As agentic AI goes mainstream, every vendor is racing to convince clients their pricing model is the “fair” one. ServiceNow and Salesforce are both pitching outcome-based pricing as the antidote to unpredictable token bills — but a closer look shows their models still carry the same volatility they claim to solve. Pega takes a fundamentally different path with Case-based pricing. Here’s the breakdown.
1. Raw Token-Based Pricing (the baseline everyone is fleeing)
Traditional LLM consumption pricing charges per token processed or generated. As agentic workflows grow more complex, agents “re-reason” at every step, and token burn scales exponentially with workflow complexity. The result: unpredictable TCO and a nightmare for ROI forecasting — exactly the pain point competitors are now trying to market their way out of.
2. Salesforce Agentforce — “Pay-Per-Conversation”
Salesforce prices Agentforce at roughly $2 per conversation, where a “conversation” is a user-initiated interaction the agent works to resolve. It’s marketed as pay-per-resolution, shifting risk to the vendor. The catch: this fee sits on top of required base licenses for Sales/Service Cloud, Data Cloud, MuleSoft, and Flow — creating a fragmented, multi-line contract where the “outcome” price is really just one more meter among several.
3. ServiceNow — “Assist Tokens” and Consumption Tiers
ServiceNow’s Now Assist uses “Assist Tokens” as a currency consumed per AI action or assist within a workflow. While ServiceNow talks about paying per transaction or business outcome, tokens are still bundled into higher-tier licenses (Pro Plus/Enterprise Plus) as a variable layer on top of the base subscription. Gartner has flagged that these multi-tiered modules plus consumption metering make long-term cost forecasting difficult as automation scales — the “outcome” framing doesn’t remove the underlying token math.
4. Pega — Case-Based Pricing (no token cost)
Pega clients pay for the business outcome unit itself — the Case (a claim processed, a loan originated, a dispute resolved) — not for API calls, tokens, or conversations. For Pega-managed models, there are no additional token charges: whether a Case Agent takes 5 steps or 50 to resolve a case, the price doesn’t move. The Agentic AI Package applies a single, fixed percentage uplift to the existing case price, bundling Case Agents, GenAI Coach, and Knowledge Buddy — no hidden meters, no stacked licenses.
Summary Thoughts: What’s Best for the Client
-
Token-based (raw LLM):
- Every token processed
- Scales with complexity
-
Salesforce Agentforce
- Per conversation, plus base license stack
- Fragmented across products
-
ServiceNow Now Assist
- Per Assist Token, layered on license tier
- Consumption still hidden inside tiers
-
Pega Case
- Based-Per business outcome (the Case)
- Flat, regardless of AI complexity
The headline claim “only pay when the outcome is completed” sounds compelling, but in practice both ServiceNow and Salesforce still require a base subscription stack and a variable metering layer — the token math just moves one level up the stack. Pega’s Case-based model is the only one of the three where cost is genuinely decoupled from AI complexity and token volume, giving CFOs a single, predictable line item tied directly to business value delivered. For clients scaling agentic automation across high case volumes, that flat-cost-per-outcome structure is the more board-ready story.
Thoughts?