AI Customer Support Pricing Index 2026: Cost per Resolution for 32 AI Agents

What is the average price per AI resolution in 2026?

The AI customer support cost per resolution, at the median published price, is $0.995, across the eight vendors that price per resolution: HubSpot ($0.45), Help Scout ($0.75), Intercom ($0.99), Lorikeet ($0.99), Gorgias ($1.00), Yuma AI ($1.20), Zendesk ($1.50) and Gladly ($1.50). The mean is $1.05. Two of the eight sit below $0.99 and two sit at $1.50, so the market's list price in September 2026 is about a dollar a resolution. Three of the eight (Gorgias, Gladly and Intercom) also bill some tickets that end with a human; the section on billing whether or not the AI resolves has the details.

Widen the net to every vendor that prints a comparable unit price, including the ones that charge per conversation, session or ticket whether or not the AI resolves it, and the set grows to 17. The range is $0.20 to $2.00 and the median is $0.90 (Siena AI's rate sits exactly in the middle). The nine vendors that charge per attempt have a lower median, $0.50, because they're pricing a cheaper event.

To compute both medians, we took each vendor's headline unit rate as printed on its own pricing page, sorted the 17 values (0.20, 0.40, 0.40, 0.45, 0.49, 0.50, 0.70, 0.75, 0.90, 0.99, 0.99, 1.00, 1.20, 1.20, 1.50, 1.50, 2.00) and took the ninth. For the resolution-only group, the middle two values are $0.99 and $1.00, so the median is their average. No estimate, discount or negotiated rate enters either number. The headline rate is the per-unit figure each vendor prints most prominently, which is the entry plan for Yuma AI and Lorikeet, the overage rate for Alhena AI, the Team pack rate for Botpress and the tier-cap rate for Macha (whose effective rate at 1,000 tickets, $0.60, would rank about seventh). The medians use that headline rate, while the cost tables below use the cheapest billing term; swapping in Tidio's annual $0.58 or Gorgias's annual $0.90 leaves the 17-vendor median at $0.90.

How was this index built?

We checked every vendor's own pricing page, documentation or marketplace listing on 24 September 2026 and copied a verbatim quote for each figure we use. The rules:

• Published list prices only. A price counts when the vendor prints it on its own site, docs or its own marketplace listing. Sales conversations, analyst reports and review sites don't count.
• US dollars. Where a page geo-switched currency (Zendesk served us GBP and Zoho served INR when read from outside the US), we read the US-rendered page or the vendor's docs for the USD figure.
• The cheapest published way to buy. Where a vendor offers monthly and annual rates, the cost tables use the cheaper one and say which it is. Committed-volume rates count when they're printed, as Zendesk's $1.50 is.
• Units are kept apart. A resolution, a conversation, a ticket and a message are different events. The ranked table holds only resolution, conversation, session and ticket pricing. Message, credit and token pricing gets its own table with the conversion assumption written out.
• Estimates are quarantined. For vendors that publish nothing, we show third-party estimates with their source and date in a separate table. They never enter a median, a chart or a ranking.
• Chat and email only. Voice agents are priced separately (Intercom's Fin Voice is $1.99 per voice outcome and Lorikeet's voice resolution is $1.50) and are left out of every table.
• Confidence is labeled. "High" means we read the figure on the vendor's page that day. "Medium" means the vendor's own pages disagree with each other or the figure came from a secondary vendor page; Zendesk, Richpanel and Gladly carry medium in this edition, and the reasons are in the section on contradictions below.

The index doesn't give a like-for-like cost per outcome across units. A vendor billing per conversation charges for every attempt, and a vendor billing per resolution charges only for successes, so 1,000 units means 1,000 attempts for one and 1,000 successes for the other. We kept the units visible in every table, and the section on per-ticket versus per-resolution pricing shows how to convert.

Which AI support agent is cheapest per resolution?

Richpanel prints the lowest unit price, $0.20 per AI-handled conversation on annual billing, though its page labels the same rate as "per resolved conversation" elsewhere, which is why it carries medium confidence. eesel AI and Macha follow at $0.40 per ticket, and both bill whether or not the AI resolves the ticket. The cheapest vendor that bills only for real resolutions is HubSpot's Breeze Customer Agent at $0.45, which needs a $90 Service Hub Professional seat and a $1,500 one-time onboarding fee first.

