Pay-Per-Resolution: Salesforce Just Made Outcome Pricing Mainstream. Now What?

In April, we argued that outcome-based pricing for AI sales tools was coming, and that the vendors would race there because per-conversation billing and credit systems were burning buyer trust. We expected the shift to take a year or two. It took ten weeks.

In June 2026, Salesforce introduced pay-per-resolution pricing for Agentforce. The new Agentforce Help Agent costs a flat $2 per resolved issue, with Data 360 and Agentforce usage unmetered during the interaction. General availability lands this month. And yesterday, July 1, Salesforce closed its acquisition of m3ter, a metering and rating platform it signed on June 8, which now powers consumption and outcome-based billing inside Agentforce Revenue Management.

Outcome pricing is no longer a startup experiment or an analyst slide. The largest CRM vendor on the planet just made it the headline model for its flagship AI product. That deserves a fair reading first, and a skeptical one right after.

What Salesforce actually changed

Agentforce has now been through three pricing models in under two years. It launched at $2 per conversation, which meant you paid the same $2 whether the agent solved the customer's problem or frustrated them into calling a human. Then came Flex Credits, a consumption system where every agent action drew down a prepaid balance, which we picked apart in our look at the Agentforce ARR fine print. Now the model is $2 per resolved issue.

Two details matter. First, unresolved conversations cost nothing. If the agent fails and escalates to a human, Salesforce doesn't get paid. Second, Data 360 queries and Agentforce usage inside the interaction are unmetered. Under Flex Credits, an agent that checked order history, looked up a warranty, and drafted a reply burned credits at every step. Under pay-per-resolution, all of that is bundled into the $2. The meter runs on the outcome, not the activity.

Credit where due: this is a genuinely better structure. Per-conversation pricing charged you for failure. Credit pricing charged you for effort, and effort is exactly the thing an inefficient agent produces in abundance. Paying for a resolved issue is the first Agentforce model where the vendor loses money when the product performs badly. That is real alignment, and Salesforce deserves recognition for shipping it at scale.

Buyers pushed the market here

Salesforce didn't do this out of generosity. Buyers have been voting against activity-based pricing for over a year, and the survey data is unambiguous.

The competitive picture reinforces it. HubSpot rolled out outcome-based pricing for its Breeze Customer and Prospecting Agents earlier in 2026. Once your two largest competitors in the mid-market are charging for outcomes, per-seat AI add-ons at $50 a head start looking like paying for potential instead of results. Salesforce had to move.

So the prediction came true, faster than we expected. Which means the interesting question is no longer whether outcome pricing arrives. It's whether outcome pricing actually works for buyers. And there, three problems remain unsolved.

Problem one: the vendor grades its own homework

Pay-per-resolution has a definitional core: what counts as resolved? In the Agentforce model, Salesforce's systems decide. If the agent marks a case resolved and the customer doesn't reopen it within the measurement window, that's $2. The customer who gave up and emailed support from a different address, or churned quietly, or got a technically-correct-but-useless answer and stopped replying? Also resolved. Also $2.

This isn't hypothetical cynicism. Every support leader knows that "resolution rate" is one of the most gameable metrics in the industry, which is why teams track reopen rates, CSAT, and first-contact resolution alongside it. An AI vendor paid per resolution has a direct financial incentive to classify ambiguous outcomes as resolutions. And when you dispute one, the evidence lives in the vendor's dashboard, computed by the vendor's classifier, presented on the vendor's terms.

Problem two: your bill now scales with your bad months

Flat pricing has one virtue no consumption model can match: the CFO knows the number in January. Pay-per-resolution replaces that number with a curve that tracks ticket volume, and ticket volume is not something finance controls.

Think about what drives support volume up. A buggy release. A confusing pricing change. An outage. A migration that breaks integrations. Every one of those is a bad month for the business, and under pay-per-resolution, every one of those is now also an expensive AI month. The billing model converts your worst product weeks into your vendor's best revenue weeks. At $2 per resolution, a company handling 50,000 AI resolutions a month is committing to $100,000 a month that rises precisely when things go wrong.

The standard vendor answer is volume commitments with tiered discounts, and that helps at the margins. But a commitment is a floor, not a ceiling. Commit low and your overage rate punishes every busy month. Commit high and you're back to paying for capacity you may never use, which is the exact prepaid-credits problem this model was supposed to fix. There is no committed-volume structure that turns an inherently variable cost into a fixed one. Finance ends up budgeting a range, then defending the top of it.

