.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Segmenting by Outcome Pricing for Business Impact
Enterprise software pricing becomes unstable when the thing being sold changes faster than the unit being charged. A seat made sense when a person logged into software and used it to do work. Consumption made sense when infrastructure use tracked both supplier cost and customer activity. AI agents complicate both assumptions because the software can now complete work itself - resolving a support case, qualifying a prospect, changing an order or booking an appointment.
Salesforce Agentforce makes the tension unusually visible. Agentforce launched in September 2024 with pricing starting at $2 per conversation. Salesforce then added action-based Flex Credits in May 2025, employee licences and unlimited-use add-ons in June 2025, and more ways to buy consumption in August 2025. By 13 August 2026, the public price list spans $2 conversations, $500 packs of 100,000 Flex Credits, $125-per-user employee add-ons, industry add-ons at $150 per user, and Agentforce 1 Editions from $550 per user.
The question is no longer whether Salesforce can meter digital labour. It clearly can. The harder question is whether it is charging for what customers ultimately value.
Monetizely's position is that customer-facing agents should be segmented by the business outcome they deliver, with a verified business outcome becoming the primary pricing meter. Flex Credits are useful operational plumbing, but an action is too far upstream from business impact to remain the main customer-facing unit of value.
Outcome pricing fails when an “outcome” is merely renamed usage
A sensible outcome meter begins with the customer's business, not with the agent's activity log. A support buyer cares whether an issue was solved. A sales leader cares whether a credible opportunity was created. A commerce operator cares whether a transaction was completed. None of those buyers becomes richer because an agent executed five actions instead of three.
That distinction is central to Monetizely's 5-Step Pricing Framework, developed further in Monetizing Agentic AI. The five steps are Goals and Segmentation, Packaging, Pricing Metric, Rate Setting, and Operationalization. Goals and Segmentation identifies which customers and jobs matter economically. Packaging turns those differences into offers a buyer can understand. The Pricing Metric chooses the unit that makes the bill grow as customer value grows. Rate Setting decides how much to charge for that unit. Operationalization then makes the design work through entitlements, metering, billing, contracting and reporting. The order matters here: choosing credits because they are easy to meter reverses the logic. We first need to decide what outcome a segment values, then determine how to measure it reliably.
Outcome pricing therefore should not mean one flat “outcome” price for every agent. A £500,000 enterprise purchase should not be priced using the same value logic as answering “Where is my order?” Segmenting the outcome is what connects price to business impact.
The practical dividing line is how directly the agent's work can be linked to a result the buyer already measures.
| Agent job | Buyer can verify | Value link | Monetizely's preferred primary meter |
|---|---|---|---|
| Customer support | Case resolved without a human taking over or the customer reopening it | High and immediate | Verified resolution |
| Sales development | Prospect meets agreed criteria and reaches an accepted next step | High, but definitions must be agreed | Accepted qualification |
| Commerce or service operations | Booking, order change, payment or other transaction completes | High and transactional | Completed transaction |
| Employee assistance | Research, drafting, summaries and internal advice | Useful but hard to isolate from human contribution | Per-user access with usage controls |
The table makes the principle concrete: move the price towards the outcome only as quickly as attribution allows. Employee copilots can remain seat-led because a better summary rarely has a clean standalone value. A customer-service agent that closes the entire case creates a much stronger charging event.
Peer-reviewed B2B pricing research points in the same direction. A March 2025 study in the Journal of Revenue and Pricing Management found that value-based pricing works better when suppliers develop the capability to quantify customer value, while research on AI business models has found that outcome contracts require explicit performance indicators, profit economics and incentive structures.
The implication for agent pricing is demanding but clear. The harder an outcome is to define and attribute, the less suitable it is as a meter. The easier it is to verify, the weaker the case for charging for the activities that produced it.
Salesforce moved faster on buying flexibility than on value alignment
Agentforce's pricing history is best read as a rapid search for a workable unit. The September 2024 launch price of $2 per conversation gave buyers a unit they understood, but conversations became less representative as Agentforce expanded from chat towards autonomous actions across service, sales, commerce and back-office workflows. Salesforce announced Flex Credits in 2025 specifically to charge for the actions agents perform, then expanded per-user packaging for employee agents.
