Per-Seat vs Per-Conversation AI Pricing: Which Costs Less?

Every vendor will tell you their pricing model costs less. Per-seat vendors say a fixed fee protects you from a surprise bill. Per-conversation vendors say you pay only for what the agent touches. The useful version is more specific. Per-conversation costs less only past a volume that depends on your resolution rate, not on the headline price. Before you sign, you need one calculation and a short list of contract terms.

Start with the vendor comparison and the speed-to-lead workflow if you want the product context. This post is the cost math.

What is the difference between per-seat and per-conversation pricing?

Per-seat charges a fixed monthly fee for each agent you deploy, no matter how many conversations it handles. Published mid-market rates on major customer-service platforms run roughly $30 to $80 per agent per month. Enterprise configurations are quoted, not listed.

Per-conversation charges for each inbound interaction the agent touches. Some vendors charge per ticket, meaning every conversation started. Others charge per resolution, meaning only the conversations the agent closes without a person. Published per-conversation rates run $0.30 to $1.00. Per-resolution rates run $0.50 to $2.00. Those ranges are what vendors publish, and you should confirm them on the vendor's own page before you treat them as your price.

Hybrid pricing is a base subscription plus a usage charge above a ceiling. You get predictability at low volume, and the vendor gets growth at high volume. Forrester analyst Lisa Singer (July 2025) describes usage-based and hybrid pricing as accelerating, because APIs and agents now do tasks on their own, which makes a price tied to a human login less relevant.

A hybrid sits between a flat fee and a pure usage bill.

Hybrid pricing is the structure you will be offered when neither pure model feels comfortable to the vendor. You pay a lower fixed amount, then a per-conversation rate once volume passes a ceiling they set. At quiet volume the bill looks like a seat. At heavy volume it looks like usage, which is how the vendor keeps the upside. Singer's July 2025 note is the reason this middle form is spreading: once software does tasks without a person logging in, a price tied to a human seat explains less of the value. That does not make hybrid automatically fair. It means you still have to read the ceiling, the rate above it, and the definition of the unit, because the headline subscription is only the first line.

The unit definition can double the bill before the rate does.

Before you compare any number, write down what the vendor means by conversation, ticket, and resolution. A conversation can be one session, a thread across three days, or anything inside 24 hours. A resolution can mean the thread ended without an escalation, the user clicked resolved, or a satisfaction score cleared a bar. Those definitions can double your effective cost. Get them in writing.

Do this before you put either proposal in a spreadsheet. Ask for the rule in one sentence, then ask for the event in their system that creates a billable unit. If a three-day thread is one conversation on one proposal and three conversations on the other, the cheaper rate can be the more expensive contract. If a resolution includes threads the customer abandoned, you will pay for silence. Write the definition into the comparison itself so the rates are rates for the same thing.

When does per-seat cost less?

Per-seat costs less when volume is low relative to the fixed fee, or when the resolution rate is low. Both produce the same result: you pay more per finished conversation than the seat would have cost.

A low resolution rate makes a per-resolution price look cheap and bill high.

Vendors selling per-resolution pricing often cite 60 to 80% resolution in their materials. Early deployments often run 30 to 45%, because the agent has not learned your knowledge base, your edge cases, or your escalation rules. If you pay per resolution and the agent resolves 35% of 5,000 monthly conversations, you pay for 1,750 resolutions. At $1.00 each, that is $1,750 a month. Two seats at $60 are $120. In that scenario the seat wins by a wide margin, about fourteen times on these round figures. This arithmetic is illustrative, not a client result.

The trap is signing on the brochure rate. A 70% assumption on 5,000 conversations is 3,500 resolutions. The same volume at a first-month 35% is half the finished work and, on a per-resolution contract, still a large bill if the unit price was set as if the agent were already good. The seat does not care that the agent is still learning. That is why a low resolution rate and a low volume push the same direction. You are paying for attempts the per-resolution meter only partly admits, and the fixed fee stays smaller until the agent actually finishes enough work to justify the meter.

Steady volume turns a seat into a budget line you can forecast.

If inbound volume is steady and you can forecast it within 20%, a seat is a budget line that does not move. A firm handling 800 conversations a month with two agents knows the cost. A per-conversation bill at that volume can be smaller, until a launch or a seasonal spike doubles it. If the sales team has already quoted a customer a fixed service price, that variability is a problem.

The friction is practical. You told a customer what the service costs. The vendor's invoice moved because a launch doubled the threads, and now the margin on that customer moved with it. A seat sized for the steady 800 does not do that. It can be the wrong choice at a much higher volume, which is the next section, but for a book of business you can forecast within about a fifth, the still number is often worth more than a slightly lower average that you cannot promise.

