Outcome-based pricing sounds ideal. "Charge me per ticket resolved" or "bill me per lead generated" creates clear alignment between cost and value. But in practice, two problems kill it.
You can't measure outcomes reliably enough to invoice them
Usage-based pricing is straightforward. Tokens consumed, API calls made, images generated - these are discrete, trackable events. Outcomes require consistent definition, tracking infrastructure, and agreement on what counts.
What does "resolved" mean? Does a ticket that gets escalated to human support count as resolved by the agent? What about partial resolutions? Edge cases multiply fast, and without infrastructure to track them consistently, you can't invoice with confidence.
Most companies don't have this infrastructure. They're building agents first and figuring out measurement second, which means they can't defend the outcome metrics when it's time to charge for them.
You're exposed to dependencies you can't control
Even when you can measure outcomes, you're often measuring things influenced by factors outside your control. If your agent's performance depends on:
- Client data quality
- Their internal processes
- Seasonal demand fluctuations
- Market conditions
- External integrations
Then your revenue becomes volatile while your costs stay relatively fixed. August hits, everyone goes on holiday, ticket volume drops, and suddenly your outcome-based revenue crashes - but your infrastructure costs don't.
This is why companies offering outcome-based pricing keep traditional models as backup. They need a revenue floor when external factors suppress outcomes through no fault of the agent's capability.
The same problem breaks internal business cases
Here's where this gets relevant for product teams: the same measurement problem that breaks vendor pricing breaks your internal business case.
Most companies try to justify AI agent investments upfront by proving both efficiency gains AND new capabilities before building anything. The business case promises 30% faster resolution times, 20% cost reduction, AND new predictive insights that enable proactive customer service.
That last part— - the new capabilities— - is where business cases fall apart. You can't measure the value of capabilities that don't exist yet. You can't baseline insights you haven't discovered. You're asking finance to approve investment based on outcomes you can't prove.
It's the same problem vendors face, just internal.
How to actually build the business case
The companies getting this right aren't trying to prove everything upfront. They're structuring investment in phases, where each phase funds the next based on proven value.
Step 1: Establish baseline metrics before you build anything
You can't prove ROI without knowing what "normal" looks like. Before writing a single line of code, instrument your current performance:
- Average ticket resolution time
- Escalation rates
- Customer satisfaction scores
- Volume patterns over time
- Cost per interaction
The first job isn't the agent. It's building the measurement infrastructure that lets you isolate what the agent actually improved versus what would have happened anyway.
Step 2: Start where you can measure: Efficiency gains
Phase 1 focuses on measurable efficiency improvements. Faster ticket resolution, reduced manual effort, lower operational costs. These are easy to quantify because you have baseline data and the outcomes are discrete.
This isn't exciting work. It's not a transformative AI use case. But it's what creates the financial headroom to fund discovery of what's actually valuable.
Your agent resolves tickets 30% faster - that's measurable ROI that justifies continued investment. Finance can see the cost reduction. Leadership can see operational improvement. You've proven the agent works reliably on defined tasks.
Step 3: Fund the observation period
Now use that efficiency-driven ROI to fund Phase 2: having someone watch what the agent reveals.
This is the critical piece most companies skip. They deploy the agent, see the efficiency gains, call it success, and move on. They never discover the new capabilities because no one is watching for them.
What you're looking for:
- Patterns in customer behavior that were buried in volume
- Correlations between support issues and other business metrics (churn, feature adoption, upsell opportunities)
- Questions that cluster in specific regions or customer segments
- Timing patterns that predict downstream problems
Example: You built an agent to resolve customer tickets faster. Once running, someone notices tickets about a specific feature spike two weeks before customers churn. That pattern was always there, but buried in volume. Now it's visible.
That's when new capabilities become possible: "Can we predict churn based on support patterns?" "Can we proactively reach out when we see the warning signs?" "Can we route feature feedback to product teams automatically?".
But you don't build those capabilities immediately. First, you validate the insight actually matters.
Step 4: Validate before you scale
Some insights will matter. Most won't. The observation period is about testing hypotheses, not building features.
Does that churn correlation hold across segments? Is it predictive or just correlated? If you acted on it, would it change outcomes? Can you measure the impact?
Only after validation do you build the capability into something measurable and scalable. Only then can you build the Phase 3 business case: "We've proven that proactive outreach based on support patterns reduces churn by X%. Here's the investment needed to scale that across all customer segments."
Now you're pricing outcomes you can actually measure because you've discovered and validated what they are.
Why this matters more than you think
Most companies are stuck at Phase 1. They've deployed agents that handle defined tasks efficiently, but they haven't unlocked new capabilities. Not because the capabilities don't exist, but because no one funded the observation period to discover them.
The business case promised transformative impact. The deployment delivered operational efficiency. The gap between aspiration and reality isn't technical - it's that companies tried to predict value they couldn't know yet, then didn't fund the discovery work to find it.
This is why AI maturity frameworks talk about stages:
- Stage 1-2: Cost savings and efficiency (easy to measure, easy to price)
- Stage 3-4: Enhanced decision-making (harder to isolate, needs baseline metrics)
- Stage 5: New business models enabled by AI (can't be priced until you know what's possible)
Most companies get stuck at Stage 2 not because they lack ambition, but because they built their business case like a Stage 5 transformation but only funded Stage 1 execution.
What to do differently
If you're building or buying AI agents, three things matter:
1. Baseline before you build. Being able to track performance effectively is essential. It’s the only way to prove your agent worked versus claiming credit for improvements that would have happened anyway.
2. Don't try to justify the whole investment upfront. Build a Phase 1 business case around measurable efficiency gains. Use those gains to fund Phase 2 discovery. Build the Phase 3 business case once you've validated what actually matters.
3. Budget for observation. Someone needs to watch what the agent reveals. That's not a free activity. It requires time, focus, and analytical capability. If you don't fund it, you won't discover the new capabilities that justify continued investment.
The bottom line
Outcome-based pricing for AI agents barely exists - not because companies lack ambition, but because they face a measurement problem. And it's the same problem breaking internal business cases.
You can't measure the value of capabilities that don't exist yet. You can't baseline insights you haven't discovered. You can't prove ROI on outcomes you can't track reliably.
But you can prove efficiency gains. You can use those gains to fund discovery. And you can build the business case for scaling what you've learned actually matters.
Stop trying to predict the future. Start funding the discovery period that reveals what's actually valuable.
