Outcome-Based Pricing
for AI Agents
When a major CX platform launched its AI agent capabilities, the biggest risk wasn't competition - it was their own pricing model. If AI agents worked as promised, customers would need fewer seats. As the company's lead pricing strategist, I designed a monetization model that turned that risk into a new and durable revenue stream, without handing competitors a price war.
The Challenge
One of the largest customer experience platforms in the world was on the verge of launching AI agents - software capable of autonomously resolving customer support tickets from start to finish. The product team was delivering something genuinely transformative. But the pricing team saw a threat hiding inside the opportunity: if customers succeeded with AI agents, they'd need fewer human agents, which meant fewer seats, which meant shrinking core revenue.
The existing seat-based model couldn't capture this new value. Charging for access to AI agents wouldn't scale with actual usage - a small team with low ticket volume would pay the same as an enterprise handling millions of interactions. And with underlying inference costs tied directly to consumption, pricing had to scale with usage or margins would collapse. The status quo wasn't an option.
The Goal
The task was to design a pricing and packaging strategy for AI agents that accomplished three things simultaneously: offset the seat-based revenue risk as adoption grew; protect margins in a cost environment dominated by expensive frontier LLMs; and establish the company as the pricing leader in an AI-first CX market - not just the first to launch, but the hardest to replicate. We had roughly ten months from initial strategy work to monetization launch.
My Approach
The first question was whether to price on usage at all, and if so, what to count. Tokens and credits were the obvious choice — they were what the underlying models actually charged for - but I ruled them out quickly. The market had no intuition for what a token was worth in a customer service context, which would make it impossible to defend a price point. And if every vendor was reselling the same frontier model capacity with a margin applied on top, differentiation would collapse into a race on price. We needed a metric that was customer-native, not infrastructure-native.
I worked with the product team to map the full spectrum of what the AI agents actually did: retrieval, reasoning, agentic workflows, partial automations, full resolutions. The most defensible and value-aligned unit wasn't a ticket touched or a message sent - it was a ticket fully resolved by AI. That became the pricing metric: Automated Resolutions.
But the definition of "resolved" mattered enormously. Previous implementations in the market had used simple customer confirmation as the threshold. I pushed for something harder to replicate: a dual-layer definition combining a deterministic resolution check with an AI-governed quality assessment - meaning we'd only count a resolution if our AI could verify the customer was actually helped. This made the metric inherently value-aligned and effectively impossible for competitors to replicate without matching the underlying quality bar.
With the metric settled, I built the financial model. Using the observed distribution of inference costs across resolutions of varying complexity, I established a cost floor. Then I modeled the implied willingness-to-pay by working backward from what customers were already spending on seat licenses to resolve equivalent ticket volumes - constructing a demand curve from real behavioral data rather than conjoint alone. I also partnered with research consultants on a market survey validating the concept with both existing customers and prospects across our target segments.
Finally, I designed the free tier: a volume of Automated Resolutions included in every plan, scaled by plan tier and seat count. The goal was to set the allowance high enough that most customers would get real value before hitting a paywall - prioritizing adoption over near-term revenue, with the thesis that adoption at scale would drive durable expansion revenue over time.
The Outcome
The launch landed as an industry-first. Competitors moved quickly to imitate the concept, but couldn't match the definition — and because the pricing metric was genuinely anchored in quality outcomes, the company was able to defend a meaningfully higher price point and avoid the commoditization trap. Margins on AI agent capabilities significantly outperformed what the industry typically achieves on AI usage revenue — which consistently benchmarks in the 40-60% range, well below traditional SaaS norms of 70-85%. The free tier strategy paid off: broad adoption drove expansion revenue at scale. The company closed 2025 with $200M in AI annual recurring revenue and is projecting $400M+ in 2026 — built almost entirely on a pricing model designed in that ten-month window.