When does per-conversation cost less?

High volume by itself does not make per-conversation cheaper. A company handling 15,000 conversations a month at a 70% resolution rate pays for 10,500 resolutions. At $0.75, that is $7,875. Ten seats at $60 are $600. In that scenario the seat costs 92% less. The example is illustrative, not a client result, and it is there to show the direction of the math.

Per-conversation wins in a narrower band: resolution above 65%, volume in the 3,000 to 8,000 range, and a unit price below $0.50. Below 3,000 conversations a month, the seat almost always costs less in absolute terms.

Three conditions have to be true at the same time.

Read the 15,000-conversation example as a warning, not as a reason to fear usage pricing in general. Ten seats at $60 stay cheap next to 10,500 paid resolutions at $0.75, which is why volume alone never settles the question. The band where the usage price can win is tighter than vendor pages imply. You want a resolution rate above 65%, so you are not paying a high unit price on a weak agent. You want volume roughly in the 3,000 to 8,000 conversations a month, large enough that the comparison is real and not so large that a seat count still wins by a mile. And you want a unit price below $0.50, so the meter cannot outrun the seats. Miss one of the three and rerun the math. Below 3,000 conversations a month, do not spend long on the usage proposal. The seat is almost always the smaller absolute number.

A seasonal business can waste a July-sized seat count in January.

Per-conversation does protect a seasonal business from paying for July's capacity in January. If the quiet month is under 40% of the peak month, model both. A business with 1,200 conversations in January and 6,000 in July pays for what it uses on a per-conversation contract, and wastes the July-sized seat count in January.

Model the year, not the peak month the vendor uses in the demo. Seats sized for July are idle money for every quiet month. Usage pricing follows January and July without asking you to guess the peak in advance. The cutoff worth testing is whether the quiet month is under 40% of the peak. If it is, put both models on the same twelve months of real counts before you decide the seat's predictability is worth the idle months. If your year is flat, this advantage disappears and you are back to the resolution rate.

How do you calculate the break-even volume?

You need four inputs: how many agents you would deploy on a seat plan (N), the monthly seat price (S), the per-conversation price (C), and your expected resolution rate (R).

Monthly seat cost is N times S. Per-ticket cost at volume V is V times C. Per-resolution cost is V times R times C. Set them equal and solve for V. Per ticket, V equals (N times S) divided by C. Per resolution, the break-even number of resolutions is (N times S) divided by (R times C).

An illustrative example, not a client result. Two agents at $60 is $120 a month. Per-resolution price $0.80. Expected resolution rate 55%. Break-even is $120 divided by (0.55 times $0.80), which is $120 divided by $0.44, which is 273 resolutions a month. At 273 resolved conversations both models cost $120. Above 273, per-resolution costs more. Below 273, per-resolution costs less. Run the same formula with your numbers before you sign.

Use a cautious rate for month one and the vendor's rate only later.

Use your first-month actual rate, not the vendor's benchmark. If you have no history, model 35% for month one, 50% for months two through four, and the vendor's claimed rate only from month five. Build all three. Do not sign on the optimistic one.

The 55% in the example is an input you replace, not a promise. A vendor slide that says 60 to 80% belongs in the month-five column until your own closed conversations produce it. If month one comes in at 35%, the break-even count moves, and a contract that looked cheaper on the slide can be the expensive one while the agent is still learning your cases. Build the three versions side by side, circle the month-one version, and sign against that one. You can reopen the comparison when the later months earn a higher rate. You cannot easily reopen a contract you signed on the highest rate.

What contract risks hide in per-conversation pricing?

Three risks show up often enough to budget for, and each can add 30 to 60% to the effective monthly cost.

An abandoned thread can be billed as a win.

Resolution definitions. If "resolved" means the conversation ended without an escalation, a customer who abandoned the thread counts as a win. A 70% rate built on abandonment is not a 70% rate built on a confirmed fix. Ask for the methodology, the API event that creates a billable resolution, and a sample of flagged conversations from current customers.

Put the sample next to the definition before you accept the rate. If the flagged threads include people who went silent, your 70% is partly a measure of giving up, and you will pay for it. Ask which API event creates the charge, in the vendor's words, and what a dispute looks like when you disagree with a batch of those events. A definition that lives only in a paragraph of the order form, with no event you can export, will be argued at renewal rather than checked in month two.

Pilot usage becomes the floor the vendor cites at renewal.