The table ranks the 17 vendors by published unit price. "AI charges at 1,000" is what the vendor bills for the AI at 1,000 units a month, including any AI platform or plan fee. "Cheapest total at 1,000" adds the minimum prerequisite, usually one help desk seat. Where a standalone agent runs on a help desk you already pay for, we say so instead of guessing your help desk bill.

The arithmetic behind the less obvious cells: Zendesk's total is one $55 seat plus 995 resolutions at $1.50, because the pricing page includes five per agent. HubSpot's total is one $90 seat plus 47,000 purchased credits at $9 per 1,000, after the 3,000 included. Gorgias at 1,000 is the Pro help desk at $300 plus the $171 annual AI bundle for 190 interactions plus 810 more at $1.50. Botpress at 1,000 is the $150 Plus plan with 250 conversations plus eight packs of 100 at $65. Yuma's figure is its $14,400 Starter bundle divided by 12, which Yuma itself calls illustrative. HubSpot credits and Freshworks sessions are sold in whole packs, so volumes that don't divide evenly round up to the next pack.

The native help desk rows include one seat because you can't buy their AI without one. If you already run that help desk, the seat is a cost you're paying anyway, so subtract it. If you don't, the standalone rows (eesel AI, Macha, Siena AI, Lorikeet, Yuma AI) need a help desk added before the comparison is fair.

Is a per-ticket price comparable to a per-resolution price?

Not directly. A per-resolution vendor bills you only for the tickets its AI closes. A per-ticket or per-conversation vendor bills every ticket its AI works on. To compare them, divide the per-ticket price by the share of handled tickets the AI resolves.

At a 50% resolution rate, a $0.40 per-ticket agent works out to $0.80 per resolution, below Intercom's $0.99 outcome. At 30% it's $1.33, above it. So the per-ticket model gets cheaper per outcome the better the agent performs, and the per-resolution model holds its unit price flat regardless.

The two models buy different things. A per-ticket charge covers tickets the AI tags, drafts for or hands off; per-resolution vendors either don't charge for that work or, like Gladly and Intercom, charge for some of it separately. Our longer treatment of the trade-off is in per-action versus per-resolution AI pricing.

How does the cost change from 750 to 10,000 tickets a month?

The table below shows AI charges alone at six monthly volumes, using each vendor's cheapest published rate. It includes AI plan and platform fees (Siena AI's $750, Lorikeet's $2,100, Yuma AI's bundles, Alhena AI's and Botpress's plans) and excludes help desk seats. Units are the vendor's own billable unit, so read each row with its unit from the ranked table.

Flat-rate vendors scale linearly. Zendesk, Gladly, Salesforce, Intercom and Help Scout multiply one rate by volume, so 10,000 resolutions cost ten times 1,000. Zendesk's pay-as-you-go rate of $2.00 would make its 10,000 column $20,000, level with Salesforce.

Platform fees punish small volumes. Lorikeet's $2,100 plan fee makes its effective cost $3.79 per resolution at 750 and $1.20 at 10,000. Siena AI goes from $1.90 per ticket at 750 to $0.98 at 10,000. These vendors are priced for teams at the right-hand side of the table.

Bucket tiers create steps. Macha, Yuma AI and Tidio sell volume in blocks. Macha's price per ticket is $0.40 at each cap (750, 1,500, 3,000, 5,000, 7,500 and 10,000 tickets), but a team using 1,000 tickets pays the 1,500 tier's $599, which is $0.60 per ticket used, and a team using 751 pays $599 for $0.80 per ticket. Yuma AI's 1,500 column is a full Growth bundle ($40,000 a year) because its annual plans pause at the bundle limit instead of billing overage.

The median AI charge at 10,000 units across the 16 vendors with a published price at that volume is $8,625 a month, running from $2,000 at Richpanel to $20,000 at Salesforce.

Which vendors charge whether or not the AI resolves the ticket?

Nine of the 17 published vendors bill per attempt, meaning a conversation, session or ticket the AI touches, resolved or not: Richpanel (on its "AI-handled" label), eesel AI, Macha, Freshworks, Botpress, Tidio, Siena AI (its page doesn't say either way, so we grouped it here), Alhena AI (which bills conversations the AI "resolves or assists") and Salesforce's per-conversation option.

