How AI customer support actually bills in 2026: every vendor's verified rate, the five pricing models, the year's pivots, and the same month costed across all of them.
Key takeaways
- 2026 is the year support AI stopped being priced like software and started being priced like labour: per resolution, per conversation, per task, with the meter running on the AI's success.
- The going rate for an AI-resolved ticket runs from about $0.49 to $2.00 depending on vendor, with $0.99 (Intercom's Fin) as the market's reference price.
- Five pricing models now coexist: flat-included, per-resolution, per-conversation, hybrid seat-plus-meter, and expiring credits. Each fails differently, and the failure is always in the busy month.
- The deepest 2026 shift: one help desk now publishes only its AI meter, the market's largest player retired its legacy bots, and the analysts expect 40% of agentic-AI projects to be cancelled by 2027. The shakeout is the story.
Table of contents
How much does AI customer support cost in 2026? The honest answer is a rate card, not a number: the market prices an AI-resolved ticket anywhere from $0.49 to $2.00, an AI-handled conversation from about $0.28 to a quota tier, and some vendors still include the AI in the seat for a flat $18: that last one is Drag, our shared inbox for Gmail (with its own web, Mac, Windows, iPhone and Android apps), at $12 to $24 a seat all-in. This report compiles every major vendor's verified pricing, first published in July 2026 and updated in November 2026, sorts the five models now competing, and costs the same support month across all of them, so the comparison is finally apples to apples.

Founder’s Take
A personal note from Nick, co-founder of Drag
Outcome is the end game. What we're currently seeing in the market is traditional customer support products iterating to outcome-based products, with few that have actually made the move (most are new startups without the constraints of legacy product).
We're riding the same wave. Varley's Agent is built for outcome-based, and our measurement of success is what percentage we can actually complete without human supervision or dependency.
I've collected the different outcome-based AI support agents in this report. It's quite definitive, and I'll aim to keep the list up to date.
That's the end game.
Methodology
Every figure below was verified directly against the vendor's public pricing page or help center, most recently at publication in July 2026, and each row carries its verification date. Where a vendor's pricing is usage-based, we state the meter and its unit, not a marketing summary of it. We sell one of the products listed (Drag), we say so where it appears, and the table treats us with the same units as everyone else. Corrections are welcome and dated updates will be noted here.
The rate table's data is published under CC BY 4.0: reuse it freely with attribution.
July 2026: initial publication; all rates verified against vendor sources at publication.
November 2026: Fin, Parahelp, eesel, Gorgias, Trengo, Zendesk and Kayako figures updated from the vendors' own pages, read 11 to 14 November 2026, and Salesforce's completed purchase of Fin noted. Lorikeet and Fini rows updated from their own pages, read 16 November 2026.
What AI support actually costs: the verified rates
| Vendor | AI pricing model | The verified rate | What the meter counts | Verified |
|---|---|---|---|---|
| Per resolution | $0.99 per resolution; 50 a month minimum when Fin runs on another help desk | A conversation Fin resolves without a human. Salesforce signed to acquire Fin for ~$3.6B in June 2026 and completed the purchase on September 10, 2026 | Nov 2026 | |
Zendesk | Per resolution + seat add-on | $1.50 committed to $2.00 pay as you go per automated resolution, above 5 included per agent a month (10 on Suite Professional), plus Copilot at $50 per agent per month | Resolutions by its AI agents; Copilot is the human-assist layer | Nov 2026 |
Kayako | Per resolution, help desk by quote | $1.00 per ticket Kay, its AI agent, resolves, with no seat fees for Kay; Kayako One, its help desk, is custom pricing | Tickets Kay resolves without a human; escalations to humans and simple deflections do not count | Nov 2026 |
| Per resolution | $1 per auto-resolved ticket, plus internal AI use at the public token price; no seat or platform fees | Auto-resolved tickets; tickets that end up with your team are not charged | Nov 2026 | |
