AI Customer Support Pricing: A Worked Cost Model (2026)

At 3,000 conversations a month with half resolved by an AI agent, the AI line outgrows the seat line on every platform priced here, Plain included. What decides your bill is what each vendor counts as billable, not its headline rate. So this page prices five vendors (Plain, Intercom, Zendesk, HubSpot and Freshdesk) on their own published price lists and their own billable definitions, with every assumption in a table. Plain is the agentic infrastructure for customer work. Teams find it today under B2B customer support and help desk software. We publish this page, Plain is one of the vendors priced, and it doesn't come out cheapest under these assumptions. Every price below is taken from that vendor's own page, checked in October 2026.

What does AI customer support cost at 3,000 conversations a month?

Under the assumptions below, monthly totals run from $1,260 (HubSpot Service Hub Professional) to $2,895 (Zendesk Suite Professional), and on every vendor priced the AI line is larger than the seat line. That second fact matters more than the ranking: once an AI agent resolves half your volume, the metered line is where the money goes. Every input is in Table 1 so you can swap in your own numbers.

Table 1. Assumptions

The non-Plain inputs come from each vendor's own pricing page, as of October 2026:

• Intercom (pricing): Advanced $85 and Expert $132 per seat a month, billed annually; Fin AI Agent from $0.99 per outcome. Fin can also run on top of another helpdesk with no seat cost, at $0.99 per outcome with a monthly minimum (the page's example is 50 outcomes).

• Zendesk (pricing): Suite Professional $115 per agent a month, billed annually, with 5 automated resolutions per agent a month in the plan and additional ones at $1.50 each committed or $2.00 pay-as-you-go.

• HubSpot (Service Hub pricing): Professional $90 per seat a month billed annually, with 3,000 HubSpot Credits a month and a required one-time $1,500 onboarding fee outside the subscription.

• Freshdesk (Omni pricing): Pro $79 per agent a month billed annually; Freddy AI Agent sessions at $49 per 100 after a one-off 500 free sessions per account.

Table 2. Monthly and annual totals at 3,000 conversations a month

Plain's line, as of October 2026, uses Horizon at $99 per user a month billed annually, which carries 2,500 credits a month. The Customer Agent (formerly Ari) uses 100 credits for each thread it resolves. To redo the line for your own volume: multiply resolutions by 100, subtract the plan's monthly allowance (Horizon 2,500, Foundation 1,000, both on annual billing), divide what's left by 1,000, round up, and multiply by $10. Volume pricing arranged through Sales can lower this; the table shows list price.

Sidekick sits outside the base as a separate assumption. If your team runs 200 Sidekick investigations a month at around 200 credits each (Plain's docs give "investigate a straightforward bug" as a roughly 200-credit example), that's 40,000 credits, or 40 top-up thousands at $10 per 1,000 credits: about $400 a month on top of the Plain line. Agent assist is a separate line at the other vendors too, as of October 2026: Intercom's Copilot is a $29 per agent a month add-on billed annually, Freshdesk's Freddy AI Copilot is $29 per agent a month, and Zendesk lists Copilot as an add-on for Professional and higher without a price on its pricing page. None of these is in Table 2.

Pylon is described here but not modelled. As of October 2026 it publishes seat prices (Professional $89 and Advanced $139 per seat, on annual contracts) and credits at $0.30 each pay-as-you-go, but no fixed credit cost per automated ticket, so there is no unit to put in Table 2.

What does each vendor actually bill its AI agent for?

Three structures cover every vendor here. Plain, Intercom, Zendesk and HubSpot bill per resolution or outcome, though each defines that word differently. Freshdesk meters per session, whether or not the agent resolves anything. Pylon charges seats plus a pooled credit balance, and Plain and HubSpot also pool credits across features: in Plain, one balance pays for the Customer Agent, Sidekick, connected agents (15 credits per discussion and 15 per thread assignment) and AI-selected labels or priority, and nothing else draws on it.

Table 3. What each vendor bills its AI agent on (as of October 2026)

On Plain, a thread uses Customer Agent credits only when the Customer Agent replied at least once, resolved the thread without a team member getting involved, and the thread moved to Done and stayed there for 7 days. A customer reply inside those 7 days takes the thread out of Done, and the clock restarts when it returns. A thread the Customer Agent hands to a person uses no Customer Agent credits, and internal notes from the team don't count as involvement. If the workspace runs out of credits, the Customer Agent stops replying and hands each thread to the team with an internal note; the full rule and the Sidekick credit examples are in Plain's credits documentation.