The pricing evolution matters because each change fixed a genuine commercial problem while moving the offer in a different direction.
| Evidence date | Agentforce pricing structure | What changed economically | Primary source |
|---|---|---|---|
| 12 Sep 2024 | From $2 per conversation | Simple customer-facing consumption unit at launch | Salesforce investor release; Wayback October 2024 archive |
| 15 May 2025 | Flex Credits introduced at $500 per 100,000; standard actions use 20 credits | Meter moved from whole interactions towards discrete agent work | Salesforce pricing history and subsequent official pricing page |
| 17 Jun 2025 | Employee add-ons from $125/user/month; Agentforce 1 Editions from $550/user/month | Employee-agent usage could be packaged around a person rather than every action | Salesforce pricing update |
| 19 Aug 2025 | PayGo, pre-commit and pre-purchase | Commitment structure became separable from the usage meter | Salesforce payment announcement |
| 13 Aug 2026 | $500/100,000 Flex Credits; $2/conversation; $125 employee add-on; $150 industry add-on; Agentforce 1 from $550/user/month | Multiple meters now coexist across customer and employee use cases | Current Salesforce pricing page |
The progression is commercially sophisticated. It also leaves Salesforce with more ways to pay without yet producing a sufficiently strong hierarchy of what customers should pay for.
Marc Benioff captured the transition directly on Salesforce's 26 February 2025 earnings call: “Agentforce is a consumption product.” On the same call, President and COO Brian Millham said, “The pricing models will change over time,” and predicted a move away from conversation pricing towards a universal credit.
Salesforce executed that change quickly. Yet the current Flex Credit definition exposes the remaining problem. On 13 August 2026, Salesforce describes an action as functions such as updating a record, summarising a case, answering a product enquiry or executing a prompt or flow. Standard Agentforce actions cost 20 credits; at $500 per 100,000 credits, that works out to $0.10 for a standard action.
Consider Salesforce's own current examples. A “Where is my order?” request uses two actions and costs $0.20. A case-management workflow uses three actions and costs $0.30. A field-service scheduling example uses five actions and costs $0.50.
Those are excellent measures of work performed. They are not necessarily measures of value created.
Our scorecard reflects that distinction. The grades cover the three parts of the 5-Step Pricing Framework that matter most to this teardown.
| 5-Step Framework step | Grade | One-line rationale |
|---|---|---|
| Packaging | A- | Salesforce now separates employee and customer-facing deployment, supports consumption and licences, and offers several commitment structures, but the menu has become harder to explain. |
| Pricing Metric | B- | Conversations and actions are far better than forcing autonomous agents into seats, yet both primarily measure activity rather than a verified customer result. |
| Operationalization | A | Digital Wallet, action-level metering, PayGo, commitments and entitlements give Salesforce unusually strong infrastructure for changing the meter again. |
The scorecard identifies the central asymmetry: Salesforce's systems are ready for a more advanced pricing model before its customer-facing metric is.
Its filings strengthen that conclusion. Salesforce's fiscal 2026 10-K says Agentforce customers can observe agents, measure performance and see what agents are doing. The April 2026 proxy then reports 2.4 billion Agentic Work Units across Agentforce and Slack and defines AWUs as tasks accomplished by AI agents.
In other words, Salesforce already knows how to measure more than logins or tokens. The next pricing reset can move from activity performed towards result verified without starting from zero.
Agentforce is autonomous enough to justify an outcome meter
Agent pricing should move towards outcomes only when the product actually performs enough of the job. The Agentic Monetization Spectrum, or AMS, answers that question through three dimensions. Zero-human ability asks how much useful work can finish without human intervention. Operational domain measures whether the agent performs one narrow task or acts across a wider business process. Output/cost ratio asks how large the customer's useful output can become relative to the cost of delivering it. Products scoring low on these dimensions behave like traditional tools and tend to fit seats or access pricing. Products scoring high behave more like labour or services, making completed work and outcomes more defensible meters.
Agentforce sits well beyond the copilot end of that spectrum. Salesforce's September 2024 launch materials described agents that could reason, make decisions and complete tasks without human intervention; the fiscal 2026 10-K describes agents that handle workflows and take action through Salesforce and MuleSoft.
The AMS score makes the pricing implication explicit.
| AMS dimension | Score | Evidence and implication |
|---|---|---|
| Zero-human ability | 4/5 | Agentforce can complete autonomous workflows, but escalation and human supervision remain material in many deployments. Salesforce reported more than four million autonomous service transactions on its May 2026 earnings call. |
| Operational domain | 5/5 | Salesforce's FY2026 10-K describes Agentforce across sales, service, marketing, commerce, integration and analytics. |
| Output/cost ratio | 4/5 | Standard actions list at roughly $0.10 while agents can execute work previously performed inside human workflows; realised customer value remains use-case-specific. |
| Total | 13/15 | Outcome-leaning pricing is justified for workflows where Salesforce can verify completion. |
A 13/15 product should not be priced primarily like an API call. Our view is that Agentforce has crossed the threshold at which Salesforce can credibly say, “Pay us when the agreed work gets done.”
