AI Workflows / For Sales and Service Leaders
Salesforce Agentforce: Pay Per Resolution AI for Sales
Salesforce now charges roughly two dollars only when its new Help Agent closes a customer issue by itself, tying the bill to outcomes instead of usage.
Salesforce Agentforce Help Agent uses pay-per-resolution pricing, reported at about $2 per successful resolution, meaning the customer is charged only when the autonomous agent resolves a customer issue from start to finish. If the issue is escalated to a human or the customer says they were not helped, there is no charge. Announced June 25, 2026, it is generally available July 2026 on the Agentforce 360 Platform.
The shift is the price tag, not the robot
Every enterprise software vendor now ships an AI agent. What Salesforce did on June 25 is more interesting than another agent launch. It changed how you pay. With the new Agentforce Help Agent, you are billed only when the agent resolves a customer issue on its own, from start to finish. If an issue is escalated to a human agent, Salesforce says it is not billed. Salesforce calls this pay-per-resolution, and as CIO reported, the price is about two dollars per successful resolution.
That pricing model warrants close attention from any leader managing a P&L. Most AI tooling bills you for consumption: tokens, seats, API calls, minutes. You pay whether or not the thing works. Outcome pricing flips the risk. The vendor only gets paid when the customer's problem is actually solved. That is a different conversation to have with your CFO, and it is why this launch matters more than the feature list.
What the Help Agent actually is
Strip away the framing and the Help Agent is a prepackaged, opinionated service agent built on the Agentforce 360 Platform. The pitch is speed to value. Instead of wiring up knowledge, actions, and channels yourself, you get a guided setup that Salesforce says deploys in minutes.
Three things make a service agent useful, and Salesforce claims to have solved all three out of the box:
- It grounds itself automatically on your Salesforce Knowledge, and you can drag in extra files or point it at a web URL to crawl. Messy data is the usual reason these agents fail, so removing that setup step is the point.
- It comes with prepackaged actions, not just answers. Out of the box it can answer questions and manage cases. You can add order management, appointment scheduling, and account management through Agentforce Builder or a coding agent of your choice.
- It works across voice, web, portal, and messaging, all switched on from a single screen. Salesforce says it will even provision a phone number for the voice channel.
There is also a redesigned Customer Service Portal built around a single conversation bar, closer to a search box that holds a real conversation than a traditional help center.
The credibility claim underneath all of this is Salesforce's own usage. According to the announcement, its help.salesforce.com portal has handled 4.3 million inquiries and resolved 70 percent of them autonomously. That is the experience the Help Agent is built on, which is a fair thing to point to.
How pay-per-resolution actually works
This is where a sales or service leader needs to read the fine print, because "you only pay when it works" is a headline, not a contract.
Salesforce's EVP and GM of Agentforce Service, Kishan Chetan, defined a resolution plainly to CIO: a successful resolution is when the user's question does not need a human escalation. If it gets kicked to a person, it does not count. If the customer explicitly says they did not get their answer, it does not count. In those cases Salesforce says the agent hands full context to your human team and you are not billed.
A few operational details from the CIO reporting that change your math:
- The price reported by CIO is about $2 per successful resolution.
- A resolution is scoped by a time window, not by question count. Actions inside a 10-minute window on a call, and a defined window for chat, count as a single resolution no matter how many questions the agent answers.
- You pre-purchase resolutions in packets, with a minimum of 1,000.
- During the interaction, both Data 360 and Agentforce are unmetered, so you are not separately tracking consumption or worrying about overages.
Read those together and the model is cleaner than usage billing but not free of judgment calls. The vendor defines what counts as resolved. Escalation to a human and an explicit "I did not get my answer" are the two outcomes Salesforce has publicly said are not billed. What it has not publicly detailed is how an ambiguous outcome is treated, for example a customer who never escalated and never complained but quietly gave up. That is not an accusation, it is the kind of edge you should confirm contractually before you sign.