Three of the eight resolution-billed vendors also charge for something short of a clean resolution:

• Gorgias bills its help desk ticket fee on every ticket, including the ones its AI Agent resolves. Its billing documentation says you're "charged both a ticket fee + automation fee if AI Agent responds to a ticket and does not hand over". A handed-over ticket pays the ticket fee only.
• Gladly charges $0.25 per AI Assist, a conversation its AI worked on before handing to a person.
• Intercom counts an outcome when Fin "completes a workflow (Procedure), including handoffs", so a conversation that ends with a human can still be billed at $0.99. Our Fin pricing breakdown walks through the definitions.

The billing unit tells you where the vendor's incentive sits. A per-resolution vendor earns more only when its AI closes tickets, which aligns it with you, but the vendor also writes the definition of "resolved". Help Scout and Intercom both count a customer who doesn't ask for more help after the AI's answer, and Gorgias counts a ticket with no human interaction within 72 hours. A per-attempt vendor earns the same on a ticket its AI fumbles as on one it closes, so the buyer carries the outcome risk and gets a predictable invoice in return. Macha is in the per-attempt group.

What hidden costs sit on top of the per-resolution price?

These are the charges we found on the vendors' own pages on top of the unit price:

• Seats you must buy first. Salesforce Service Cloud Core at $195 a user, Gladly at $120 a team member, HubSpot Service Hub Professional at $90, Richpanel Pro at $69, Zendesk Suite Team at $55, Intercom Essential at $29, Help Scout Standard at $25 and Freshdesk Growth at $19.
• Platform and plan fees. Lorikeet's Start plan is $2,100 a month, Siena AI's platform fee is $750 a month and is charged at zero volume, and Yuma AI's smallest bundle is $14,400 a year paid upfront.
• One-time fees. HubSpot requires a $1,500 Professional Onboarding purchase.
• Overage premiums. Gorgias charges $1.50 per automated interaction past the bundle, 50% above the in-bundle $1.00. Zendesk's pay-as-you-go rate is $2.00 against $1.50 committed. HubSpot's pay-as-you-go credit rate is $0.010 against $0.009 on annual packs.
• Annual lock-in. Richpanel says "All subscriptions are on yearly contracts." Lorikeet and Yuma AI bill annually; Yuma AI's plans carry a 12-month commitment.
• Expiring units. Freshworks session packs expire with the billing cycle, and Botpress conversation packs expire monthly.
• Model usage. Botpress includes $0.10 of LLM usage per conversation and bills AI spend above that in $10 top-ups sold at cost.
• Double counting. Gorgias's ticket fee on AI-resolved tickets, described above.

Macha has no seat, platform or onboarding fee, and setup and monitoring by the Macha team are included in the tier price. It also publishes no top-up price: when a month's tickets run out, agents pause until the next billing date unless you move up a tier or buy a top-up, and the top-up rate isn't on the pricing page.

How do message- and credit-priced AI agents compare?

Four vendors price a smaller unit than a conversation: an AI reply, a credit or a block of LLM tokens. Their unit prices look tiny next to the ranked table because they aren't the same unit, so we never put them in it. To make them roughly comparable we had to assume how many units one customer conversation uses. Each row states its assumption.

Chatbase's model multiplier moves the bill most. The same 1,000 three-reply conversations cost $150 on a 1-credit model and $350 on Claude Sonnet at 3 credits a reply, and Chatbase lists its top models at 6 credits. Copilot Studio's $0.01 is a price per credit; Microsoft's own billing rates page prices a classic answer at 1 credit, a generative answer at 2 and an agent action at 5, so a conversation that looks up an order costs more than one that answers from a help article.

Which AI support vendors don't publish a price?

Eleven vendors in the index publish no usable list price. Everything in the estimate column below comes from a third party, is labeled with its source and date, and is left out of every median, chart and ranking above.

A vendor that prices each deal separately can charge each buyer what that buyer will bear, which is the incentive behind quote-only pricing, and the Vendr ranges above (Ada from $36,132 to $300,000 a year) show how wide that spread gets. Several of these vendors also let the contract define what a billable "resolution" is, so two customers paying the same rate can pay for different things. Forethought is no longer independent. Zendesk completed its acquisition on 26 March 2026, and since 4 June 2026 Zendesk customers can buy "Forethought AI agents by Zendesk" as an add-on, per Zendesk's announcement. Neither announcement prints a price, so Forethought stays in this table and Zendesk's $1.50 automated resolution stays the only Zendesk AI price in the ranked one. Front is in this table because $0.05 is a floor, not a rate you can buy at, and Front doesn't define the conversation it bills.