This is the same forecasting problem we flagged in our piece on usage-based AI pricing, moved one layer up the stack. Metering outcomes instead of tokens is fairer per unit, but it's no more predictable in aggregate. Any CFO running the evaluation framework from our guide on evaluating sales AI should model the P90 volume scenario, not the average, before signing.

Problem three: sales outcomes don't count like support outcomes

Here's the part that matters most for revenue teams. Pay-per-resolution works for support because a support resolution is discrete, high-volume, and mostly attributable to the agent that handled it. One ticket, one interaction, one outcome, countable in the thousands per month.

Sales outcomes are none of those things. A closed deal is the product of a six-month, multi-touch process: an SDR sequence, three discovery calls, a champion who sold internally, a proposal, a security review, a negotiation. Humans and AI share every stage. If a vendor charges per closed deal, what exactly did its software cause? The honest answer is "some unknowable fraction," and the pricing answer will be "all of it."

Follow that to its conclusion. Outcome pricing on sales results is a commission paid to your software vendor. A percentage of closed-won, or a flat fee per deal, paid to a tool that touched the deal alongside five other tools and four humans. No sales leader would pay a rep full commission for forwarding one email in a deal thread. That's the deal outcome-priced sales AI is asking for.

The m3ter acquisition tells you the roadmap

Read the July 1 close of the m3ter deal in this light. m3ter is not a support tool or an agent framework. It's a metering and rating engine, infrastructure for measuring granular events and turning them into invoices. Salesforce didn't buy it to power one $2 SKU. It bought it, and wired it into Agentforce Revenue Management, because the plan is to meter more things: more agents, more outcome types, more billable events across the platform.

That's not sinister, it's strategy, and Salesforce has been transparent about wanting consumption and outcome revenue to grow. But buyers should be clear-eyed about the direction. The vendor that owns the meter decides what gets metered next. Every capability that today sits inside your license is a candidate to become an outcome SKU tomorrow. The pricing model you sign in 2026 is the thin end of the model you'll be renewing in 2028.

What buyers should do now

The practical takeaway is not to reject outcome pricing. Where it fits, it's the fairest model on offer. The takeaway is to be deliberate about where it fits.

• Use outcome pricing for discrete, high-volume tasks. Support resolutions, meeting bookings, record enrichment. Clear definitions, countable events, attribution that mostly holds. Cap the monthly spend so a volume spike can't blow up the budget.
• Keep the core system of work on flat pricing. The CRM, the pipeline, the AI your reps use fifty times a day. This is infrastructure, and infrastructure billed per event punishes adoption. You want reps using AI more, not a meter that makes every use a line item.
• Demand the resolution definition in the contract. Not in the docs, not in the dashboard. The contract. What counts as resolved, the reopen window, your audit rights over the underlying interaction data, and the dispute process. A vendor confident in its resolution quality will put the definition in writing. A vendor that won't is telling you something.

How PipeLance approaches this

PipeLance charges flat per-seat pricing: Core at $69/user/month and Pro at $149/user/month. All 33 capabilities and all 119 AI tools are included in the license. There are no meters, no credits, no per-resolution charges, and no outcome SKUs. A rep who runs 400 AI actions in a day pays the same as one who runs 4.

This is a deliberate incentive choice, not a pricing shortcut. When a vendor bills per resolution, its revenue grows with billable events, so its incentive is to maximize the count. When a vendor bills flat and eats its own inference costs, its incentive is the opposite: make the AI efficient, resolve the issue in one pass, chain ten steps into one action, and stop. Our margin improves when the AI does more with less. Under a meter, the vendor's margin improves when you consume more. Buyers should ask which of those incentives they want pointed at their business.

Zoom out and the June announcement is still a win. The market spent two years charging for conversations and credits, buyers pushed back with their preferences and their renewals, and the biggest vendor in the category blinked. Pricing is moving toward value, which is where it always eventually goes. But "pay for outcomes" is a slogan, and the three hard questions underneath it, who defines the outcome, who absorbs the volatility, and which outcomes can honestly be attributed to software, are all still open. The vendors just bought a metering company. Buyers should bring a contract lawyer.