The math you should run before you buy
Do not evaluate this on vibes. Run the comparison against what a resolution costs you today. Here is a decision framework you can use in one working session:
1. Pull your current fully loaded cost per resolved contact. Take agent salaries, benefits, tooling, and management overhead, then divide by resolved tickets. Most teams land somewhere well above two dollars for anything a human touches. 2. Estimate your automatable share. Look at your ticket mix and flag the repetitive, knowledge-answerable, low-emotion issues. Password resets, order status, appointment changes. That is the pool the agent can realistically close. 3. Model three scenarios at your real volume. Multiply automatable volume by an assumed resolution rate (be conservative, do not assume Salesforce's 70 percent) and by two dollars. Compare to today's cost for that same pool. 4. Price the escalation path. Every unresolved contact still costs you a human. Make sure your model counts the deflection failures, not just the wins, so you are comparing total cost, not best case. 5. Stress test the definition of resolved. Ask Salesforce in writing how a silent abandon is billed versus a human escalation, since that treatment is not publicly detailed. Your unit economics depend on that answer.
If you lead a team through AI adoption decisions like this one, our [AI course for managers](/ai-for-managers) walks through exactly how to build the cost model, set guardrails, and roll a tool like this out without breaking your CSAT. You can also take the [two-minute quiz](/quiz) to find the right starting point for your team.
Where this fits in the bigger agent stack
The Help Agent is not an isolated product. Salesforce also signed a definitive agreement on June 15 to acquire Fin, a customer agent platform aimed at small and medium businesses that Salesforce says is trusted by more than 30,000 companies. Salesforce expects that deal to close in the fourth quarter of its fiscal year 2027, subject to regulatory clearance. The direction is obvious: more prepackaged agents, faster time to value, and pricing that follows outcomes rather than seats.
For sales and service leaders, the strategic read is that autonomous agents are moving from custom builds to installable products, and the vendors are starting to compete on how they charge, not just what they ship. If you are building an operating system for how your team runs AI, this is the pattern to plan around. Our [AI for managers operating system](/ai-workflows/ai-for-managers-operating-system) breaks down how to fit tools like this into a repeatable stack instead of a pile of pilots.
The honest caveats
A few things to keep in front of you. This is a July 2026 general availability, and outcome pricing is new enough that the real per-resolution economics at scale are not yet public beyond the reported two-dollar figure and the 1,000-resolution minimum. The 70 percent resolution rate is Salesforce's own number on its own portal, which is a strong signal but not a guarantee for your ticket mix. And the definition of a resolution is set by the vendor, which is the single most important term to negotiate and monitor. Trust the model, verify the meter.
None of that makes the launch weak. It makes it worth a serious, numbers-first evaluation rather than a demo-driven yes.
Frequently Asked Questions
Does pay-per-resolution really mean I pay nothing when the agent fails?
That is what Salesforce states. If the issue is escalated to a human agent or the customer explicitly says they were not helped, Salesforce says it is not billed, and the agent passes full context to your team. The catch is that Salesforce defines what counts as a resolution, and it has not publicly detailed how a silent abandon or other ambiguous outcome is billed, so confirm that contractually before you commit.
What does a resolution actually cost?
CIO reported about $2 per successful resolution, purchased in packets with a minimum of 1,000. A resolution is scoped by a time window, so multiple questions answered inside a 10-minute call window count as one resolution, not several.
How fast can we really deploy it?
Salesforce positions it as deployable in minutes because it grounds automatically on your Salesforce Knowledge and ships with prepackaged actions and one-screen channel setup. Real-world timelines depend on how clean your knowledge base is and how many custom actions you add, so treat "minutes" as the setup wizard, not the full rollout.
Is this only useful for support, or does it help sales?
It is a customer service agent at its core, but for revenue leaders it matters because deflecting routine service load frees human time for higher-value conversations, and the portal can trigger workflows like scheduling and ordering inside the chat. Evaluate it as a service tool with sales-adjacent upside, not a seller.
What is the biggest risk for my team?
The definition of "resolved" is the vendor's, and your budget is tied to it. If the detection of unhappy or abandoned customers is loose, you could pay for interactions that did not truly help. Put the resolution definition and its measurement in writing.
When can we get it?
The Help Agent, the redesigned Customer Service Portal, and pay-per-resolution pricing are all expected to be generally available in July 2026.
Sponsored Training
Informational tool analysis for working professionals, not legal, medical, or financial advice. AI tools do not replace your professional judgment.