Intercom bills two kinds of event. Fin counts an outcome when the customer asks for nothing more after its final answer, which covers a customer who confirms and a customer who simply leaves (an "assumed" resolution), and it also bills a Procedure set up to pass the conversation to a person or a workflow; a resolution the customer later reopens is deducted (Intercom's outcomes article). Zendesk goes the other way: since the tiers it introduced on May 18, 2026, only a verified resolution, where the customer didn't ask for more and an LLM check confirms it, counts against the allowance, while contained conversations and assisted escalations don't (Zendesk's automated resolution tiers). HubSpot's customer agent uses 50 credits per resolution, and the charge is stopped only by a visitor-initiated handoff; a manual assignment to a human, including through a workflow, doesn't by itself prevent the conversation counting as resolved once 72 hours pass after the visitor's final message (how HubSpot counts a resolution).

Freshdesk bills the attempt, not the result. Its billing unit is the session: on chat, everything within 24 hours of the start; on email, a 72-hour window from the customer's first message. A session is billed whether or not the agent resolves it. At a 50% resolution share you pay for 3,000 sessions to get 1,500 resolutions, which is the whole of Freshdesk's $1,470 AI line in Table 2.

The question to put to every vendor is "show me the event that creates a charge", not "what is the rate". A rate you can read is only worth something if the count behind it is one you can check.

Do people who only answer a few tickets need a paid seat?

In at least 14 conversations with B2B support leaders and engineers at 13 companies (June 2025 – September 2026), buyers raised the cost of paying a full seat for people who answer support only occasionally: engineers on rotation, on-call staff, CSMs and other light users. That includes objections to seat pricing in general, Plain's included. It's a raw count, not a share, but it comes up often enough that the seat rules deserve the same scrutiny as the AI unit.

On Plain a member seat is what a user needs to reply to customers. Viewer seats are free and unlimited on every plan, so engineers, product managers and account owners who read threads and add internal context cost nothing. Foundation is $45 per user a month (billed annually) for up to 5 member seats; Horizon is $99 per user a month (billed annually) for 3 to 10; Frontier covers 10 or more member seats, annual billing only, by quote.

As of October 2026, Intercom gives Advanced 20 free Lite seats and Expert 50 for internal collaboration; Lite seats can't reply to customers, and replying needs a Full seat. Pylon charges for seats and AI credits and requires new customers to buy a minimum number of seats. Zendesk, HubSpot and Freshdesk price per agent or per seat on the plans modelled here.

Fly.io migrated from Help Scout to Plain and built a custom support portal using Plain's API, streamlining workflows for their 9-engineer support team and rotating on-call engineers. Plain's case study is titled How Fly.io saves 200+ hours of engineering time per year with Plain (Fly.io on plain.com). The question to settle before you price seats is who on your rota needs to reply, and who only needs to read.

The modelling consequence is simple. Table 2 prices only the 6 replying seats. If your engineers reply directly to customers, they are member seats on Plain and Full seats on Intercom, and the seat line grows with them whatever the AI agent does.

What gets gated behind a higher tier?

Each row in Table 2 sits on a specific plan because of what the entry tier leaves out, as of October 2026:

• Plain: Horizon carries SLAs, business hours, escalation paths, Microsoft Teams, CRM integrations, suggested responses and reporting themes. Foundation caps at 5 member seats and Horizon at 10. Frontier, by quote, carries SSO and SCIM, Discord and custom channels.

• Intercom: SLAs and SSO sit on Expert, which is why Table 2 shows both the Advanced and the Expert row.

• Zendesk: SLAs are on Suite Professional, not Suite Team; Copilot can be added on Professional and higher; Suite Enterprise is by quote.

• HubSpot: the customer agent needs Professional or Enterprise; Enterprise adds conditional SLAs, skill-based routing and a sandbox; both carry a one-time onboarding fee, $1,500 and $3,500.

• Pylon: SAML SSO is an add-on on Advanced and included on Enterprise, which is a custom quote.

Across these five, the tier you need tends to be set by SLAs and SSO more often than by the AI agent itself. If you're on a contract with response-time commitments, price the plan that carries them, not the one that first lists the agent.

What does it cost over three years at 15% quarterly growth?

The three-year model assumes volume starts at 3,000 a month and grows 15% each quarter (month 36 is about 4.65× month 1), a resolution share fixed at 50%, one member seat per 250 human-handled conversations rounded up, each quarter as three months at that quarter's volume, and list prices unchanged. On that path the volume is 3,000 conversations, 1,500 AI resolutions and 6 seats in the first quarter; 5,247, 2,624 and 11 in the fifth; 7,980, 3,990 and 16 in the eighth; and 13,957, 6,979 and 28 in the twelfth.

Table 4. Three years at 15% quarterly growth (USD, annual list prices)

This model passes Horizon's 10-seat ceiling in the fifth quarter, so Plain's list prices cover year one only, and years two and three need a Frontier quote. The Plain AI line in years two and three is shown at list top-up price, before any Frontier allowance or volume pricing. The same limit applies to everyone: Zendesk Suite Enterprise, HubSpot Enterprise, Intercom's high-volume Fin pricing, Pylon Enterprise and Plain Frontier are all quote-only, so no three-year model of list prices is complete past the point you need them.

