How Much Does an AI Support Agent Cost in 2026? Four Pricing Models, Worked Out
September 15, 2026 · 8 min read · by the hiy team
The short answer
We make hiy, which sells a Support Agent, so read what follows knowing that. An AI support agent in 2026 is priced one of four ways: per outcome (Intercom's Fin charges $0.99 each time it counts a conversation as handled, on top of any seats), per seat (helpdesk suites that bundle AI into what you already pay per human agent), per message credit (Chatbase's tiers run from $40 for 700 credits to $500 for 15,000), or flat (hiy's Founding plan is $49 a month for 2,000 visitor messages across every agent you own, after a 14-day trial). Which is cheapest depends almost entirely on volume and on what the vendor counts: at three hundred conversations a month a flat plan wins clearly; at three thousand, a flat plan with a cap does not fit at all, and per-outcome pricing is the honest comparison. Every figure below was read from the vendor's own pricing page on 15 September 2026, and none of them is the real cost — the real cost is the documentation work and the person behind the handoff, which no price list shows.
The four pricing models
| Model | What you pay for | When it bites |
|---|---|---|
| Per outcome | Each conversation the vendor counts as handled — resolved, or the customer stopped asking | It gets more expensive precisely as the agent gets better, and the definition of "handled" is the vendor's |
| Per seat | The human agents in your helpdesk; AI is bundled or added per seat | You pay for people to unlock software meant to need fewer people; small teams pay for suites they do not use |
| Per message credit | A monthly allowance of credits, consumed per message or per model call | You have to know how many credits a conversation burns, and the pricing page may not say |
| Flat | One monthly fee for a stated allowance | The allowance is a ceiling, and a busy month can hit it |
Per outcome: Intercom's Fin
Read from intercom.com/pricing today: $0.99 per outcome, the same rate on every plan and on Fin bought standalone. An outcome, in the page's own words, is counted when "a customer confirms their issue is resolved, or they don't ask for more help after Fin responds, or Fin completes a workflow (Procedure), including handoffs." You are charged once per conversation even if several questions were answered. Standalone Fin has no seat costs and no setup or platform fees, but carries a minimum monthly commitment (the page's example is 50 outcomes). Intercom's seat plans, which all include Fin access, are Essential at $29, Advanced at $85, and Expert at $132 per seat per month.
Two things to notice in that definition. First, "they don't ask for more help after Fin responds" is a deflection-style count — it bills the conversation where the customer got a wrong answer and left, as well as the one where they got a right answer and left. Deflection vs resolution is the longer version of why that matters. Second, a handoff completed through a Procedure is itself a billable outcome, so the cost of the agent giving up gracefully is not zero.
Per-outcome pricing is honest in one way that flat pricing is not: you pay for what happened. It is awkward in another: the vendor decides what happened.
Per seat: the helpdesk suites
The large helpdesk platforms mostly sell AI as part of a per-agent seat, or as a per-seat add-on to one. We have not read their pages today, so no numbers here — but the model is worth understanding. It ties the price of the automated layer to the number of humans, which is backwards for a small team: two founders doing support pay for two seats to unlock an agent whose purpose is to need fewer of them. It suits teams that already run the suite for its inbox, routing and reporting, and want the agent inside the tool they live in.
Per message credit: Chatbase
Read from chatbase.co/pricing today: Free at $0 with 50 message credits a month (and agents deleted after 14 days of inactivity), Hobby at $40 for 700 credits, Standard at $150 for 4,000, Pro at $500 for 15,000, and auto-recharge at $40 per 1,000 extra credits. The page does not say how many credits a message consumes, and credit models commonly charge more credits for more capable models — so the number you need before you can compare, credits per conversation, is one you will have to measure yourself.
Credits are a fair way to charge for something whose cost varies. They are also the model most likely to surprise you in the third month.
Flat: hiy
hiy's paid plan is Founding Creator at $49 a month — a founding rate that the list price is stated to rise toward $79 once outcome proof exists, with founding subscribers keeping theirs. It carries 2,000 visitor messages a month, and that allowance is one pool for the whole account: a second agent divides the month rather than doubling it. A Support Agent is part of Founding, and creating one starts a 14-day trial with no card — one trial per account, ever. Citations on every grounded answer, the honest "I don't have that", and the handoff that records the conversation are on every plan, and no plan removes them. The Free plan is $0 with 300 messages a month for your own public twin; a Support Agent needs Founding or the trial that stands in for it.
