AI Agent Pricing Models Compared: Per-Seat vs Per-Resolution vs Per-Token

AI agent pricing is a mess. Vendors use different models, bundle costs differently, and make direct comparison deliberately difficult. One vendor charges per seat, another per resolution, and a third per conversation minute. The cheapest option on paper can be the most expensive in practice—and vice versa.

Here's how each pricing model works, when each makes sense, and how to model costs for your specific situation.

Per-seat pricing

How it works: You pay a fixed monthly fee per user (agent, admin, or team member) who has access to the platform. Price typically ranges from $30–$200/seat/month depending on the tier and features included.

Common with: Horizontal agent platforms, CRM-integrated sales agents, team productivity tools. Examples include platforms that bolt AI capabilities onto existing per-seat SaaS products.

Advantages:

• Predictable budgeting. You know exactly what you'll pay regardless of volume.
• Simple to model. Multiply seats by price and you have your annual cost.
• Often includes unlimited usage within the plan, so high-volume teams get better unit economics.

Disadvantages:

• You pay the same whether a seat is used heavily or barely at all. Seasonal businesses and teams with variable workloads overpay during quiet periods.
• Per-seat pricing incentivizes vendors to limit seats rather than limit value. You may need to buy seats for people who only occasionally need access.
• Hidden caps are common—"unlimited" often means "up to 10,000 conversations per seat per month" buried in the terms of service.

Best for: Teams with consistent, predictable usage patterns and a fixed number of power users. Mid-market companies with 10–50 seats typically get fair value from per-seat pricing.

Per-resolution pricing

How it works: You pay only when the AI agent successfully resolves an interaction without human intervention. Prices range from $0.50–$5.00 per resolution, depending on complexity and the vendor's definition of "resolution."

Common with: Customer support agents, IT help desk agents, and voice agents for appointment booking. Vendors like Intercom Fin and similar support-focused platforms popularized this model.

Advantages:

• Direct alignment with value. You only pay when the agent delivers a result, not when it escalates or fails.
• Low risk for pilot programs. If the agent doesn't resolve anything, you pay nothing.
• Easy to calculate ROI. Compare the per-resolution cost against your current cost-per-ticket with human agents.

Disadvantages:

• "Resolution" definitions vary and matter enormously. Does a one-message FAQ answer count the same as a complex multi-turn troubleshooting conversation? Some vendors charge the same for both; others have tiers.
• Costs scale linearly with volume. High-volume support teams can end up paying more than a flat-rate model. A team handling 50,000 tickets/month at $1.50/resolution pays $75,000/month—potentially more than a per-seat plan.
• Vendors are incentivized to classify interactions as "resolved" aggressively. Track reopen rates and CSAT alongside deflection rate to ensure resolutions are genuine.

Best for: Companies deploying support agents with moderate volume (1,000–20,000 tickets/month) who want to pay for outcomes rather than access. Excellent for pilots where you need to prove value before committing to a larger investment.

Usage-based (per-token / per-minute) pricing

How it works: You pay based on actual consumption—tokens processed, API calls made, minutes of voice interaction, or messages exchanged. Pricing is granular: $X per 1,000 tokens, $Y per voice minute, $Z per API call.

Common with: Voice agents (per-minute), custom-built agents on LLM APIs, and developer-focused platforms. Low-level pricing where you assemble the agent yourself and pay for raw compute.

Advantages:

• Maximum flexibility. You pay precisely for what you use, with no wasted spend.
• Transparent cost drivers. You can optimize by switching models, caching prompts, or reducing token usage.
• Lowest floor. Small-scale deployments can cost single-digit dollars per month.

Disadvantages:

• Unpredictable costs. A sudden traffic spike, a chatty agent loop, or a poorly optimized prompt can cause bills to explode. Voice agents are particularly volatile—a 3-minute average call at $0.10/minute is manageable; a 15-minute edge case is not.
• Requires monitoring and optimization expertise. You need to understand tokens, caching, model routing, and cost-per-task to manage spend effectively.
• Comparison shopping is harder. Comparing $3/million tokens on one provider against $0.08/voice-minute on another requires modeling your specific usage pattern.

Best for: Engineering teams building custom agents who want granular control, startups with unpredictable or rapidly growing volume, and voice agent deployments where per-minute pricing is the industry standard.

Hybrid and tiered models

Many vendors combine elements:

• Base platform fee + per-resolution: A fixed monthly fee for the platform, integrations, and analytics, plus a per-resolution charge for AI interactions.
• Tiered usage: Included resolutions in the base plan (e.g., 1,000/month), with overage charges for additional volume.
• Model-dependent tiers: Different prices based on which AI model the agent uses (GPT-4o vs. a smaller model), letting you optimize cost versus quality.

Hybrid models are increasingly common and often the most fair—you get predictable base costs with variable costs that scale with value delivered.

How to model costs for your situation

Run this analysis before committing to any vendor:

1. Estimate your monthly volume. How many interactions, tickets, leads, or calls will the agent handle? Use your last 6 months of data and project 20% growth.

2. Estimate resolution rate. Conservatively assume 40–50% AI resolution rate in the first 3 months, growing to 60–70% as you optimize. Don't use the vendor's "up to 80%" claim—use your own pilot data.

3. Calculate cost under each model. For a team with 20 seats, 10,000 monthly tickets, and 50% AI resolution:

• Per-seat at $100/seat: $2,000/month regardless of volume
• Per-resolution at $1.50/resolution: 5,000 resolutions × $1.50 = $7,500/month
• Usage-based at $0.05/conversation: 10,000 × $0.05 = $500/month (but add LLM costs, infrastructure, and engineering time)

4. Account for total cost of ownership. Per-seat and per-resolution prices usually include infrastructure, while usage-based pricing requires you to manage and pay for infrastructure separately. A fair comparison adds $1,000–$5,000/month in ops overhead to usage-based pricing for non-trivial deployments.

5. Negotiate. AI agent pricing is rarely fixed. Volume commitments, annual contracts, and competitive quotes all create leverage. Vendors routinely discount 20–40% from list pricing for committed annual deals.

The pricing model tells you something about the vendor

Per-seat vendors are betting their agent is sticky enough that you'll renew. Per-resolution vendors are confident their agent actually resolves issues. Usage-based vendors are targeting builders who want raw components, not packaged solutions.

Match the pricing model to your confidence level: if you're experimenting, per-resolution or usage-based pricing de-risks the investment. If you've proven value and want predictability, lock in a per-seat or committed annual deal.