Where do the vendors' own pages contradict each other?

Four vendors in the ranked table publish figures that disagree across their own pages, or publish them somewhere unusual. We rank them on the figure described below and flag it.

Zendesk (medium confidence). The US pricing page lists automated resolutions at $1.50 committed and $2.00 pay-as-you-go, with 5 included per agent per month on Suite Team. Zendesk's help center, updated for the resolution tiers introduced on 18 May 2026, describes the allowance as a dollar amount instead: "Team $2 per agent seat, per month", up to "$5000" a year, in Managing resolution allowances. The help center's tier article explains which tiers draw on that allowance, and the help center describes its example prices as placeholders. We used the pricing page's $1.50 and five included resolutions. Our Zendesk AI pricing guide tracks the change.

Richpanel (medium confidence). We re-read richpanel.com/pricing on 24 September. The visible page is consistent: Pro at $69 per agent a month "on yearly contracts", and a Customer Service AI card at $0.20 "per AI-handled conversation" that says "You pay only for the conversations they handle." Hidden markup on the same page still carries an older tier list with $59 and $99 seat prices and a line reading "About $0.20 per resolved conversation", and Richpanel's llms.txt describes "about $0.20 per AI-resolved conversation". We ranked Richpanel on the visible handled-conversation label with the $69 seat. If Richpanel confirms it bills only resolutions, it becomes the cheapest resolution-only vendor by a wide margin.

Gladly (medium confidence). Gladly's own website lists no price and asks for a demo. The $1.50 per AI Resolution, $0.25 per AI Assist and $120 per team member come from Gladly Inc's Shopify App Store listing, which is a plan for Shopify merchants. The "$250 / month" on that listing is a spending cap the merchant sets, with usage billed as you go ("unlimited interactions up to your spending cap"), so running 1,000 resolutions means raising the cap to cover the $1,620 total. The free plan's one included seat carries only 100 AI interactions, so our total prices the paid seat at $120. Enterprise Gladly is quote-only, and a Vendr figure for it (median $24,150 a year) belongs with the estimates.

Gorgias (high confidence, unusual billing). The figures are consistent, but the model charges twice for AI-resolved tickets, as covered above. Treat the help desk's ticket fee as part of the AI's cost.

How does AI cost per resolution compare to a human agent?

A human-handled resolution costs several times the AI list prices above. Zendesk's own cost per resolution guide works an example of $250,000 in monthly support costs over 50,000 verified resolutions, which is $5.00 each, and its cost base counts salaries, managers, software, training, overhead and reopened cases. Unthread's compilation of channel benchmarks puts human email support at $8 to $15 a ticket, live chat at $10 to $16 and phone at $17 to $25. Those are third-party benchmarks rather than list prices, so they sit outside the index, but they explain why a $1.50 resolution still sells.

The per-unit price only becomes a budget once the help desk is in it. Take a team of 5 agents on Zendesk Suite Team ($55 a seat on annual billing, $275 a month) handling 3,000 tickets a month, with the AI working every ticket:

At 50%, the two per-ticket agents cost the least. At 30%, Fin's per-outcome price wins by about $300 a month, because the per-ticket agents still bill the 2,100 tickets their AI didn't close. Fin also bills completed Procedures that end in a handoff, so treat its line as a floor. Its standalone plan carries a minimum monthly commitment, which this volume clears, and Zendesk's 25 included resolutions are its five per agent.

Where does Macha sit in this index?

Macha publishes this index and is one of the 32 vendors in it. Its row follows the same rules as every other row, and the figures come from our pricing page as it read on 24 September 2026.

Macha charges per ticket: one thread between Macha and one person is one charge, however many messages it takes, whether or not the AI resolves it. The plan is one tier ladder, billed monthly: $299 for 750 tickets, $599 for 1,500, $1,199 for 3,000, $1,999 for 5,000, $2,999 for 7,500 and $3,999 for 10,000, with custom pricing above that. Each cap works out to $0.40 a ticket. Between caps you pay the next tier up, so the effective price per ticket used is higher. At 1,000 tickets you pay $599, or $0.60 each. Setup and monitoring by the Macha team are included, there are no seat or platform fees, and the trial is $50 of free usage (about 125 tickets), no credit card, no time limit. There's no published top-up price.