Over three years the AI line is the larger line for every vendor priced. Growth in AI resolutions, not headcount, is what moves the total.

Which pricing structure will you regret?

Per-session metering, if your resolution share is low. You pay for every attempt. In Table 2 that means 3,000 billed sessions for 1,500 resolutions, and the ratio only improves as the agent gets better at closing threads on its own. If you're starting with an agent that resolves a third or less, a session meter charges you most heavily during the period you're least sure it works.

A per-resolution or per-outcome price with a loose definition. An "assumed" resolution, or a billed handoff, turns a price you can read into a count you can't audit. The fix is procedural: before you sign, ask for the event log that the invoice is built from, and check a sample of charged events against the conversations behind them. Plain itemises credit use in Settings → Usage, and Plain's docs define the billable event (replied at least once, no team member involved, Done for 7 days).

Paying a vendor's agent to do what your own agent already does. If you've built an agent of your own, the cheapest AI line may be the one you don't buy. Plain lets you define, build and run your own agent and connect it to the same queue through Plain's API; it uses a fixed 15 credits per thread assignment and 15 per discussion, whatever the work, from the same balance as the Customer Agent (connecting your own agent). Model that line from your own resolution path before you price a vendor's agent on top of it.

Seat-led pricing when engineers answer customers. The seat line follows headcount you didn't hire for support. Viewer seats, Lite seats and similar help only if those people never reply; the moment an on-call engineer answers a customer directly, that person is a replying seat on Plain and a Full seat on Intercom. Price the seat rule against how your team actually works, not against the org chart.

Here's what this costs Plain to say. On the numbers in this model Plain is not the cheapest option. Its AI line rises with every thread the Customer Agent resolves, like every per-resolution model. The case for Plain is a billable definition you can audit, viewer seats that stay free as engineers rotate through support, and the Customer Agent, Sidekick and your own agents running on one queue and one credit pool. It isn't a lower bill. If you're still choosing tools rather than pricing them, the guide to AI support tools for B2B SaaS covers what each one does.

Sidekick is Plain's agent-facing assistant. It has no channel to the customer: it investigates across your integrated tools and hands off to a person or to the Customer Agent, which writes and sends the reply. It draws on the same credits, typically 15 to 300+ per request depending on effort, which is why the model prices it separately, after Table 2, outside the base.

If you want to run your own volumes against Plain's credit model rather than ours, book a demo and bring your conversation counts and resolution share. There's also a 7-day free trial, no credit card required.

Frequently asked questions

Which support platform's pricing stays predictable as you scale?

None of them stays flat once an AI agent resolves a large share of volume; every vendor here meters the AI line. Predictability comes from a billable definition you can audit and a usage view you can read. On Plain, a thread is charged only after it stays in Done for 7 days, and credit use is itemised in Settings → Usage. HubSpot Service Hub Professional has the lowest total in this model.

What does AI customer support cost for 3,000 conversations a month?

Under the assumptions in Table 2 (6 replying seats, half of conversations resolved by the AI agent, annual list prices), monthly totals run from $1,260 on HubSpot Service Hub Professional to $2,895 on Zendesk Suite Professional. Plain Horizon comes to $2,074, Intercom Advanced $1,995 and Freshdesk Omni Pro $1,944. On every vendor priced, the AI line is larger than the seat line.

Is per-resolution pricing cheaper than per-conversation or per-session pricing?

It depends on your resolution share. At 50%, a session meter bills twice as many events as a resolution meter, because every attempt counts. Per-resolution prices rise as the agent improves, so a better agent means a bigger AI line, and the definition of "resolved" decides how much of that growth is real. Check the definition before the rate.

What is usage-based helpdesk pricing?

Seats plus a metered AI unit. The unit differs by vendor as of October 2026: Plain charges 100 credits per thread the Customer Agent resolves; Intercom Fin charges per outcome from $0.99; Zendesk charges per verified automated resolution beyond its allowance; HubSpot's customer agent uses 50 credits per resolution; Freshdesk charges per session at $49 per 100; Pylon charges credits at $0.30 pay-as-you-go.

How does Plain charge for the Customer Agent?

The Customer Agent uses 100 credits for each thread it resolves, drawn from the plan's monthly allowance: 2,500 credits a month on Horizon and 1,000 on Foundation, both on annual billing. Top-ups cost $10 per 1,000 credits and never expire. A thread counts as resolved when the agent replied at least once, no team member got involved, and the thread stayed in Done for 7 days.

Do engineers who only read tickets need a paid seat on Plain?

No. Viewer seats are free and unlimited on every Plain plan, so engineers, product managers and account owners who read threads and add internal notes cost nothing. A member seat is needed only to reply to customers. Foundation covers up to 5 member seats at $45 per user a month billed annually; Horizon covers 3 to 10 at $99 per user a month billed annually.