Flat pricing is simple to budget and simple to outgrow. The cap is a real ceiling — past it, your agents stop answering visitors until the month turns over, and there is no overage to buy — and it is the reason the next section has two rows.
A worked example at two volumes
Assume a conversation is about three visitor messages, and the agent handles 60% of conversations without a person — both assumptions, not vendor figures.
| Monthly volume | Per outcome (Fin, standalone) | Flat (hiy Founding) |
|---|---|---|
| 300 conversations · ~900 messages | 180 outcomes × $0.99 ≈ $178, above the minimum commitment | $49, well inside the 2,000-message pool |
| 3,000 conversations · ~9,000 messages | 1,800 outcomes × $0.99 ≈ $1,782 | Does not fit: 9,000 messages is over the 2,000 cap, and hiy has no overage tier today |
Chatbase is not in the table because the credits-per-conversation figure it needs is not on the pricing page. If it is one credit per message, 900 messages lands on the $150 Standard tier and 9,000 on the $500 Pro tier; if it is more, move up a row.
The point of the table is not that flat is cheaper. It is that the models cross. Below roughly a thousand conversations a month a flat plan is hard to beat; well above it, the flat plan either does not exist or you are on a custom quote, and per-outcome pricing is the number to compare.
The costs no price list shows
- The documentation. An agent answers from your docs and is exactly as good as they are. Budget days, not hours, for making them answerable.
- The person behind the handoff. Every model above hands the stuck customer to someone. If that someone is you, the agent's cost includes your evenings.
- The test. Twenty questions before launch, and the queue of misses after. Cheap in money, real in time.
- The migration. A per-outcome vendor's transcripts, a credit vendor's trained agent — neither travels well. Ask what you can export before you sign.
Five questions to ask before you sign
- What exactly is counted — an outcome, a message, a credit — and who decides?
- Is a wrong answer the customer walked away from billed the same as a right one?
- Is the handoff free, or is giving up an outcome too?
- What happens in a month over the allowance: overage, a hard stop, or an upgrade?
- Are citations and the honest refusal on every tier, or is trust a paid feature?
The honest version
Our price is a founding rate, and it is low because the product is early: no resolution analytics, no helpdesk integrations, no self-service secret rotation, and a 2,000-message ceiling with nothing above it yet. Fin's $0.99 buys things hiy does not have — an inbox your whole team works in, Procedures, a decade of helpdesk. If you are a support organisation, the suites are priced for you and we are not. If you are a product team with docs, no support desk, and a few hundred conversations a month, the arithmetic above is the honest reason to look at a flat plan, and the reason we priced one.
Prices change; the four we quoted were read on 15 September 2026, and a vendor's own pricing page outranks this post the day it differs.
Where to start
Count last month's conversations and multiply. If the number is in the hundreds, run the five questions on a flat plan and a per-outcome plan and compare the answers, not the headlines. hiy's pricing page has the two plans and what every plan keeps; the Support Agent page has the live demo, so you can judge the answers before you judge the price.
Questions people ask
How much does Intercom Fin cost per resolution?
$0.99 per outcome on every plan, read from intercom.com/pricing on 15 September 2026. An outcome is counted when the customer confirms the issue is resolved, doesn’t ask for more help after Fin responds, or Fin completes a Procedure including a handoff — once per conversation. Standalone Fin carries no seat cost but a minimum monthly commitment.
Is per-outcome or flat pricing cheaper for an AI support agent?
Below roughly a thousand conversations a month a flat plan is usually cheaper — at 300 conversations, 180 outcomes at $0.99 is about $178 against a $49 flat plan. Well above that, a capped flat plan may not fit at all and per-outcome pricing is the honest comparison.
What does hiy’s Support Agent cost?
It is part of the Founding Creator plan at $49 a month — a founding rate — with 2,000 visitor messages a month pooled across every agent on the account. Creating a Support Agent starts a 14-day trial with no card. Citations, the honest refusal and the recorded handoff are on every plan.