The remaining constraint is attribution, not autonomy.
The market is already segmenting outcomes by business impact
Salesforce would not be making this move in isolation. By August 2026, B2B AI vendors are running a live market test of how far pricing can move from access, through usage, and towards results.
The most interesting signal comes from vendors that distinguish different outcomes by their economic meaning, rather than attaching a single price to every automated interaction.
| Vendor | Evidence date | AI pricing unit | Relationship to business impact |
|---|---|---|---|
| Intercom Fin | 26 Jun 2026 | $0.99 for a resolution, procedure handoff or disqualification; $9.99 for a qualification | Strong - different verified outcomes carry different prices. |
| Zendesk AI Agents | Jul 2026 page, accessed 13 Aug 2026 | Successful resolutions, with outcome tiers based on value delivered | Strong - explicitly segments resolutions by value. |
| Salesforce Agentforce | 13 Aug 2026 | $2 conversation or Flex Credits at $500/100,000; standard action 20 credits | Medium - work is measurable, but the customer-facing meter usually stops before business impact. |
| Freshworks Freddy AI | 13 Aug 2026 | First 500 sessions included; another 100 sessions for $49 on current Freshdesk Omni plans | Lower - price follows AI sessions even though the product is marketed around end-to-end resolution. |
| HubSpot Breeze | 13 Aug 2026 | Credits at $0.010 each; 1,000-credit packs cost $10 | Medium - credits allow scaling, but consumption remains the main charging unit. |
The market evidence does not suggest that every vendor should copy Intercom. It shows something more useful: outcomes themselves can be segmented.
Intercom's June 2026 pricing is particularly instructive. Fin charges $0.99 for a support resolution but $9.99 for a qualified prospect routed to an agreed next step. Failed attempts are not charged as outcomes.
That is a tenfold difference between two pieces of completed agent work. The customer is not paying ten times more because qualification needs exactly ten times the tokens or model calls. Intercom is distinguishing what the completed job is worth.
Practitioners elsewhere are feeling their way towards the same question. On HubSpot's 8 May 2025 earnings call, CEO Yamini Rangan described its philosophy as delivering value first and monetising after the company sees repeat value; HubSpot subsequently expanded credits for Breeze Customer Agent.
Freshworks CEO Dennis Woodside gave an equally revealing counterpoint on the company's Q3 2025 earnings call. Freshworks had seen some customers reach up to 80% deflection with newer agents, yet he said the company was “not quite to the point to move to full resolution-based pricing.” Freshworks retained session pricing while acknowledging that more agentic interactions were more valuable.
Taken together, Benioff, Millham, Rangan and Woodside describe the same transition from different angles. SaaS operators know that the economic unit is changing. What separates the better models is not enthusiasm for consumption or outcomes, but whether the vendor can prove which result happened and price that result differently from a lower-value one.
Salesforce's next reset should make verified outcomes the primary meter
Salesforce gets three important things right.
First, it refused to preserve seat pricing simply because seats were familiar. Second, it separated employee-facing agents from customer-facing agents, recognising that a worker using an assistant is economically different from an autonomous agent serving thousands of customers. Third, Salesforce built the operational machinery required for variable pricing. Digital Wallet provides near-real-time monitoring and alerts; PayGo, pre-commit and pre-purchase separate payment terms from the underlying meter.
What Salesforce gets wrong is subtler. It says Flex Credits align cost to business value, but its own definition bills an action such as a record update or prompt execution. An action can be necessary for value without being the value itself.
A support agent that tries five actions and fails to solve the customer's problem may consume more Flex Credits than an agent that solves it in two. The existing meter can therefore make a less efficient agent generate more revenue from the same attempted job. That incentive becomes harder to defend as autonomy rises.
Monetizely's recommended reset is one architecture: verified business outcome becomes the primary customer-facing Agentforce meter, while Flex Credits remain the underlying cost and usage ledger rather than the headline value unit.
For service, the charging event should be a resolution that passes agreed verification rules. For sales, it should be an accepted qualification or another defined progression event. For transactional service and commerce, the unit should be a completed booking, change, recovery or purchase. Prices should differ because those outcomes produce different business impact.
Employee-facing Agentforce remains appropriately packaged around user access where attribution is weaker. That does not dilute the recommendation. It is exactly what segmentation is meant to accomplish: price autonomous customer work by completed business result; price human-assistance products around the human who receives the assistance.
