The $0.99 Resolution: What 18 Months of Outcome-Based AI Pricing Reveals About Billing Infrastructure
Outcome-based pricing sounds like the perfect alignment of incentives. Pay only when AI delivers results. But real-world data from Intercom, Zendesk, and others reveals billing complexity that creates new challenges for finance teams.
Last Updated: January 2026
Quick Reference
Key findings from 18 months of outcome-based AI pricing:
- Better AI = Higher costs: Improving your AI's effectiveness directly increases your bill
- "Resolution" definitions vary: Soft resolutions (customer silence) count as billable outcomes
- Forecasting is fundamentally different: Costs depend on customer behavior, AI effectiveness, and seasonality
- Finance teams need new controls: Budget ranges, rolling forecasts, and variance reserves replace fixed budgets
- Independent tracking is essential: Your resolution counts won't match provider invoices without reconciliation
The Promise vs. The Reality
Outcome-based pricing for AI seemed like the perfect solution to the unpredictability problem.
Instead of paying for tokens (confusing), seats (doesn't scale with value), or API calls (disconnected from outcomes), you'd pay only when the AI actually delivered results. Customer issue resolved? That'll be $0.99. Sales conversation completed? $2.00. Problem solved, value delivered, payment made.
Intercom pioneered this model when Fin launched in 2023 with per-resolution pricing. Zendesk followed with outcome-based pricing for their AI agents. Salesforce introduced Agentforce at $2 per conversation. The industry seemed to be converging on a better way.
After 18 months of real-world deployment, the data tells a different story. Companies report billing increases of 100%+ within a year. Finance teams describe budgeting as "impossible." One business owner reported going from $200/month to $1,400/month during a product launch. A Reddit post titled "My Intercom billing shot up by 120%" captured the sentiment that's become widespread.
Outcome-based pricing didn't eliminate billing complexity. It transformed it into something finance teams have never had to manage before.
The Core Problem: Better AI = Higher Costs
Here's what outcome-based pricing data reveals:
When your AI gets better, your costs go up.
Traditional cost structures work the opposite way. Better software is usually more efficient, reducing costs. Better employees produce more value per dollar. Better tools decrease time-to-completion.
But with per-resolution pricing:
- Improve your knowledge base → AI resolves more issues → costs increase
- Train the AI on better responses → resolution rate climbs → costs increase
- Reduce friction in the chat widget → more users engage → more resolutions → costs increase
A support team shared their experience: they invested time improving their help center content, which made Fin more effective at resolving issues. Their resolution rate jumped from 40% to 65%. Great for customer experience. Terrible for their budget—costs increased by over 60% with no corresponding increase in support volume.
This creates a counterintuitive incentive structure. Do you deliberately keep your AI less effective to control costs? Do you gate improvements behind budget approvals? The alignment becomes more complex than the model suggests.
The Definition Problem: What Counts as a "Resolution"?
The word "resolution" seems straightforward. The AI resolved the customer's issue. But dig into the definitions, and you find a range of edge cases.
Intercom's Definition
Intercom counts a resolution when:
- Hard Resolution: The customer clicks "That helped 👍" or replies with an affirmative response
- Soft Resolution: The customer exits the conversation without requesting further assistance within 24 hours
The soft resolution is where things get complex. If a customer asks a question, gets an answer they don't find helpful, and leaves without finding their answer? That's still counted as a resolution. You pay $0.99 for an interaction that didn't actually resolve anything.
One Intercom community post highlighted an even more problematic edge case: Fin counts a resolution even when a human takes over from the AI chatbot, as long as the original AI response happened first. The company initially promised to address this, then walked back the commitment.
Zendesk's Approach
Zendesk charges for an "Automated Resolution" when:
- The conversation was not handed over to a human agent, AND
- Zendesk's AI evaluation determines the request was "satisfactorily resolved"
Notice the second criterion: Zendesk uses AI to evaluate whether the AI successfully resolved the issue. This creates a black-box billing situation where you're paying based on an algorithm's judgment of another algorithm's performance.