eesel AI | Expiring credits | $299 a month for 500 credits, then $0.80 per extra credit; no seat or platform fees; unused credits do not roll over | Each ticket or chat the agent handles is 1 credit, however many replies touch it | Nov 2026 |
Gorgias | Per resolution | $0.90 per automated interaction within its AI Agent tiers on annual plans ($1.00 monthly), then $1.50 each past the tier's limit | Resolutions by its AI agent | Nov 2026 |
Freshdesk (Freddy) | Hybrid: sessions + seat add-on | About $0.49 per Freddy session, plus Copilot at $29 per agent per month | Freddy self-service sessions; Copilot per assisted agent | Nov 2026 |
Tidio (Lyro) | Per conversation, quota tiers | Lyro sold as an add-on with monthly conversation allowances; quota tiers step hard with volume | Each conversation Lyro handles | Jul 2026 |
Trengo | Per conversation surcharge | $0.28 (annual) to $0.33 (monthly) per AI-handled conversation on top of the plan, charged from the first conversation | AI-handled conversations, counted in 7-day windows | Nov 2026 |
| Expiring credits | AI runs on credits included in workspace bundles ($45 to $295 flat); unused credits expire monthly | Credits drawn per AI action | Nov 2026 | |
Featurebase | Per resolution | $0.49 per AI resolution, billed on top of the plan price (no separate AI seat fee), the lowest published per-resolution rate | Resolutions by its AI | Jul 2026 |
Salesforce (Agentforce) | Per conversation or per action | $2.00 per conversation, or Flex Credits at about $0.10 per action (introduced after customer pushback) | Conversations, or individual AI actions | Jul 2026 |
| Per conversation | $0.50 per conversation (50 HubSpot Credits at $0.01), billed whether or not the AI resolves it; platform tiers apply beneath | Conversations the agent handles, resolved or not | Jul 2026 | |
Fini | Per resolution | $0.89, $0.69 or $0.49 per resolved ticket, by plan; the 90-day pilot and its 90% guarantee are Enterprise only | Resolutions, with no AI seat fees | Nov 2026 |
Lorikeet | Per resolution, plus a monthly plan | Start $2,100 a month plus $0.99 per resolution (chat, email, SMS); Scale $5,100 a month on annual billing plus $0.90; voice $1.50 and $1.20; no separate platform fee; escalations never bill | Resolved tickets | Nov 2026 |
| Sierra · Decagon · Ada | Outcome-based, unpublished | Per-resolution or outcome contracts negotiated in sales cycles; no public rates | Negotiated outcomes, vendor-defined | Jul 2026 |
Drag | Flat, included | Varley, Drag’s AI, included in the seat from $18 with monthly AI credits (1,000 a seat on Plus, 2,000 on Pro); teams that use more top up | Conversations Varley finishes with no one on your team involved; hand-offs and rejected actions are free | Sep 2026 |
Sortd | Announced | AI features listed on paid tiers with the fuller set marked launching soon; no usage pricing published | n/a | Jul 2026 |
The reference price the market keeps quoting each other is Fin's $0.99; everything else positions against it.
The five ways AI support bills, and how each one fails
Flat-included. The AI is in the seat price (Drag at $18, with monthly AI credits; HubSpot partially, by tier). Fails by ceiling: at very high AI volume, the included amount runs out and top-ups start.
Per-resolution. You pay when the AI succeeds (Fin $0.99, Zendesk $1.50 to $2.00, Gorgias $0.90, Kayako $1.00, Featurebase $0.49). Fails by success: deflection improving from 30% to 60% doubles the AI line, and the invoice punishes exactly what you wanted.
Per-conversation. You pay when the AI participates, resolved or not (Trengo's surcharge, Lyro's quotas, HubSpot's per-conversation credits). Fails by ambiguity: a conversation is a fuzzier unit than a resolution, 7-day windows and quota steps decide your bill, and busy months spike it.
Hybrid seat-plus-meter. A per-agent add-on for the human-assist layer plus usage for the autonomous layer (Zendesk's $50 Copilot plus resolutions; Freshdesk's $29 plus sessions). Fails by opacity: two meters, one invoice, and nobody in the room can predict next quarter's number.
Expiring credits. Usage credits that vanish monthly (Crisp, eesel). Fails by waste: quiet months forfeit what busy months needed.
The pattern across all five: the question is no longer what the software costs, it is who carries the risk of the AI working. Metered models put that risk on the customer; flat models put it on the vendor.
Drag · with Varley
The inbox your team and Varley work in together
Shared inbox and live chat for Gmail teams, with Varley, Drag’s AI, included from $18. Plus an MCP server your own AI tools can drive.
