Per-seat charges a flat rate per agent regardless of usage. Per-resolution charges per ticket AI successfully resolves, with Intercom Fin published at $0.99 per resolution. Per-investigation charges only when AI actively investigates a ticket by querying production systems. For B2B technical support where 20-40% of tickets need investigation, per-investigation pricing can cost less at scale.
Per-seat pricing
Per-seat pricing is the easiest model to explain: you pay a flat monthly fee for each support agent who gets access to the AI product. Published examples include Kustomer IQ at $89 per agent per month and Zendesk AI add-ons starting around $50 per agent per month. The bill is tied to headcount, not actual AI work completed.
The advantage is predictability. If you run a 10-agent team on a $50 per-seat plan, your monthly cost is $500 whether the queue is quiet or overloaded. That makes budgeting simple, especially for finance teams that prefer software contracts to look like other SaaS tools. It also works well when ticket volume is stable, AI usage is modest, and the main goal is giving every agent the same tooling layer.
The tradeoff is idle spend. Ten agents at $50 each is still $500 per month even if AI only touches 100 tickets. If adoption drops, or seasonality cuts volume, the bill does not move down with usage. Per-seat pricing is usually the safest fit when you want a team-level AI layer, already know your headcount, and care more about cost certainty than aligning spend to actual automation output.
Per-resolution pricing
Per-resolution pricing charges only when the AI closes a ticket without human intervention. Intercom Fin is the clearest published example at $0.99 per resolution. Intercom also reports a 56% average containment rate, which gives a useful public baseline for estimating spend on a support queue with a similar ticket mix.
The benefit is that you are paying for visible outcomes rather than licenses. If the bot does not resolve the conversation, there is no resolution charge. For teams focused on ticket deflection, that can feel fairer than buying seats and hoping adoption follows. It also maps neatly to customer support KPIs such as containment rate, deflection rate, and cost per resolved conversation.
The math gets expensive fast at higher volume. On 1,000 tickets per month with 56% containment, you would see 560 billable resolutions. At $0.99 each, that is $554.40 per month. At 5,000 tickets, 2,800 resolutions cost $2,772 per month. In other words, per-resolution pricing can start below a seat-based contract at low volume and then overtake it once resolution count climbs. For a 10-agent team at $50 per seat, the crossover is roughly 450-500 monthly resolutions.
| Monthly tickets | Fin containment (56%) | Per-resolution cost @ $0.99 | Per-seat (10 agents @ $50) |
|---|---|---|---|
| 500 | 280 | $277.20 | $500 |
| 1,000 | 560 | $554.40 | $500 |
| 2,000 | 1,120 | $1,108.80 | $500 |
| 5,000 | 2,800 | $2,772 | $500 |
Per-investigation pricing
Per-investigation pricing charges only when the AI actively investigates a support ticket. That means querying production systems, pulling account state, tracing usage patterns, or helping diagnose root cause. The billable event is not a deflected FAQ conversation. It is a real investigation step on a ticket that needs system-aware analysis.
That distinction matters for B2B support. Many SaaS queues contain a large tier-1 layer of password resets, documentation questions, plan details, and status checks. Those tickets may be deflected at scale, but they usually do not need investigation. Investigation-heavy tickets are a smaller slice, often around 20-40% of B2B technical support volume depending on product complexity, integration depth, and customer profile.
For example, on 1,000 monthly tickets, a 30% investigation rate means 300 investigation events. On the same queue, a per-resolution product operating at 56% containment would bill 560 resolution events. That lower denominator is the core argument for per-investigation pricing. It works best when investigation cost is the real problem, not raw deflection volume. For teams dealing with billing issues, API failures, broken integrations, and account state problems, this model can align spend to the work that actually consumes technical support time. Pricing is usage-based, so the right way to compare is against your measured investigation rate.
Main pricing model comparison
| Pricing model | How you pay | Published example | Monthly cost (1K tickets, 56% resolution, 30% investigation) | Scales with | Key risk |
|---|---|---|---|---|---|
| Per seat | $/agent/month flat | Zendesk ~$50/agent, Kustomer IQ $89/agent | $500-$890 (10 agents) | Agent headcount | Fixed cost in low-volume months |
| Per resolution | Per ticket AI resolves | Intercom Fin $0.99/resolution | $554 (560 resolutions × $0.99) | Volume × resolution rate | Expensive at high volume |
| Per investigation | Per investigation event | Altor (usage-based) | Based on 300 investigation events | Investigation count | Requires measuring investigation rate |
Choose the model that matches the work
Choose per-seat if
You have predictable volume, want team-level AI tooling, and expect low variation in usage month to month.
Choose per-resolution if
Deflection is the main goal, you already run support on Intercom, and your monthly volume is still manageable.
Choose per-investigation if
You run B2B technical support, a meaningful share of tickets need investigation, and you want spend tied to actual technical work.
The vendor markup vs API cost gap
The API cost difference between GPT-4o and GPT-4o mini is tiny compared with the contract difference between pricing models. At 1,000 tickets per month, a GPT-4o workflow at roughly $0.01 per ticket costs about $10 in model spend. A GPT-4o mini workflow at roughly $0.0006 per ticket costs about $0.60. That is a $9.40 gap. A per-resolution contract at $0.99 on 560 resolved tickets adds about $554 instead. The pricing model creates cost differences that are 50-100× larger than model selection on the same ticket volume. Optimize the pricing model first, then optimize prompts, caching, batching, and model mix.
Model your queue before you sign a pricing contract
Compare your containment rate, investigation rate, and ticket volume against the contract denominator before you commit.
FAQ
What is per-resolution pricing in AI support?
Per-resolution means you pay a fee for each ticket the AI resolves without human intervention. Intercom Fin charges $0.99 per resolution. Cost scales with both ticket volume and containment rate. At Intercom's reported 56% average containment, 1,000 tickets per month costs about $554.
Is per-seat or per-resolution pricing cheaper?
For a 10-agent team at $50 per seat, or $500 per month, per-resolution at $0.99 becomes more expensive above about 505 resolutions per month. At 56% containment on 1,000 tickets, 560 resolutions cost $554.40, so per-resolution is already more expensive than per-seat for that team size.
What is per-investigation AI support pricing?
Per-investigation pricing charges for each ticket requiring active AI investigation — querying production systems, pulling account state, diagnosing root cause. Unlike per-resolution, it does not charge for simple FAQ deflections. This model aligns cost to the highest-value work in B2B support queues.
Does Intercom Fin charge per resolution or per ticket?
Intercom Fin charges $0.99 per resolution — per conversation that Fin handles without routing to a human agent. Tickets that Fin cannot resolve and passes to a human are not charged as resolutions.