At a volume that sits on a cap, Macha's per-ticket rate ties eesel AI's for second-cheapest in the index; between caps it costs more than eesel AI's straight $0.40. Because it bills attempts, its cost per resolution depends on how many tickets the agent closes, per the conversion table above. Macha fits teams already running Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom who want agents working inside the existing ticket queue and want someone else to build and monitor those agents. If you need an AI that bills nothing for failures, a per-resolution vendor from the ranked table is the better fit. The Macha on Zendesk page shows how the agent layer sits on a help desk.

How will this index be updated?

We publish a new edition every quarter. The next one is due in December 2026, and each edition re-reads every vendor page from scratch rather than carrying figures forward. Between editions, we correct a row when a vendor changes a published price, and the change is noted with its date in the row.

Vendors that want a row corrected, or quote-only vendors that decide to publish a rate, can reach us through the contact details on this site; we'll only use a figure the vendor has put on a public page. For the conceptual background on how these models differ, see our guides to AI agent pricing models and the economics of credits versus cost per resolution.

Frequently asked questions

How much does an AI support agent cost per resolution in 2026? The median published price is $0.995 per resolution across the eight vendors that price per resolution, with a range of $0.45 (HubSpot) to $1.50 (Zendesk and Gladly). Across all 17 vendors with a comparable published unit price, including per-conversation and per-ticket pricing, the range is $0.20 to $2.00 and the median is $0.90.

Which AI customer support agent is cheapest? Richpanel prints the lowest unit price at $0.20 per AI-handled conversation on an annual contract, at medium confidence because its page contradicts itself. eesel AI and Macha follow at $0.40 per ticket, though Macha's $0.40 applies at a tier cap and 1,000 tickets work out to $0.60 each on its 1,500-ticket tier. HubSpot's Breeze Customer Agent is the cheapest resolution-only option at $0.45, once you've bought a $90 seat and $1,500 onboarding.

Does Zendesk charge $1.50 or $2.00 per automated resolution? Both. Zendesk's pricing page lists $1.50 per automated resolution when you commit to a volume and $2.00 pay-as-you-go. Its help center describes a newer dollar-based allowance, so we rate the Zendesk figures medium confidence in this edition.

How much does Intercom Fin cost per resolution? Intercom charges $0.99 per Fin outcome, and an outcome includes a completed Procedure even when it ends in a handoff. On Intercom's own help desk you need at least one seat, from $29 a month on annual billing, so 1,000 outcomes cost $1,019 a month.

Is per-resolution pricing cheaper than per-ticket pricing? It depends on your resolution rate. Divide the per-ticket price by the share of handled tickets the AI resolves: $0.40 per ticket is $0.80 per resolution at 50% and $1.33 at 30%. Above roughly a 40% resolution rate, a $0.40 per-ticket agent costs less per resolution than a $0.99 per-resolution one. At $0.60 per ticket, which is Macha's effective rate at 1,000 tickets, the break-even is about 61%.

What does AI support cost at 10,000 units a month? Across the 16 vendors with a published price at that volume, AI charges run from $2,000 (Richpanel) to $20,000 (Salesforce), with a median of $8,625. Zendesk and Gladly land at $15,000 per 10,000 resolutions and Intercom at $9,900 per 10,000 outcomes. Per-attempt vendors bill tickets, so at a 50% resolution rate their cost per resolution doubles: Macha's $3,999 for 10,000 tickets is $0.80 per resolution, or $3,999 for about 5,000 resolutions.

Why don't Ada, Decagon and Sierra publish prices? They sell annual enterprise contracts priced per deal, and several let the contract define what counts as a resolution. Third-party estimates put Ada at a Vendr median of $72,000 a year and Decagon at $400,000, but those are estimates and they aren't ranked here.

How often is the AI Customer Support Pricing Index updated? Every quarter. This is the September 2026 edition, checked on 24 September 2026, and the next edition is due in December 2026.

To price Macha against your own queue, start a trial with $50 of free usage (about 125 tickets), no credit card, no time limit, or compare the tiers on the pricing page.