Agentforce was inside it, but nobody outside the company could see how much.

The September 2025 release spelled out the scope. Data Cloud and AI ARR was the annualised value of active Data Cloud and certain generative AI subscriptions, including Agentforce and standalone generative AI products. In other words, the metric always held more than agents. That matters later, when the company began quoting a growth rate for Agentforce on its own.

The Agentforce ARR era: December 2025 onward

The first standalone Agentforce figure arrived at the October 2025 investor day. Salesforce said that, combined with Agentforce, its total agentic AI ARR in Q2 was approximately $440 million. Note the wording. It is an agentic AI basket, not a pure Agentforce SKU count.

From December 2025 the quarterly release carried an Agentforce line. It went from over $0.5 billion to $800 million, then $1.2 billion, then over $1.5 billion. In the same period the headline metric was renamed Agentforce and Data 360 ARR and absorbed Informatica, which the separate analysis of how much of that figure was bought rather than built takes apart in detail. This post does not repeat that subtraction. It asks a different question: what did each release choose to show, and what did it stop showing?

The Agentforce metrics Salesforce stopped reporting

A company choosing which numbers to publish is normal. A company swapping its adoption metric every quarter or two is a pattern worth recording, because the metric that disappears is usually the one that stopped flattering the story. We are not claiming that is what happened here. We are claiming you cannot tell from the outside, which is reason enough to keep the ledger.

Deals and paid deals

Deal counts were the first Agentforce scoreboard. The cumulative total rose from 5,000 in February 2025 to over 29,000 in February 2026. The quarterly additions grew each time: about 3,000, then 4,500, then 6,000, then 10,500.

The paid share was steady while it lasted. It was about 60% in February 2025, half in May, 48% in September and 51% in December 2025. Then the paid figure disappeared, one release before the total deal count also disappeared. Neither has returned.

Our read is that deal counts were always a weak metric. A free pilot and a seven-figure contract both count as one deal. Dropping them is defensible. Dropping them without a paid customer count to replace them is less so, because paid adoption breadth is exactly what a buyer wants to know.

Tokens, then Agentic Work Units

Token counts had an even shorter life, and the three figures do not sit on one scope. In December 2025 Salesforce said Agentforce had processed more than 3.2 trillion tokens through its LLM gateway. In February 2026 the company said Salesforce had processed more than 19 trillion tokens to date, up 5 times year on year. That is a six-fold jump in one quarter, and the subject of the sentence changed from Agentforce to Salesforce.

In May 2026 the release gave 28.6 trillion tokens to date, up 152% quarter on quarter. The cumulative total rose about 51% from 19 trillion, so the 152% must describe something else, probably tokens in the quarter. The release does not say. By August 2026 tokens were gone.

Agentic Work Units replaced them. Salesforce defines an AWU as a measure of discrete tasks executed by AI agents in production across its platform, including Agentforce and Slack. The cumulative count went from 2.4 billion to 3.8 billion to 7.0 billion, with 3.2 billion in the July 2026 quarter. Simple subtraction puts the April quarter at about 1.4 billion, so 3.2 billion would be growth near 129%. Salesforce reported 97%. We could not reconcile the two, and the gap is too large to be rounding.

What the growth rates imply about earlier Agentforce ARR

Salesforce never published Agentforce ARR before October 2025. It did publish growth rates against year-ago figures. Divide each disclosed figure by one plus its growth rate and you recover the base the company was comparing against. Nobody else appears to have published this series, so here it is.

Zan Digital calculation: disclosed ARR divided by (1 + stated growth). Because Salesforce writes "over" for both some figures and some growth rates, each implied base is approximate. The July 2025 result matches the roughly $440 million Salesforce disclosed at its October 2025 investor day, which is a useful check on the method.

Why the October 2024 base matters

The first implied point is the strangest. Salesforce's February 2025 release counted Agentforce deals "since October", so the product had barely launched when the quarter ended on 31 October 2024. Yet the growth rate implies roughly $116 million of ARR at that date.

The likely explanation is scope, not error. The investor day wording, "combined with Agentforce, its total agentic AI ARR", suggests the series includes generative AI products that existed before Agentforce. If so, the 330% growth rate is measured partly against older AI add-ons. That is a fair way to build a metric. It is not the same as Agentforce growing from zero, and we think the release should have said so.

The middle of the series is the other surprise. Between January and July 2025, implied ARR rose from about $297 million to about $441 million, which is slow for a product the company called its fastest-growing. The acceleration came later, from late 2025, which lines up with the edition and pricing changes covered in the breakdown of Agentforce editions and bundled credits.

Agentforce ARR as a share of a $41.5B revenue base

Salesforce booked $41.5 billion of revenue in fiscal 2026, the year to 31 January 2026. Against that base, more than $1.5 billion of Agentforce ARR is about 3.6%. Against the latest quarter's run rate, it is about 3.3%. Either way, Agentforce is a small business inside a large one.

The ratio below divides each quarter's AI ARR by that quarter's total revenue multiplied by four. ARR is contracted run rate and revenue is recognised income, so this is a scale comparison, not an accounting identity. It is still the clearest way to see how much of Salesforce the AI story actually covers.