Resolution Definitions Compared
| Provider | Price Point | Resolution Definition | Key Risk |
|---|---|---|---|
| Intercom (Fin) | $0.99/resolution | Hard (customer confirms) OR Soft (no follow-up in 24h) | Pays for abandoned conversations |
| Zendesk | Varies by tier | No human handoff AND AI determines "satisfactorily resolved" | Black-box AI evaluation |
| Salesforce (Agentforce) | $2/conversation | Completed agent conversation | Charges per conversation, not resolution |
The Verification Gap
This is the billing infrastructure challenge: How do you independently verify resolution counts?
With token-based billing, you can count tokens yourself (roughly). With seat-based billing, you know how many seats you have. But with resolution-based billing, you're trusting the provider's definition and counting methodology.
Companies report discrepancies between their internal metrics and provider invoices. Your system shows 1,847 AI-handled conversations. The invoice shows 2,103 resolutions. Which is right? How do you reconcile?
Need independent resolution tracking? Bear Billing provides conversation-level tracking that reconciles against provider invoices automatically. See how it works.
The Forecasting Challenge
Traditional SaaS costs are predictable. You have 50 seats at $100/month = $5,000/month. Done.
Outcome-based AI costs are anything but predictable. They depend on:
- Customer behavior: How many customers reach out for support?
- Issue complexity: Simple issues resolve faster (and cheaper) than complex ones
- AI effectiveness: As mentioned, better AI = more resolutions = higher costs
- Seasonality: Product launches, holidays, and promotional periods spike volume
- Product changes: New features generate new questions generate new resolutions
An analysis of 200+ Capterra reviews found a consistent theme: "founders and operators can't predict their customer support bills anymore. This makes budget allocation and other financial processes needlessly difficult."
One company's experience illustrates the problem: they budgeted $3,000/month for Fin based on historical support volume. During a product launch, volume spiked, resolution rate was high, and the bill came to $8,500. A 183% budget overrun from a single month.
The Compound Forecasting Problem
It gets worse when you consider that resolution-based costs compound with growth:
- More customers → more support interactions → more resolutions
- Better product adoption → more usage → more questions → more resolutions
- Successful marketing → user growth → support growth → resolution growth
Your AI support costs scale with your success. Unlike infrastructure costs (which have economies of scale) or human support costs (which scale sub-linearly with good tooling), resolution-based costs scale linearly—or worse—with volume. This mirrors the power user problem that many AI products face with usage-based pricing.
The "Soft Resolution" Loophole
Let's examine the most complex aspect of outcome-based pricing: soft resolutions.
A soft resolution typically means: the customer didn't ask for more help within some time window (usually 24 hours). The assumption is that silence equals satisfaction.
But silence can mean many things:
- Satisfaction: The answer worked, customer moved on
- Abandonment: The answer didn't help, customer gave up
- Distraction: Customer got pulled away, forgot about the issue
- Channel switching: Customer called phone support instead
- Delayed discovery: Customer tried the solution, found it didn't work, but it's been 25 hours
All of these count as "resolutions" under most providers' definitions. You're paying for outcomes that might not actually be outcomes.
One support team ran an experiment: they followed up with customers who had "soft resolution" conversations to ask if their issue was actually resolved. The result? Only 62% reported their issue was truly resolved. They were paying full price for a 38% false-positive rate.
The Budget Control Challenge
How do you control costs when they're driven by customer behavior rather than your decisions?
Rate Limiting AI
Some companies implement "resolution caps"—after X resolutions per month, Fin gets turned off and human agents take over. This defeats the purpose of AI automation and creates a terrible customer experience on the 1st of each month when caps reset.
Deflection Before Resolution
Others try to deflect customers to self-service before they reach the AI. The irony: you're now investing in infrastructure to prevent customers from using the AI you're paying for.
