Three Models, Three Incentives

Every AI support tool uses one of three pricing models. The model they choose tells you what they optimize for — and what costs you more than it should.

Model 1: Per-Seat (Traditional)

How it works: You pay a monthly fee per agent/user who has access to the tool. $15 to $165/user/month depending on the vendor and tier.

Who uses it: Zendesk, Freshdesk, HubSpot Service Hub, and most traditional help desks.

The incentive problem: Per-seat pricing incentivizes limiting access. You don't give your engineers access to the support tool because that's another $55/month. But your engineers SHOULD see bug reports directly. Per-seat pricing creates information silos.

When it works: Large teams (20+ agents) where the cost per seat amortizes over high volume. At 1,000 tickets per agent per month, $55/seat is $0.055/ticket — cheap.

When it hurts: Small teams (1-5 people) where each seat is a significant cost. A 3-person team on Zendesk Suite Growth pays $267/month regardless of whether they get 50 tickets or 500.

Model 2: Per-Resolution (Usage-Based)

How it works: You pay when the AI successfully resolves a customer question without human involvement. $0.20 to $1.50 per resolution depending on the vendor.

Who uses it: Intercom ($0.99), Zendesk AI ($1.50+), Gorgias ($0.90), Supp ($0.20).

The incentive alignment: You pay for results. If the AI resolves a ticket, you pay. If it escalates, you don't. This aligns the vendor's incentive with yours — they want the AI to work because they only get paid when it does.

The catch: Some vendors charge for "resolutions" even when the quality is questionable. If the AI gives a technically-complete-but-unhelpful answer and the customer doesn't follow up, that counts as a "resolution." Watch your resolution quality, not just quantity.

When it works: Any volume level, but especially for teams with predictable support patterns where automation can handle 60%+ of messages.

When it hurts: Very high volume without good automation. If the AI can't resolve and you're paying $0.99 per attempt, costs balloon. (Though most per-resolution models only charge for successful resolutions.)

Model 3: Per-Conversation (Hybrid)

How it works: You pay per customer conversation, whether it's resolved by AI or a human. Some tools count a "conversation" as a thread that might span multiple messages.

Who uses it: Intercom (base plan uses conversation-based billing), some newer tools.

The problem: This model charges you for conversations where the AI fails AND the human has to step in. You pay for the attempt and the fallback. It's the worst of both worlds if your automation rate is low.

When it works: High automation rates (80%+) where most conversations are genuinely resolved by AI.

When it hurts: Low automation rates. If the AI handles 40% and humans handle 60%, you're paying per-conversation for conversations that require human labor anyway — essentially double-paying.

The Math by Scenario

Scenario: 500 support messages/month, 3-person team, 70% automation rate

The difference between the cheapest ($87.50) and most expensive ($792) option is $704.50/month — $8,454/year. For the exact same outcome: 350 messages auto-resolved, 150 handled by humans.

What to Look For

Watch for hidden costs. Some per-resolution tools charge a base subscription ON TOP of per-resolution fees. You're paying twice.

Understand what counts as a "resolution." Is it one message classified? A full conversation completed? An issue marked as solved? The definition changes your effective cost per ticket.

Check the escalation cost. Per-resolution tools typically don't charge for escalations (the AI couldn't handle it, so no charge). Verify this. If they charge for failed attempts, that's a red flag.

Ask about volume discounts. Some tools offer lower per-resolution rates at higher volume. If you're processing 2,000+ messages/month, negotiate.

The Bottom Line

Per-seat pricing is the worst model for small teams. You pay for access, not results. You pay the same whether your team handles 50 tickets or 500.

Per-resolution pricing aligns incentives. You pay for automation that works. But the price spread — $0.20 to $1.50 per resolution — means your choice of vendor matters a lot.

Per-conversation pricing is the riskiest. You pay even when automation fails. Unless your automation rate is very high, you'll overpay.

For small teams, per-resolution at the lowest rate you can find is almost always the best deal. The math is simple and the cost scales proportionally with the value you receive.