Which AI Support Tools Charge Per Conversation, Resolved or Not? (2026)

Which AI support tools charge per conversation?

Macha publishes this index and is one of the vendors below. These are the per-attempt vendors from the AI Customer Support Pricing Index, with the unit each one bills in its own words. We checked each price on the vendor's own page on 24 and 25 September 2026 and list the headline rate.

Front sits just outside the list. Its Autopilot agent is "Starting at $0.05 /conversation" beside a Talk to Sales button, and Front doesn't define the billable conversation, so the index doesn't count it as a usable price. See the Front pricing guide.

How we classified each vendor

A vendor is per-attempt when its own pricing page or billing docs name a conversation, session, ticket or task as the unit, rather than a resolution or outcome. Where the page also says whether unresolved or escalated attempts are billed, the table quotes it. Where it doesn't, the row says so. Richpanel is the hard case. Its visible pricing card and calculator code price "AI-handled" conversations at $0.20, while its llms.txt says "about $0.20 per AI-resolved conversation". We classed it on the pricing page and carry it at medium confidence.

What does the per-conversation price cost per resolution?

Divide the per-attempt price by the share of handled tickets the AI resolves. At 50%, every price doubles. The break-even columns show the resolution rate above which the per-attempt vendor costs less per resolution than four per-resolution prices: HubSpot's $0.45, Help Scout's $0.75, Intercom Fin's $0.99 and Zendesk's $1.50 committed rate.

Against HubSpot's $0.45, only Richpanel and the $0.40 agents ever come out ahead, and the $0.40 agents need an 88.9% resolution rate to do it. Against Zendesk's $1.50, five of the nine win at a 35% resolution rate (Richpanel, eesel AI, Macha at a tier cap, Freshworks and Botpress), and six win at 50%. The break-even ignores fixed fees, so the worked month below adds them back. The eight per-resolution vendors themselves are compared in our cost per resolution breakdown.

What resolution rates do teams actually report?

The break-even only helps if you know your rate, and vendors' marketing numbers aren't a good guide. Merchants reviewing AI agents on the Shopify App Store in 2026 report a wide spread. Tediber said Yuma AI closes "30% of our tickets without any human interaction" (20 March 2026), and MESHKI US reported "35% automated resolutions in less than 6 months" (11 February 2026). Joli Rêve wrote that "with proper setup and time investment, you can realistically aim for 50% full automation" (19 March 2026). On Gorgias, Alessi USA reported exceeding "a 40% automation rate" at peak (19 June 2026), and Alessi's EU business reported automating "over 50% of customer inquiries" (1 September 2026). At the top, Happy Hygge Gifts said its "AI resolution rate is around 90%" with Tidio's Lyro and Flows (16 September 2026). The reviews are on the Yuma AI (5.0 from 9 reviews), Gorgias (4.3 from 705) and Tidio (4.8 from 1,342) listings, ratings as of 25 September 2026.

Read against the break-even table, a team at 30% to 35% pays less per resolution on a $0.40 per-ticket agent than on Zendesk, and more than on Fin. At 50% it beats Fin too. At 90%, even Tidio's $0.70 works out to $0.78 per resolution. One merchant review also shows how per-conversation billing confuses buyers: Direct Discount Furniture & Mattress wrote that Tidio's "plans, billable conversations, and included features are presented differently depending on where you look" (21 May 2026).

What should you ask a per-conversation vendor?

• What starts a billable unit? One reply, two user messages, a 72-hour window or a ticket.
• Does spam or bot traffic count? Alhena AI says spam and irrelevant queries cost zero credits; ask everyone else.
• Does a conversation that spans two months bill twice? Botpress's FAQ says it counts "in both months".
• Do anonymous or repeat sessions count separately? Botpress counts "multiple anonymous sessions" as separate conversations.
• Is a handoff to a human billed? It is in most per-attempt models; Alhena AI skips immediate handoffs.

What does a month of 1,000 tickets cost at 45% resolved?

Take 1,000 tickets a month with the AI resolving 450. Each total includes the one seat or plan the vendor requires, following the index's method; the four standalone agents also need the help desk they run on, which isn't included.

At 45%, the two lowest per-attempt prices (Richpanel and eesel AI) come in under Fin. Freshworks, Tidio, Macha and Botpress land between Fin and Zendesk, and the vendors with a platform fee or a $2 conversation cost the most. Our 1,000-tickets-a-month breakdown has the same vendors at 1,000 billable units, and the cheapest AI support agent ranking sorts them by total.

Why do some vendors charge per attempt?

The billing unit decides who carries the outcome risk. A per-attempt vendor earns the same on a ticket its AI fumbles as on one it closes, so its revenue follows your volume, not its performance, and you get a predictable invoice in return. A per-resolution vendor earns only on successes, which lines its incentive up with yours, but it also writes the definition of "resolved", and the definitions differ: Help Scout and Intercom count a customer who doesn't ask for more help, and Gorgias counts a ticket with no human interaction for 72 hours.