Selective AI Deployment
Some deploy AI only for certain issue types, customer segments, or times of day. Low-value customers get AI; high-value customers get humans. Simple issues get AI; complex issues go straight to humans.
This creates operational complexity that traditional support models don't have. And it still doesn't make costs predictable—it just shifts the unpredictability to "which issues end up with AI?"
The Annual Bucket Solution
Intercom introduced "annual buckets" to address some of this: buy a set number of resolutions upfront, burn them down over the year. This provides more flexibility than monthly commitments—if you have a spike in April, you're not paying overage rates.
But it creates new challenges:
- How many resolutions should you buy? (Back to forecasting problem)
- What happens if you burn through your bucket early? (Overage pricing)
- What happens if you don't use them all? (Wasted spend)
- How do you budget when you're prepaying for uncertain volume?
Finance Team Challenges
Let's talk about what this means for finance teams trying to manage AI costs.
Monthly Close Becomes Unpredictable
With seat-based pricing, you know your software costs before the month ends. With resolution-based pricing, you don't know the final number until the provider closes their billing cycle—often days after month-end.
Finance teams report delaying monthly closes waiting for AI vendor invoices. Or closing with estimates that require true-ups later. Neither is acceptable for serious financial management.
Variance Analysis Is Impossible
When your AI bill is 40% higher than last month, what caused it?
- More total support interactions?
- Higher resolution rate?
- Change in issue mix?
- AI model update from the provider?
- Seasonal factors?
- Product changes?
With traditional costs, variance analysis is straightforward. With outcome-based pricing, you need granular data that most providers don't expose.
Budgeting Becomes Fiction
How do you set an annual budget for resolution-based AI costs?
"We'll have 20% user growth, which should drive 15% more support volume, and our resolution rate should stay around 55%, so we're looking at approximately..."
This requires fundamentally different budgeting approaches than traditional SaaS spend.
What Billing Infrastructure Needs to Handle This
If outcome-based AI pricing is here to stay—and it appears to be—billing infrastructure needs new capabilities:
1. Resolution Tracking (Your Numbers, Not Just Theirs)
Track conversations and outcomes in your own system:
- Conversation opened
- AI responded
- Customer replied / abandoned / escalated
- Human took over (yes/no)
- Your determination: was this actually resolved?
Compare your counts to provider invoices. Identify discrepancies. Build leverage for disputes.
2. Resolution Quality Scoring
Not all resolutions are equal. Build your own quality assessment:
- Customer returned within 24-48 hours with same issue? (False resolution)
- Customer gave explicit positive feedback? (True resolution)
- Issue required human follow-up later? (Partial resolution)
- Customer churned shortly after? (Resolution didn't matter)
Use this to calculate "quality-adjusted resolution costs"—what you're really paying for outcomes that stick.
3. Predictive Cost Modeling
Build forecasting models that account for:
- Historical resolution rates and trends
- Seasonal patterns in support volume
- Correlation with product releases and marketing campaigns
- AI improvement trajectory (costs go up as it gets better, remember?)
Show finance teams a range: "Based on current trends, next month's AI resolution costs will be $4,200-$5,800 with 80% confidence."
4. Real-Time Budget Monitoring
Seat costs don't need real-time monitoring. Resolution costs do.
- Current month resolution count
- Projected month-end resolution count (based on daily trends)
- Budget remaining / projected overage
- Alert thresholds when projected costs exceed budget
5. Resolution Attribution
Which customers are driving resolution costs?
- Customer A: 47 resolutions, $46.53 in AI costs
- Customer B: 312 resolutions, $308.88 in AI costs
Now you can make decisions:
- Is Customer B's pricing tier appropriate for their support load?
- Should you introduce usage-based support pricing?
- Which customers are profitable after AI support costs?