Forrester's read of first-half 2026 enterprise software earnings treats consumption pricing as a renewal floor. Credits and usage tiers feel flexible during a pilot. They also give the vendor a baseline at renewal. Ungoverned trial usage is a trail the vendor will use to set the next minimum.

The pattern Forrester describes is easy to miss while the pilot feels free. Credits, units, and tiers let you try the agent without a hard cap, and the usage you generate becomes the number the vendor brings to the renewal as the least you should commit. A trial with no owner, no cap, and no decision about which threads are in bounds is not flexibility you keep. It is evidence the vendor keeps. Decide the cap for the pilot in the same week you start it, and export the usage yourself so the renewal conversation uses your count.

Minimums and overages hide inside the advertised rate.

Minimums and overages. Many contracts make you pay for at least 2,000 conversations whether you hit them or not, and charge a higher rate above a ceiling. The effective price, once you include both, is rarely the advertised rate.

Read both numbers before you compare vendors on the unit price alone. A minimum of 2,000 means a quiet month is billed as if you had the traffic, which removes the seasonal advantage you thought you were buying. An overage rate above a ceiling means the busy month is billed at a worse price than the one in the proposal. Add the minimum, the included band, and the overage to a normal month and to a peak month. The rate you should compare is the one those months actually produce.

When is outcome-based pricing worth considering?

Outcome-based pricing ties the fee to a named result: a qualified appointment booked, a billing dispute resolved, a support case that does not come back. McKinsey's April 2026 survey of 419 B2B pricing executives found that 10 to 30% of organizations have adopted generative or agentic AI in pricing today, and 65 to 85% expect to within one to three years. Outcome-based structures are the model that survey ties most closely to a measurable result.

It works when the result is binary and the vendor cannot edit it.

It works when the outcome is binary, timestamped, and stored somewhere the vendor cannot edit. A calendar entry is the cleanest example. Softer outcomes, such as satisfaction or revenue, need an attribution agreement before you sign, and those agreements get hard at renewal. For a first agent, a seat is simpler to audit and per-conversation is simpler to model against volume. Outcome-based pricing is worth a look after you have 90 days of resolution data. The side-by-side of the structures is in how agent pricing models compare.

Appointment booking survives an argument because the entry is either on the calendar or it is not, the time is on it, and the vendor does not control the calendar. A claim about satisfaction, churn, or revenue needs a model of what else could have caused the change, and you will be negotiating that model again when the contract is up. Wait until you have 90 days of your own resolution data before you add that argument. Until then, a seat is easier to audit and a per-conversation price is easier to put next to volume. Outcome-based pricing is a later conversation, not the first contract.

What should you ask before you sign?

What event creates a billable unit? If they cannot name the API call, the definition is contractual, which means it is disputable. What is the monthly minimum, and what is the overage rate? Can the vendor use your consumption as the floor at renewal? For per-resolution contracts, who decides a conversation was resolved, how is a dispute handled, and how long is the appeal window?

Four questions depend on which model is in the contract.

Ask what event creates a billable unit, and ask them to name the API call. If they cannot, the definition is only in the contract, which means you will argue about it later. Ask the monthly minimum and the overage rate above the ceiling, as numbers. Ask whether the vendor can use your consumption as the floor when the contract renews. On a per-resolution contract, ask who decides that a conversation was resolved, how you dispute a batch, and how long the appeal window stays open. Each answer should fit in a sentence you can attach to the order. A paragraph that restates the brochure is not an answer.

Three questions apply no matter which model you pick.

What data do they retain, and can you export the logs when the contract ends? What is the latency commitment, and what credit do you get when the agent is down? Is the unit price fixed for the term, or can they reprice it at renewal? Each question takes a few minutes. The answers decide whether the contract protects you.

When you want the formula run against your actual volume, get your free Systems Plan. We will use your conversation count, not a vendor's benchmark.

How we know this.

McKinsey, April 2026, n=419 B2B pricing executives, is the source for the 10 to 30% and 65 to 85% adoption figures. Forrester's July 2025 note from Lisa Singer, and Forrester's first-half 2026 earnings note, are the source for the shift toward usage pricing and the renewal-floor pattern. The $30 to $80, $0.30 to $1.00, and $0.50 to $2.00 ranges are published vendor rates and should be rechecked on each vendor's site. Every worked example in this post is illustrative arithmetic, not a client result, and none of the figures are a price for our work. The break-even uses two seats at $60, a $0.80 per-resolution price, and a 55% resolution rate, and it is illustrative, not a client result.