Per-attempt billing also pays for work that isn't a resolution. A per-ticket agent that tags, drafts a reply for a human or collects order details before a handoff has done work the per-resolution model gives away or bills separately. Botpress's pricing FAQ says "AI-only and human-assisted conversations are counted the same", and any exchange with at least two messages from the end user counts. Our longer comparison of the two models is in per-action versus per-resolution AI pricing.

What changed with Salesforce's $2 per conversation?

Salesforce's Agentforce pricing page still carries a Conversations card at "$2 USD/Per conversation" in its source. On 25 September 2026, the page's own stylesheet hides that price on screen and shows "For existing Agentforce Conversations customers. Contact your Account Executive for more information." in its place. Its FAQ also tells customers on conversation pricing that switching to Flex Credits means swapping "all of your existing Agentforce Conversation SKUs".

A new buyer's published route is Flex Credits: $500 per 100,000 credits, and "Agentforce actions are 20 Flex Credits", which is $0.10 an action. A conversation that takes four actions costs $0.40 under that model, and one that takes twenty costs $2.00. The index keeps Salesforce's $2 row for existing customers and flags it.

Which tools charge per message, credit or token?

Four vendors bill a unit smaller than a conversation, so the charge moves with how chatty the AI is rather than with how many customers write in:

• Chatbase bills a message credit per AI reply: 1 credit on basic models and 2 to 6 on premium ones, from its $40 Hobby plan with 700 credits.
• Microsoft Copilot Studio bills Copilot Credits at $0.01 pay-as-you-go: 1 for a classic answer, 2 for a generative one and 5 for an agent action.
• Voiceflow bills $0.005 per user message plus LLM tokens by model.
• Zoho Desk bills Zia Agents by LLM token, with 30 million tokens a month free on its standard models.

None of the four charges more for a resolved conversation than an unresolved one of the same length.

Which per-resolution vendors still charge when a human finishes the ticket?

Two of the eight resolution-billed vendors bill some conversations that end with a person:

• Intercom counts a Fin outcome when Fin "completes a workflow (Procedure), including handoffs", so a conversation a person finishes can still bill $0.99. The Fin pricing breakdown walks through it.
• Gladly charges $0.25 per AI Assist on its Shopify app, for a conversation the AI worked on before handing to a person.

Gorgias is a different case. Its AI fee applies only to automated interactions, but its help desk ticket fee applies to every ticket, and its billing documentation says an AI-resolved ticket pays both. The Gorgias pricing guide has the plan details.

Seat-priced AI add-ons bill regardless of outcome too. Freddy AI Copilot is $29 per agent a month on Freshdesk, and Front's Copilot is $20 per seat a month, per their pricing pages on 25 September.

Where does Macha fit?

Macha publishes the index and is in the per-attempt group. It bills per ticket, one thread with one person however many replies it takes, whether or not the AI resolves it. Each tier cap works out to $0.40 a ticket, and between caps the effective rate is higher: 1,000 tickets fall on the 1,500 tier at $599, or $0.60 each, which is why it loses to Fin at 45% resolved in the worked month. There are no seat or platform fees, and setup and monitoring by the Macha team are included. Macha fits teams already running Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom whose volume sits near a tier cap and whose AI resolves well above the break-even for that tier, or who want the agent to tag, draft and route the tickets it doesn't close. If you expect a low resolution rate and want to pay nothing for failures, a per-resolution vendor is the cheaper fit.

Frequently asked questions

Which AI support tools charge per conversation? Nine of the 17 vendors with a comparable published price bill per conversation, session or ticket: Richpanel, eesel AI, Macha, Freshworks, Botpress, Tidio, Siena AI, Alhena AI and Salesforce. Their list prices run from $0.20 to $2.00 per unit.

Is per-conversation pricing cheaper than per-resolution pricing? Only above the break-even resolution rate. A $0.40 per-ticket agent beats Fin's $0.99 above 40.4%, Zendesk's $1.50 above 26.7% and HubSpot's $0.45 only above 88.9%.

Does Freshdesk charge per resolution? No. Freddy AI Agent bills per session at $49 per 100, the first 500 sessions are a one-time allowance per account, and an email session covers 72 hours from the customer's first email.

Can new customers still buy Agentforce at $2 per conversation? Salesforce's pricing page labels that option "For existing Agentforce Conversations customers" as of 25 September 2026. New buyers are pointed to Flex Credits at $0.10 per standard action.

Which vendors bill only real resolutions? Help Scout, Zendesk, Lorikeet, Yuma AI and HubSpot publish per-resolution prices with no charge for handoffs. Most still have a fixed cost first: a help desk seat (Help Scout, Zendesk, HubSpot, plus HubSpot's $1,500 onboarding), a plan fee (Lorikeet's $2,100 a month) or an annual bundle (Yuma AI from $14,400).

Published list prices checked 24 and 25 September 2026. To see where your own queue lands against the break-even table, start a trial with $50 of free usage (about 125 tickets), no credit card, no time limit, or compare tiers on the pricing page.