Want to see resolution costs by customer? Bear Billing attributes AI costs at the customer level automatically, so you know exactly which accounts drive support spending. Request early access.
6. Provider Reconciliation
Automated reconciliation between your resolution tracking and provider invoices:
- Match your conversation IDs to their resolution charges
- Flag discrepancies for review
- Build audit trail for disputes
- Track resolution definitions changes from providers
The Coming Evolution: Outcome Verification
The next frontier in outcome-based pricing is AI verification of outcomes. Instead of trusting that silence equals satisfaction, use AI to evaluate whether resolutions were actually successful.
Intercom's Head of Pricing predicted "three big shifts: a move to more outcome-based pricing, differentiated pricing for different types of agents, and AI verification of resolutions."
This creates a fascinating recursion: AI judging AI. And it raises new questions:
- Who controls the verification AI?
- What if the verification AI is wrong?
- Does the verification AI have incentives aligned with yours or the provider's?
- How do you audit verification decisions?
Billing infrastructure will need to incorporate verification tracking, confidence scores, and dispute mechanisms for AI-evaluated outcomes.
Strategic Implications
For AI-Powered Companies
If you're building products that use outcome-based AI services, you need to decide: Do you pass through the pricing model or abstract it?
Pass-through: Charge your customers per resolution/outcome. Aligns your costs with your revenue. But exposes your customers to the same unpredictability you experience.
Abstraction: Charge your customers a predictable fee (subscription, seat-based) and absorb outcome-based cost volatility. Simpler for customers. Riskier for your margins.
Most companies are choosing abstraction, which means they need robust margin protection. Understanding the full cost structure—including resolution cost variability—is essential for sustainable pricing. For a deeper dive into this challenge, see our analysis of unit economics for AI products.
For Finance Teams
Outcome-based AI costs require new financial controls:
- Budget ranges, not budget points: Accept that AI costs will vary and plan for it
- Rolling forecasts: Update projections monthly based on actual trends
- Variance reserves: Set aside contingency for AI cost overruns
- Provider diversification: Consider multiple AI providers to reduce concentration risk
- Contract negotiation: Push for annual buckets, overage caps, and price protection clauses
For Procurement
Outcome-based pricing requires different vendor management:
- Definition audits: Understand exactly what counts as a billable outcome
- Resolution verification rights: Can you audit/dispute resolution counts?
- Definition change notifications: Require advance notice if they change what counts as resolution
- Competitive benchmarking: Compare effective cost-per-resolution across providers
Resolution Cost Management Checklist
Before signing or renewing an outcome-based AI contract:
- Document the resolution definition in writing, including soft resolution criteria
- Calculate effective cost-per-resolution using your historical support volume
- Model worst-case scenarios for product launches and seasonal spikes
- Set up independent resolution tracking in your own systems
- Establish reconciliation processes to compare your counts to provider invoices
- Negotiate definition change notifications requiring 90+ days advance notice
- Request audit rights to dispute resolution counts
- Budget in ranges, not point estimates, with 20-30% variance reserves
The Bottom Line
Outcome-based pricing for AI is not a simpler billing model. It's a fundamentally different billing model that requires fundamentally different infrastructure.
The companies succeeding with outcome-based AI costs share common practices:
- They track resolutions in their own systems, not just in provider dashboards
- They forecast based on multiple variables, not just historical averages
- They monitor in real-time, not just at month-end
- They attribute costs to customers and use cases, not just departments
- They have mechanisms to verify, dispute, and reconcile provider charges
The $0.99 resolution sounds cheap. At scale, with compounding growth, in an unpredictable support environment, it can become the most expensive line item in your AI budget.
Build the infrastructure to understand it, and you'll make outcome-based AI work for you. Ignore the complexity, and you'll face the same unpredictability others have reported.
Bear Billing tracks resolutions in your system, forecasts costs based on trends, and reconciles against provider invoices automatically. Request early access for visibility into your outcome-based AI costs.