On September 16, 2026, Salesforce showed investors a four-stage curve for its customers. At the last stage, a company running Agentforce "wall to wall internally and externally," the projected ARR uplift is "3x-4x+."
We read that slide as the buyer's budget forecast, not only the seller's. A business case built on removed seats now meets a vendor model that pairs some seat optimization with a larger bill.
What Salesforce filed
The deck is Exhibit 99.1 to a Form 8-K dated September 16, 2026. It names four stages. "Fundamental" customers average three clouds, primarily on mid-tier editions. "Pre-Agentic" customers add clouds and Industries, at a 1.2x ARR uplift.
"Agentic" customers adopt Agentforce "in support and core expansion, with some seat optimization," at 1.5x to 2x. The last stage reaches 3x to 4x or more. A footnote bases the uplift on "compounding growth projections" for a customer cohort (Salesforce, Investor Day deck). These are projections, not prices.
One measured point sits beside them. For its top 100 customers by agentic work units, the same deck reports more than 2x ARR uplift in the 18 months since Agentforce launched.
Where the business sits today
In the quarter ended July 31, 2026, Agentforce ARR exceeded $1.5 billion, up more than 240 percent (Salesforce, Q2 fiscal 2027 results). From that quarter, the measure includes Slackbot and Headless 360, so part of the growth is scope. Fiscal 2027 revenue guidance is $46.1 billion to $46.4 billion.
Set side by side, Agentforce ARR is about 3 percent of that guidance, on two different measures. The Agentforce Apps line, which groups Sales, Service, Marketing, Commerce, flex credits and Slack, grew 8 percent to $7.19 billion in the quarter (Salesforce, Form 10-Q). The investor curve describes how Salesforce plans to move the rest of its base.
The same 10-Q names the tension. Attrition can rise with "decreases in the number of users at our customers." It also states: "We have limited experience with determining optimal pricing for our consumption-based contracts." The seller is still learning the price. So is the buyer.
The break-even the slide implies
What follows is our analysis, on an illustrative account. Take a customer with $10 million of Salesforce ARR, and read each multiple as applying to that starting ARR. At 1.5x to 2x, the annual bill rises by $5 million to $10 million. At 3x to 4x, it rises by $20 million to $30 million.
The Bureau of Labor Statistics put the mean annual wage of US customer service representatives at $46,590 in May 2025, across 2,595,750 jobs (BLS, OEWS). In wages alone, before benefits, $20 million equals about 429 representatives. $30 million equals about 644.
Research on assisted support gives a second yardstick. Across 5,179 agents, a generative AI assistant raised issues resolved per hour by 14 percent on average (Brynjolfsson, Li and Raymond, QJE 2025). If a gain of that size had to fund $20 million, the payroll behind it would be about $143 million, or some 3,066 representatives.
At assistant-grade gains, each extra dollar of software needs about seven dollars of support payroll behind it. The case has to rest on work the agent finishes alone. Hikari Blue · operator note
That is where pricing now sits. Salesforce lists Flex Credits at $500 per 100,000 credits, with 20 credits per Agentforce action, which is $0.10 an action (Salesforce, Agentforce pricing). At list, $20 million buys 200 million actions a year. The seller prices the finished unit of work. The buyer should measure it the same way.
The honest counterpoint
The multiples describe spend, not margin. A customer whose ARR triples may also have grown, added business lines or retired other vendors. The top 100 customers chose to more than double their spend, and some likely measured returns we cannot see. Sales and marketing agents can add revenue rather than remove cost. We use support labor because it is the most common agentic case, not the only one.
What management does now
Ask the account team where your company sits on the four stages, and what ARR its three-year plan assumes for you. On our reading, the investor curve is the sum of those account plans.
Rebuild each agentic case on cost per resolved case: labor, seats, credits and integration, for the same queue in both periods. A case that counts only removed seats omits the credits that replace them.
Then write the contract to match. Fix the price per action for the term. Set consumption caps and alerts. Tie seat reductions at renewal to measured deflection. Keep agent logic and case history portable. In its results release, Salesforce says its AIforce harness makes its data and actions "available to every agent, model, and interface." For a buyer, that sentence makes the agent layer comparable across vendors, which is the premise of a model-agnostic architecture.
What slows execution
Seat reductions arrive at renewal. Consumption bills arrive monthly. Labor savings need workforce redesign, usually through attrition, which takes quarters. Deflection is hard to count when agents hand off half-finished cases. Without a baseline set before launch, neither side can prove the saving later.
Holding the vendor curve, the labor ledger and the deflection rate in one view is ordinary work for an AI operating layer. It is also the AI strategy question a CFO asks first.