Zan Digital calculation from Salesforce results releases. Revenue is total revenue as reported. The July 2026 Informatica figure is an assumption, because that release did not state it. Shares use Salesforce's rounded ARR figures.

Two things stand out. First, the headline AI share jumped from 3.4% to 6.5% in one quarter, and most of that step was the Informatica closing. Second, the non-Agentforce, non-Informatica slice barely moved as a share of the business. Agentforce's gain explains almost all of the organic increase.

We think that second point is underrated. Data 360 was the original half of this story, and as a share of Salesforce it has not grown much. Some of that is definitional, because Slackbot and Headless 360 moved into Agentforce in July 2026. Some of it may simply be a product that grows at a normal software pace.

Where this Agentforce ARR ledger is weakest

Every number above is Salesforce's own. That is the strength of the ledger and its main limit. There is no independent source for Agentforce adoption, and three caveats apply to everything here.

ARR is not revenue, and Salesforce chooses the scope

ARR is not audited and has no standard definition. Salesforce has moved the Agentforce boundary at least once without restating the prior year. The share-of-revenue ratios compare a contracted run rate with recognised revenue, which is why we label them directional. Our implied early series also inherits every rounding word Salesforce used, so each point could be off by tens of millions of dollars.

The $53,000 ARR per paid deal is weaker still. Deals are not customers, a customer can sign several deals, and Agentforce ARR in that quarter included products sold outside any Agentforce deal. Treat it as an order of magnitude, not as an average contract size.

The case that disclosure drift is normal

There is a fair defence of Salesforce. Early in a product's life, deal counts and token counts are the only signals available. Once revenue is material, ARR is the better measure, and switching to it is maturity rather than evasion. Agentic Work Units also try to measure work done, which is closer to customer value than tokens consumed.

Salesforce also discloses more than most peers. Many large software vendors give no standalone AI ARR at all, as the ranking of AI revenue disclosure metrics across vendors shows. A vendor that publishes a changing set of numbers still beats one that publishes none. Our objection is narrow: every metric change should come with one restated prior period, so the series stays readable.

What to check in the Q3 fiscal 2027 Agentforce ARR print

Salesforce's third quarter of fiscal 2027 ended on 31 October 2026. As of 2 November 2026 the company had not announced the results date. In each of the past two years it reported third quarter results on 3 December. Guidance calls for revenue of $11.42 billion to $11.50 billion, including slightly above 4 points of Informatica contribution.

The useful number is the dollar add, not the growth rate. The year-ago base was "surpassed half a billion", so almost any figure will produce a growth rate above 200%. The table below sets out what different Agentforce ARR readings would mean. These are scenarios for reading the print, not predictions.

Share uses the midpoint of Q3 revenue guidance, $11.46 billion, multiplied by four. Scenario values are illustrative reading thresholds chosen by Zan Digital, not company guidance or estimates.

Four lines to search for in the release

First, the definitions footnote. Salesforce named two pending acquisitions, Contentful and Fin, in the August release. If either closes and lands inside an AI ARR metric, the add will include bought ARR again. Second, whether an Informatica Cloud ARR figure returns. Third, whether any deal or paid customer count reappears. Fourth, whether the Agentic Work Unit history is restated, which would explain the growth rate we could not reconcile.

If none of the four appears, the Agentforce line is still informative. It just carries less weight than its growth rate suggests. We would watch the third scenario row most closely, because a large add with a quiet definition change is the outcome the ledger exists to catch.

What the Agentforce ARR ledger means if you are buying

You are probably not trading Salesforce stock. You may be in an Agentforce renewal or pilot conversation where "tens of thousands of companies" is part of the pitch. The ledger gives you better questions than the slide does.

A product at about 3% of its vendor's revenue is still early. That cuts both ways. Salesforce has strong reasons to discount and bundle to grow the line, which helps your negotiation. Pricing and packaging are also still moving, which hurts your forecasting, as the arithmetic of Agentforce's per-conversation pricing shows.

The paid-deal ratio is the other lesson. In every quarter it was published, roughly half of Agentforce deals were paid. If your account team cites adoption, ask which half your peers are in. The broader question of whether Salesforce ends this cycle as the agent platform or as the incumbent agents route around is argued in the case for and against Salesforce in the agent era.

The same discipline applies to every incumbent. ServiceNow gives a different AI metric on a different basis, as covered in the analysis of ServiceNow's Now Assist ACV target. Put the two side by side and neither tells you what your own deployment will return.

Frequently asked questions

Where to start before the December print

Copy the ledger table into your own vendor file now, while the last eight releases are easy to find. When the Q3 release lands, add one row and fill only the cells Salesforce fills. Leave the blanks blank. The blanks are the information.

If Agentforce is on your renewal list, ask your account team one written question before the print: how many paying Agentforce customers does Salesforce have today? A vendor confident in breadth of adoption can answer it. Record the answer, or the refusal, next to the ARR figure.

Related reading