Salesforce's own corporate metrics point in the right direction. Its April 2026 proxy does not call Agentic Work Units “tokens consumed”; it calls them tasks accomplished by an AI agent. The company had delivered 2.4 billion of them by the end of fiscal 2026 and put AWUs into executive performance metrics for fiscal 2027.
The next conceptual step is small but economically important: a task accomplished tells Salesforce that the agent worked. A verified outcome tells the customer that the work mattered.
Leaders should design the price around the business result
The lesson extends beyond Salesforce. Outcome pricing is not a licence to pick the biggest customer KPI on a slide and charge against it. Revenue, churn and customer satisfaction usually contain too many variables outside the software vendor's control.
Pricing leaders should instead choose the nearest result that is both valuable and provably caused by the product. Intercom can verify whether Fin qualified and routed a prospect. Zendesk can identify an AI resolution. Salesforce can observe agent actions, workflows, completed tasks and customer handoffs.
For operators making the shift, Monetizely's position leads to four higher-level decisions:
Set a target for how much autonomous-agent revenue should come from verified results. Treat the mix of outcome-priced ARR as a strategic KPI rather than allowing credit consumption to become the end goal.
Build the product roadmap around outcomes worth buying. A pricing team cannot create a high-value qualification, successful recovery or completed transaction if the agent itself stops at drafting text.
Make unit economics visible by outcome class. A support resolution and a qualified enterprise lead should each have their own revenue, delivery cost and gross-margin view rather than disappearing inside one pool of AI credits.
Report customer impact alongside agent activity. Billions of tasks and rising consumption prove adoption. Resolutions, accepted opportunities, completed transactions and avoided manual work prove monetisable value.
Salesforce has already done much of the difficult work. It has moved beyond seats, learned to meter digital labour, built controls for consumption, created separate packages for humans and agents, and accumulated enough operating data to measure autonomous work at enormous scale. Its fiscal 2026 10-K reported $41.5 billion of total revenue, while the 2026 proxy reported $800 million of Agentforce ARR and more than 29,000 Agentforce deals in the first 15 months.
The pricing opportunity now sits one level higher. Do not ask customers to buy more AI activity as their agents become more capable. Ask them to buy more successful work. Segment that work by business impact, verify it rigorously, and let price rise when customer value rises.
That is how outcome pricing stops being a slogan and becomes a durable revenue model.
Assumptions
AMS scores and framework grades are Monetizely assessments based on publicly available product behaviour, pricing and operating disclosures as of 13 August 2026. Dollar-to-action conversions use Salesforce's current public rate of $500 per 100,000 Flex Credits and 20 credits for a standard action. Historical pricing points are corroborated through dated Salesforce disclosures; Wayback links below are month-specific archive indexes because individual archive timestamps can vary by geography and page capture.
Footnotes
https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
https://investor.salesforce.com/news/news-details/2024/Salesforce-Unveils-AgentforceWhat-AI-Was-Meant-to-Be/default.aspx
https://web.archive.org/web/202410*/https://www.salesforce.com/agentforce/pricing/
https://www.salesforce.com/in/agentforce/pricing/
https://web.archive.org/web/202505*/https://www.salesforce.com/agentforce/pricing/
https://www.salesforce.com/in/news/stories/pricing-update-2025/
https://www.salesforce.com/news/stories/new-agentforce-payment-options/
https://www.sec.gov/Archives/edgar/data/1108524/000110852426000060/crm-20260131.htm
https://www.sec.gov/Archives/edgar/data/1108524/000110852426000085/crm-20260416.htm
https://www.fool.com/earnings/call-transcripts/2025/02/26/salesforce-crm-q4-2025-earnings-call-transcript/
https://www.fool.com/earnings/call-transcripts/2026/05/28/salesforce-crm-q1-2027-earnings-transcript/
https://www.marketbeat.com/earnings/reports/2025-5-8-hubspot-inc-stock/
https://www.investing.com/news/transcripts/earnings-call-transcript-freshworks-beats-q3-2025-estimates-stock-gains-93CH-4335884
https://www.intercom.com/help/en/articles/8205718-fin-ai-agent-outcomes
https://www.zendesk.com/service/ai/ai-agents/
https://www.freshworks.com/freshdesk/omni/pricing/
https://www.hubspot.com/products/artificial-intelligence/credits
https://ir.hubspot.com/news-releases/news-release-details/hubspot-credits
https://doi.org/10.1057/s41272-025-00530-z
https://link.springer.com/article/10.1007/s11846-022-00521-z