Agent Usage Outran Agent Revenue Nearly Four to One. That Gap Is Repricing Voice of Customer.

What did Salesforce actually report on 26 August?

Salesforce closed its second quarter on 31 July and reported on 26 August. Revenue was 11.345 billion dollars, up 10.83% year on year. Adjusted earnings per share were 5.90 dollars and net income was 4.844 billion dollars, up 73.31%. Current remaining performance obligation stood at 33.5 billion dollars, up 14% in constant currency. The company also booked a 2.6 billion dollar gain on strategic investments, which included its stake in Anthropic.

Two figures matter more than the headline. Agentforce annual recurring revenue reached 1.5 billion dollars. And customers drove 3.2 billion Agentforce usage units in the quarter, which the company reported as up 97% quarter on quarter.

The gap between what customers used and what they paid

Agentforce annual recurring revenue was 1.2 billion dollars a quarter earlier. Against the 1.5 billion reported for the second quarter, that is growth of 25% quarter on quarter. This next step is our arithmetic rather than a disclosed figure: usage rising 97% while revenue rises 25% means consumption grew close to four times faster than the revenue attached to it.

There are innocent readings. Committed-use contracts get consumed unevenly. Trials and internal usage inflate unit counts before they attach to a line item. Bundled entitlements let a customer raise consumption without raising spend. Every one of those readings has the same consequence for a buyer, which is that the effective price of a unit of automated work fell during the quarter, and is likely to keep falling while vendors compete on consumption growth as a proof point.

What the same period did to the measurement layer

While the resolution layer was being bid up, the feedback layer was being written down, and the numbers are not close.

On 22 April, Thoma Bravo agreed to hand Medallia to its lenders in a debt-for-equity swap. The transfer wiped out 5.1 billion dollars of equity against a 6.4 billion dollar take-private completed in 2021, with 2.8 billion dollars of debt outstanding and a syndicate including Blackstone, KKR, Apollo Global and Antares Capital converting into majority ownership. It is the largest equity write-off in Thoma Bravo's history and among the largest private credit restructurings recorded.

On 18 May, Qualtrics finalised its 6.75 billion dollar acquisition of Press Ganey Forsta. Reporting on the financing describes banks carrying more than 500 million dollars in paper losses on a 5.3 billion dollar debt deal, and JPMorgan halting that deal over concerns about disruption from artificial intelligence.

Set those two against each other, which is again our arithmetic. Qualtrics paid more for Press Ganey Forsta alone than Thoma Bravo paid for the whole of Medallia four years earlier. Both companies were named Leaders in the same Gartner Magic Quadrant for voice of customer platforms published in March.

Why the two halves moved in opposite directions

The convenient explanation is that artificial intelligence commoditised feedback analysis, and there is something in it. The more uncomfortable explanation sits in the category's own research.

Medallia's published research finds that between 30% and 40% of departments take no action on the customer experience insights they receive. Its 2026 State of Customer Experience report has 66% of brands believing experience is improving against 17% of consumers who agree. Qualtrics reported in its own 2026 consumer trends work that roughly one in five consumers saw no benefit at all from artificial intelligence in customer service, and that 30% say nothing after a bad experience.

Read together, those are vendors documenting the weakness that repriced them. A category whose own evidence says a third or more of its output is never acted on is selling reporting rather than outcomes, and reporting is exactly what gets cheaper when a model can do the reading. The resolution layer is being bid up because a resolved conversation is self-evidently an outcome. Nobody has to be persuaded that it happened.

What a buyer should do differently this quarter

Do not sign a multi-year committed-use rate on agentic customer service at this quarter's unit price. The disclosed direction of travel is that unit economics are moving in the buyer's favour, and a three-year rate locks you out of that.

Price measurement and resolution separately even when a single vendor sells both, because the two are being valued in opposite directions and a bundle conceals which half carries the cost. Ask for a written definition of a resolved conversation before comparing any two resolution rates, including how handoffs, abandonments and reopened tickets are treated. And when a platform in the measurement layer changes hands, ask the new owner about roadmap funding rather than continuity of service, because a creditor recovering a position and a sponsor pursuing growth do not invest alike.

What we could not establish

Several things, and they bound everything above. The 76% end-to-end resolution rate published for Fin is a vendor claim that cannot be checked without the denominator. Medallia and Qualtrics are both privately held, so there is no filed revenue, retention or customer count to test either the write-down or the purchase price against. Salesforce does not break out how much of the 1.5 billion dollars of agent revenue is net new against migrated from existing service licences, so the growth rate cannot be separated from repackaging. And a usage unit is a vendor-defined measure, which means the 97% figure is precise but not necessarily comparable to anything outside Salesforce.

We have published the numbers we could source and named the ones we could not. The reasoning above stands or falls on figures that are all traceable, and where a figure is ours rather than a company's, the sentence says so.

FAQ

Why does the gap between usage and revenue matter to a buyer?

Because it prices your next renewal. Salesforce reported agent usage up 97% against agent revenue up 25% in the same quarter, which is our arithmetic on its own two disclosures. Consumption running roughly four times ahead of revenue means the effective price per unit of work is falling. A multi-year committed-use rate signed at this quarter's price is unlikely to look competitive by the time it renews.

Is the 76% resolution figure independently verified?

No. It is Fin's own published figure for its AI Agent, and we have reproduced it as a vendor claim rather than a measured result. The number that would make it checkable is the denominator, meaning what the vendor counts as a resolved conversation and whether a handoff, an abandonment or a reopened ticket counts against it. No vendor in this category currently publishes that definition.

Does the Medallia restructuring mean the platform is going away?

Nothing in the announcement says so. Ownership moved from Thoma Bravo to the lending syndicate through a debt-for-equity swap, which changes who owns the business rather than whether it operates. What it does change is the pressure the business is under, because the new owners are creditors recovering a position rather than a sponsor pursuing growth.

What should a shortlist look like after a quarter like this?

Split it. Price measurement and resolution as two purchases even when one vendor sells both, because the two halves are being repriced in opposite directions and bundling hides which half you are paying for. Then ask each vendor for its resolution definition in writing. A vendor that will not put the denominator in a contract is quoting a marketing number.