Salesforce made pay-per-resolution the new default for AI service agents in July 2026. Agentforce Help Agent went GA on a model where customers pay only when the agent autonomously resolves an issue end to end, with no charge when feedback is negative or the case escalates to a human. Data 360 and Agentforce run unmetered during the interaction, so there’s no consumption forecasting and no overages to manage.
The pricing is straightforward for buyers to accept, and whether each resolution actually earns a margin is a separate question that plays out entirely on the cost side.
Every unsuccessful resolution now sits on the vendor's ledger. That’s the intended design of the model, and it’s also why outcomes-based pricing makes cost attribution more important than ever.
Outcome pricing is moving fast
ICONIQ's January 2026 State of AI snapshot put outcome pricing at 18% of AI companies, up from 2% six months earlier. Salesforce turning it into a default for service agents will accelerate that curve. Seat-based and token-metered models are getting harder to defend in front of buyers who want to pay for outcomes.
The shift changes what needs to be measured. Seat and token models rewarded adoption and throughput, which are easy to count. Outcomes-based pricing rewards successful work and absorbs the cost of every attempt that doesn’t finish, which is much harder to count. That’s a different economic model, and it demands a different instrumentation layer underneath it.
What a resolved case actually costs
Knowing what a company charges for a resolved case says little about whether that case cleared margin. A single resolved ticket can run up model calls, retrieval steps, and repeated tool invocations when the agent retries. Two customers on the same SKU can produce very different cost curves depending on how their users prompt and how much human escalation gets baked into the path that ends in a resolution.
Pricing gets locked in at the contract table. Cost accumulates one transaction at a time, buried inside logs that rarely get aggregated by customer or by workflow. By the time finance sees the quarterly margin, the pricing decision is already six months old.
Product leaders usually feel this before anyone else does, when the unit economics stop lining up with the pitch. Usage patterns that looked harmless in a pilot start eating margin at scale, the gross margin line drifts, and the team struggles to identify which customers or workflows are responsible.
What has to sit under outcome pricing
For pay-per-resolution to hold up as a durable model, a few things need to be true at the transaction level.
- Cost per resolution is measured with the same precision as price per resolution, broken out by customer and by workflow.
- Margin traces back to the specific feature and customer segment that produced it, instead of being averaged across the whole book.
- Pricing decisions get revisited on real evidence as usage scales, so early pilot economics aren’t assumed to hold forever.
This is the layer Revenium provides, giving teams transaction-level cost attribution by customer and by workflow, tied to the outcome being priced. It reports economic intelligence in the terms product leaders actually run their business on, such as ROI by workflow and margin by customer.
Without that layer, outcomes-based pricing amounts to a bet that average behavior stays close to pilot behavior, and it rarely stays that close as volume grows.
The board question
Outcomes-based pricing gives boards a new question to ask. The focus moves away from ARR per seat or aggregate token spend toward whether each priced outcome cleared margin, and which cohort of customers keeps producing the ones that didn’t.
Answering that question requires cost data at the granularity of the outcome itself. When the finance team can only see aggregate spend by model provider, the board question gets a hedge rather than a real answer.
Salesforce set the benchmark for AI service pricing
With this move, Salesforce set an expectation that AI service pricing will be judged by whether the vendor absorbs the cost of failure. Every AI product team pricing against that benchmark now needs to know, in real time, what a resolution costs before deciding what to charge for it.
For most teams, the first obstacle is measurement, and pricing can only be set correctly once that measurement exists.
Revenium connects the two, mapping cost per resolution to margin per customer so that outcomes-based pricing becomes a model teams can monetize with confidence instead of a margin gamble.
Measure, prove and monetize every AI interaction.
Know What Every Resolution Costs Before You Price It
If you’re pricing against Salesforce's pay-per-resolution benchmark, you’re agreeing to eat the cost of every attempt that doesn’t finish. Making that a durable model means seeing cost per resolution with the same precision you already see price per resolution.
Revenium is that layer. We attribute every model call, tool invocation, and human escalation back to the customer and workflow that triggered it, then map that cost to the outcome you’re charging for. You see cost per resolution and margin per customer in real time, well before the quarterly margin review.
That’s how the board question finally gets a real answer, showing which cohorts produce the outcomes that clear margin and which ones erode it, backed by evidence rather than a hedge.
See how Revenium maps cost-per-resolution to margin-per-customer.
Book a demo at revenium.ai/demo, or start free at app.revenium.ai/sign-up.
