Pega vs. the "Pay-Per-Outcome" Crowd: What's Really Driving AI Cost Predictability?

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?