The Pricing Model Actually Changed
Salesforce quietly did something significant with the Agentforce Help Agent launch on June 25: they moved from per-conversation billing — where you paid regardless of outcome — to pay-per-resolution at a flat $2, charged only when the agent closes an issue autonomously. No resolution, no charge. Escalations are free.
That is not a marketing reframe. It is a structural shift in who absorbs execution risk. Under the old model, a 30% autonomous resolution rate still costs you 100% of the conversation volume. Under outcome-based pricing, that same 30% rate costs you 30% of what you'd expect. The vendor now has direct financial skin in the resolution game.
What the Data Actually Says
Salesforce isn't running this blind. On help.salesforce.com, the agent handled 4.3 million inquiries with a 70% autonomous resolution rate. That's the internal benchmark they're pricing against. If your support profile is messier — more edge cases, more custom configurations, more escalation-prone tickets — your effective cost and resolution rate will diverge from that benchmark. Model it explicitly before July GA.
The agent ships pre-wired to Salesforce Knowledge, which matters for setup speed but also means resolution quality is directly tied to how clean and current your Knowledge base is. Garbage in, escalations out — and at least those escalations are now free.
How to Evaluate This Before You Renew
If you're currently on Flex Credits or per-conversation pricing, run three numbers before July:
- Your actual monthly conversation volume from the last 90 days
- Your estimated autonomous resolution rate — be conservative; use 40-50% if you don't have clean data
- Your blended cost per conversation under the current plan
Compare that against $2 × (volume × estimated resolution rate). The delta tells you whether outcome-based pricing wins immediately or only after you've invested in Knowledge base cleanup.
More importantly, this pricing structure gives procurement and finance a clean story: you're buying verified outcomes, not activity. That changes the internal approval dynamic. Budget owners who hesitated on AI support spend because of unpredictable cost-per-conversation now have a defensible unit economics model to approve against.
The practical action for any operator running on Salesforce: pull your support ticket data this week, build the comparison model, and enter the July GA conversation with numbers rather than vendor-supplied benchmarks.