If our Agentforce spend reaches the stage Salesforce projects for us, which cost line falls by at least as much, and who measures it?
Carry one number into the next budget review: total cost per resolved case, labor plus software, for the same queue, month by month. When software rises faster than labor falls, the gain is accruing to the vendor, not to the buyer.
- Salesforce, Inc. (September 16, 2026). Form 8-K, Item 7.01, and Exhibit 99.1, Investor Day presentation, furnished to the US Securities and Exchange Commission. Primary source for: the four customer stages, "Fundamental" ("Averages 3 clouds, primarily on mid-tier editions"), "Pre-Agentic" (1.2x ARR uplift), "Agentic" ("Adopts Agentforce in support and core expansion, with some seat optimization," 1.5x-2x ARR uplift) and "Agentic Enterprise" ("Adopts Agentforce wall to wall internally and externally," 3x-4x+ ARR uplift); the footnote "uplift based on compounding growth projections from cohort of customers in the Agentic scenario"; and the top 100 customers by agentic work units with ">2x ARR uplift in 18 months since Agentforce launch." Read on the text of the exhibit; our reading is that each multiple applies to the ARR of a customer at the starting stage. sec.gov, Salesforce Form 8-K, September 16, 2026 · sec.gov, Exhibit 99.1 Investor Day deck
- Salesforce, Inc. (August 26, 2026). Second quarter fiscal 2027 results, Exhibit 99.1 to Form 8-K. Primary source for: Agentforce ARR above $1.5 billion, up over 240 percent year over year, and the note that "Effective Q2 FY27, Agentforce ARR includes our AI offerings, Slackbot and Headless 360"; full year fiscal 2027 revenue guidance of $46.1 billion to $46.4 billion; and the statement that AIforce is "making it available to every agent, model, and interface." sec.gov, Salesforce Q2 fiscal 2027 results
- Salesforce, Inc. Form 10-Q for the quarterly period ended July 31, 2026. Primary source for: Agentforce Apps subscription and support revenue of $7,193 million against $6,682 million, 8 percent growth, and its composition (Agentforce Sales, Service, Marketing, Commerce, Agentforce Apps Flex Credits and Slack); attrition factors including "decreases in the number of users at our customers" and "the increased prevalence of consumption-based pricing models"; and "We have limited experience with determining optimal pricing for our consumption-based contracts." sec.gov, Salesforce Form 10-Q, quarter ended July 31, 2026
- US Bureau of Labor Statistics. Occupational Employment and Wage Statistics, May 2025, customer service representatives (SOC 43-4051), national, all industries. Primary source for: mean annual wage of $46,590 (series OEUN000000000000043405104) and employment of 2,595,750 (series OEUN000000000000043405101). Wages exclude benefits. api.bls.gov, OEWS mean annual wage series · api.bls.gov, OEWS employment series
- Brynjolfsson, E., Li, D. and Raymond, L. R. (2025). Generative AI at Work. The Quarterly Journal of Economics, 140(2), 889-942; NBER Working Paper 31161. Primary source for: 5,179 customer support agents; access to a generative AI conversational assistant "increases productivity, as measured by issues resolved per hour, by 14% on average, including a 34% improvement for novice and low-skilled workers." The study measures an assistant to human agents, not autonomous agents; we use it as a yardstick, not as a forecast for Agentforce. nber.org, Working Paper 31161
- Salesforce, Inc. Agentforce pricing page, read on September 18, 2026. Vendor list prices, "subject to change": Flex Credits at $500 per 100,000 credits; "Agentforce actions are 20 Flex Credits"; $2 per conversation. salesforce.com, Agentforce pricing
- Our arithmetic, on an illustrative account of $10 million ARR that is not drawn from any filing. Uplift: 1.5x to 2x adds $5 million to $10 million; 3x to 4x adds $20 million to $30 million. Wage equivalents: 20,000,000 divided by 46,590 is 429.3; 30,000,000 divided by 46,590 is 643.9. Assisted-productivity yardstick: 20,000,000 divided by 0.14 is 142.9 million; divided by 46,590 is 3,066; 1 divided by 0.14 is 7.1 dollars of payroll per dollar of software. The yardstick assumes, as a simplification, that a 14 percent gain in resolutions per hour converts one for one into wage savings. Price per action: 500 divided by 100,000 is $0.005 per credit; times 20 is $0.10; 20,000,000 divided by 0.10 is 200 million actions. Agentforce ARR against revenue guidance: 1.5 divided by 46.4 is 3.2 percent; 1.5 divided by 46.1 is 3.3 percent; ARR and revenue are different measures, so the ratio is an order of magnitude only.
The Hikari Blue team · Austin, September 2026