Why the seat stopped making sense
Per-seat pricing worked because seats correlated with value. More people using the software meant more work being done in it, so charging per person was a rough but honest proxy.
An AI agent breaks that correlation completely. The whole proposition is that the work happens without a person, which means the better the product performs, the fewer seats it should need. A vendor charging per seat for agentic software is charging you less as its product becomes more useful, which no vendor will do for long.
So the unit moved. The interesting part is that vendors did not all move it to the same place, and the differences are not cosmetic.
Conversation, resolution, outcome: three units that are not interchangeable
| Vendor | Published rate | Unit | Billed when |
|---|---|---|---|
| Intercom Fin | $0.99 | Per outcome | The AI resolves without a human handoff |
| Zendesk AI Agents | $1.50 committed, $2.00 pay as you go | Per automated resolution | The AI resolves without a human handoff |
| Salesforce Agentforce | $2.00, or Flex Credits | Per conversation | The conversation happens |
Per conversation
You are billed for the attempt. Every inbound conversation the agent handles is chargeable regardless of what happens next. If the AI resolves it, you paid $2.00 and saved an agent's time. If the AI flounders and escalates, you paid $2.00 and a human handled it anyway. Failure is billable.
Per resolution and per outcome
You are billed for the result. Conversations that escalate to a human, that the customer abandons, or that the AI cannot close are not charged. The vendor absorbs the cost of its own failures.
The naming difference between Zendesk's resolution and Fin's outcome is largely branding. The mechanism is the same one: payment on successful end-to-end automated closure.
Why the same number is not the same price
Zendesk pay as you go and Salesforce Agentforce both publish $2.00. They are not comparable. At a 50% automated resolution rate and ten thousand conversations, Agentforce bills all ten thousand at $2.00 for $20,000, while Zendesk bills five thousand at $2.00 for $10,000. Same headline figure, double the bill.
This is the single most important thing to understand about the category, and it is invisible if you compare pricing pages as a list of numbers.
The variable that decides your bill is yours, not theirs
Under per-seat pricing you could calculate your bill from the pricing page. Under outcome pricing you cannot, because the multiplier is your own automated resolution rate, and no vendor can tell you what yours will be.
At Fin's $0.99 per outcome, with ten thousand conversations a month:
- At 30% resolution, three thousand outcomes, $2,970 a month.
- At 50%, $4,950.
- At 70%, $6,930.
- At 85%, $8,415.
Read that column carefully, because it contains something counterintuitive: your bill rises as the product gets better. That is the model working as designed. You are paying more because more work is being done without a person, and the comparison that matters is not last month's invoice but the cost of the human hours those resolutions replaced.
It does mean budgeting is different. A finance team used to a fixed per-seat line item will see a variable cost that grows, and will need the deflection figures alongside it to read the number correctly. Bring both to that conversation or the good news looks like overspend.
Pricing model, published limits and what each tier excludes, on the same fields for every product.
What counts as a resolution, and why you must ask
When the unit of billing is an outcome, the definition of that outcome is the contract. These are the questions that move a bill materially, and the answers vary by vendor.
- Does an abandoned conversation count? A customer who asks a question, gets an answer and closes the tab without confirming has arguably been resolved. Some models bill it, some do not.
- Does a reopened ticket bill twice? If a customer returns two days later on the same issue, is that a second resolution or a failure of the first?
- Is a partial handoff chargeable? The AI collects the details, identifies the issue, and passes a well-prepared case to a human. Real value delivered, and under a strict end-to-end definition, not billable. Under a looser one, billable.
- What happens to conversations the agent refuses? Out-of-scope questions routed straight to a human should not be chargeable, and generally are not, but confirm it.
- Is there a minimum commitment? Zendesk's $1.50 is explicitly the committed rate against $2.00 pay as you go. A 25% discount for commitment is real money, and a commitment you overshoot or undershoot has its own consequences.
- What does the platform underneath cost? None of these rates is standalone. Intercom's helpdesk starts at $29 per seat per month, and the AI pricing sits on top. The per-outcome figure is an addition to a per-seat bill, not a replacement for it.
That last point deserves emphasis. The seat has not disappeared, it has been supplemented. Most buyers in 2026 are paying per seat for the platform and per outcome for the automation, which is a more complex bill than either model alone.
This is the leading edge of a much wider repricing
Support is where outcome pricing landed first, because support has an unusually clean success condition: the case is closed or it is not. Most software does not.
But the pressure applies everywhere. Any category where AI reduces the number of people needed has the same problem with per-seat pricing, and the vendors in those categories are all looking for a defensible unit. Some of what they land on will be genuine outcome pricing. Some will be consumption pricing wearing an outcome label, where you pay per action taken rather than per result achieved, and the distinction will be exactly as consequential as conversation versus resolution is here.
Across the wider catalogue we maintain, the pattern of metering on something other than a seat is already common and predates AI: per node, per location, per device, per geographic region, per managed endpoint. What is new is the unit being tied to a result rather than to a countable object. That is a genuine improvement in alignment when it is real, and a rebrand when it is not.
The test is simple, and it is worth applying to any vendor claiming outcome pricing: does the vendor lose revenue when the product fails? If a failed attempt still bills, it is consumption pricing with better marketing.
Questions people ask
- Is per-resolution pricing cheaper than per-seat?
It depends entirely on your resolution rate and your conversation volume, which is why no honest comparison can be made from the pricing pages alone. What per-resolution pricing reliably changes is the risk profile: you stop paying for failed automation. Whether the total is lower depends on how much of your volume the AI actually closes.
- What is the difference between per conversation and per resolution?
Per conversation bills on arrival, so a conversation the AI cannot handle is still charged and is then also handled by a human, meaning you pay twice. Per resolution bills only on successful end-to-end automated closure. Salesforce Agentforce publishes $2.00 per conversation and Zendesk publishes $2.00 per resolution pay as you go: identical numbers, materially different bills.
- Why does my bill go up when the AI gets better?
Because you are being charged for work completed rather than for licences held. A higher automated resolution rate means more outcomes billed, and it also means more cases handled without a person. The invoice needs reading against the human hours displaced, not against last month’s invoice.
- Do I still need to pay for seats?
In most cases yes. These per-outcome rates sit on top of a platform subscription rather than replacing it. Intercom’s helpdesk starts at $29 per seat per month, with the AI priced separately. Budget for both.
- How do I estimate my automated resolution rate before buying?
Take your last quarter of tickets and classify them by whether a well-written help centre article would have answered them completely. That is a rough but useful proxy, and it is usually a lower number than vendors quote from their own benchmarks. Then run a paid pilot on real traffic, because the rate depends on your product, your customers and your documentation, none of which a benchmark accounts for.
Sources
Every price, limit and date above was checked against these pages on the day shown. Where a figure has since moved, the vendor’s own page is the authority and this one is a snapshot.
- 1AI customer service agent pricing comparison · Fin AIStates Fin at $0.99 per outcome, Zendesk at $1.50 per resolution committed or $2.00 pay as you go, Salesforce Agentforce at $2.00 per conversation or Flex Credits, and Intercom helpdesk from $29 per seat per month. Note this is a vendor publishing a comparison including its own product.Checked
- 2AI customer service agent pricing comparison · IntercomThe same comparison on Intercom’s own domain.Checked
- 3Agentforce pricing · SalesforceVendor page for the per conversation and Flex Credits models.Checked
- 4Zendesk AI agents pricing · ZendeskVendor page for automated resolution pricing.Checked