Salesforce reported second quarter fiscal 2027 results on 26 August 2026. Revenue came in at $11.345 billion, up about 11% year over year. Agentforce annual recurring revenue passed $1.5 billion, up more than 240% year over year. The stock rose roughly 11% in premarket trading and management raised full year guidance to a range of $46.1 billion to $46.4 billion.
Those are the numbers that led the coverage. For anyone who owns a Salesforce budget, they are not the important ones.
The figure that matters sits further down the release. Customers consumed 3.2 billion Agentic Work Units in the quarter, up 97% from the previous quarter. Not year over year. Quarter over quarter.
Agentic Work Units, usually shortened to AWUs, are the unit Salesforce meters Agentforce against. ARR growth tells you how many companies signed contracts. AWU growth tells you how hard the agents are working inside those contracts, and that is the line that shows up on your invoice.
What is an Agentic Work Unit?
An AWU is the billing unit for agent activity in Agentforce. Where the older seat model charged per user, the consumption model charges for work performed: an agent retrieving records, reasoning over them, calling an action, producing a response.
The practical consequence is that your Agentforce cost is no longer set at contract signing. It is set by usage, and usage is decided by how many agents you run, how many things each one can do, and how often people or systems trigger them. Those three variables move independently of your headcount.
We covered how the pricing model works in Agentforce pricing explained, and the wider cost framework in the AI agent pricing and TCO guide. This piece is narrower. It is about what the Q2 number implies for your own forecast.
What Salesforce actually reported
The relevant figures from the quarter:
- Agentforce ARR above $1.5 billion, up more than 240% year over year
- Agentforce and Data 360 combined ARR near $3.9 billion, up more than 210% year over year
- 3.2 billion AWUs consumed, up 97% quarter over quarter
- Current remaining performance obligations of $33.5 billion, up 14% in constant currency
- Bookings from premium Agentforce and Slack bundles more than doubled sequentially
Read the ARR and the AWU figures together. ARR grew 240% over twelve months. Consumption grew 97% over three. Even allowing for new customers arriving during the quarter, consumption per contract is climbing, not flattening.
That is the part worth planning around. A platform metric is not your metric, and your own curve may look nothing like the aggregate. But if the direction holds across the installed base, the safe assumption for a 2027 budget is that per agent consumption rises over time rather than settling.
Why consumption growth matters more to you than ARR growth
ARR is a vendor metric. It measures what Salesforce booked. Consumption is a customer metric. It measures what you used, and under a consumption model those are the same thing as what you owe.
The gap between them is where budget surprises live. A team can sign a contract in Q1 with a credit allocation that looks generous, deploy three agents in Q2, add actions to those agents through Q3 because adding actions is easy, and arrive at Q4 having consumed the allocation with a quarter still to run. Nothing went wrong in that sequence. No one made a mistake. The cost curve simply followed the capability curve, and nobody was tracking the capability curve.
This connects to something we wrote about earlier in the month. Salesforce's own Agentic Enterprise Index showed the average org going from five agents to thirteen, and average skills per agent going from two to six. That is roughly an eightfold increase in the number of distinct things an org's agents can do. We called it the agent action surface and argued it was a security problem.
The Q2 consumption number suggests it is also a cost problem. More agents with more skills produce more AWUs. The security exposure and the invoice are driven by the same underlying growth, which means one inventory answers both questions.
How to forecast your own AWU consumption
Four steps. None of them require a finance background.
Establish your current run rate. Pull your AWU consumption for the last three months from your Agentforce usage reporting. Monthly, not quarterly, because quarterly averages hide the trend inside them. If you cannot get monthly granularity, that gap is itself worth raising with your account team before you plan anything else.
Divide by agent, then by action. Total consumption is not actionable. Consumption attributed to a specific agent, and ideally to a specific action within that agent, tells you which capability is expensive. In most orgs the distribution is heavily skewed. A small number of high traffic actions account for most of the spend, and finding them takes an afternoon.
Model the capability curve, not the user curve. This is where most forecasts go wrong. Teams project agent cost from expected user growth, because that is the habit the seat model taught. Under consumption billing the driver is how many agents you run multiplied by how many actions each can take, multiplied by trigger frequency. Forecast those three, not headcount.
Add a planned increment. Your agents will gain actions over the year, because that is what the platform is designed to make easy and because the Index shows it happening across the installed base. A forecast that assumes today's capability set is a forecast of a system nobody intends to keep static. Decide the increment deliberately and write it down, rather than discovering it in Q4.
What to put in front of your CFO
Three things, and they fit on one page.
The first is your current monthly AWU run rate with the trend line across the last three months. The second is the concentration figure: what share of consumption comes from your top five actions. The third is your forecast under two scenarios, one where capability stays flat and one where it grows at the rate you actually expect.
The gap between those two scenarios is the number to discuss. It is also the number that justifies governance work, because the cheapest way to control consumption is to stop deploying actions nobody uses, and that requires the same action inventory that security asked for.
Two things not to do. Do not present the platform wide 97% figure as your forecast, because it is an aggregate across a customer base that includes companies at very different stages. And do not present ARR growth to a CFO at all. It is Salesforce's number, it is a good number, and it has nothing to do with your budget.
The honest caveats
Salesforce does not publish an average AWU price, and pricing varies by agreement, bundle and volume commitment. So the 3.2 billion figure cannot be converted into a per customer cost from public information. Treat it as a directional signal about consumption behaviour, not as an input to a spreadsheet.
The 97% figure is also a single quarter. One quarter is a data point, not a trend, and it covers a period when Agentforce was expanding rapidly across new accounts. Some of that growth is new customers rather than existing customers consuming more. Salesforce did not break out the split, and until it does, the honest reading is that consumption per contract is rising without knowing precisely how fast.
What is not ambiguous is the direction, or what it means for planning. Under a consumption model, the budget question is not what you signed. It is what your agents do, how often they do it, and whether anyone is watching that number between contract renewals.
If nobody at your organisation currently owns the AWU run rate, that is the gap to close before the next planning cycle. It is a smaller job than it sounds, and it is considerably smaller than the conversation that follows an overrun.
For the governance structure around this, see our practical AI governance programme. For the security side of the same inventory, see securing AI in Salesforce.
*Figures from Salesforce's second quarter fiscal 2027 results, published 26 August 2026, and its 2026 Agentic Enterprise Index. Enterprise Dreamin' has no commercial relationship with Salesforce and this is analysis rather than investment advice.*