The word "resolution" is doing all the work
A disclosure: since 28 September 2026 Varley’s Agent resolutions draw on the AI credits included in Drag’s seats, so we have a stake in how a resolution is defined. We publish our own definition on the AI credits page (a conversation Varley finishes with no one on your team involved; hand-offs and rejected actions are free) and hold it to the same four lenses as everyone else. The published rates above span $0.49 to $2.00 for what is nominally the same unit, a four-fold spread that tells you the unit is not standardized. Four lenses decide what a rate actually costs:
What counts. The quiet industry default is assumed resolution: if the customer stops replying within some window, the ticket bills as resolved, including the customer who gave up and the one who got a wrong answer. A vendor counting confirmed end-to-end fixes will invoice a fraction of one counting silence, at the same sticker rate.
Who decides. Almost everywhere, the vendor defines the unit and controls the counter, so ask for the definition in writing before comparing any two rates.
Do escalations bill. A conversation the AI touched and then handed to a human is a resolution on some meters and free on others. This single clause can move a bill by half.
Does a return visit bill twice. The customer who comes back tomorrow with the same issue is a new billable resolution on most meters. Independent benchmarking puts repeat-contact rates after AI handling at roughly 2.3x, which means deflection that fails can quietly raise total cost while looking cheap on the dashboard.
The practical rule this section reduces to: never compare sticker rates; compare definitions, and get them in writing. And a growing tier of agent vendors (Sierra, Decagon, Ada) publishes no rates at all, which makes the definition question unanswerable before a sales cycle: factor that cost in too.
What changed in 2026
Kayako put the meter on its front page. By November 2026, Kayako's pricing page published one number: $1 for each ticket Kay, its AI agent, resolves, with no seat fees for Kay. Kayako One, the help desk around it, is custom pricing, quoted by sales. The bet is that AI resolution volume is the headline, with the seat price left to a sales call.
Zendesk turned the meter on. The market's largest player moved its automation to outcome pricing ($1.50 to $2.00 per resolution) while retiring its legacy bots between August and December 2026, forcing thousands of teams to re-decide their AI stack this year, which is precisely why this report exists.
The consolidation accelerated, then detonated. Forethought was acquired by Zendesk in March; Cognigy went to NiCE; and in June, Salesforce signed a definitive agreement to acquire Intercom's Fin, the vendor whose $0.99 rate is this market's reference price, to fold it into Agentforce, and the deal closed on September 10. Independent AI-support pricing is disappearing into the suites, and the reference price itself now has a new owner with a $2.00-per-conversation product to protect.
The analysts called the shakeout. Gartner expects over 40% of agentic AI projects to be cancelled by end of 2027 on cost and unclear value. Read against the table above: the cancellations will concentrate where the meter made value hardest to prove.
One support month, costed across the market
Take a five-person team whose AI handles 500 conversations a month, and hold everything else equal.
| Model | Vendor example | The month's AI bill | Plus |
|---|---|---|---|
| Flat-included | Drag | Included in the seat, with top-ups at high volume | Five Pro seats, $120 |
| Per-resolution, low rate | Featurebase | ~$245 | The plan price on top |
| Per-resolution, reference | ~$495 | Seats from $29 to $132 each on annual billing | |
| Per-resolution, premium | Zendesk | $712.50 to $950 (475 billable after 25 included) | Suite seats from $55; Copilot ($50 per agent) needs Suite Professional at $115 |
| Per-resolution, help desk by quote | Kayako | ~$500 | Kayako One, its help desk, at a custom price |
| Per-conversation | Trengo | ~$140 to $165 surcharge | The $329 Boost plan it sits on |
| Expiring credits | eesel | $299 (500 credits: each ticket or chat is 1 credit) | No seat or platform fee |
Two honest readings. Against a human agent's cost, even the premium meter is cheap: $950 for 500 resolutions undercuts any salary. Against seats that include the AI, every premium meter is expensive: on Drag the AI is included in the seat, with top-ups only at high volume. The model question is really a forecast question: if you believe your AI volume will grow, buy the model whose bill grows slowest with it. Run your own numbers in the cost calculator.
Cost comparator: your month, across the models
| Vendor | Model | This month |
|---|---|---|
| Per resolution | $245 | |
| Per conversation | $250 | |
| Flat-included | $270 | |
| Per session | $390 | |
| Per conversation | €424 | |
| Per resolution | $445 | |
| Per resolution | $450 | |
| Per resolution | $495 | |
| Per resolution | $495 | |
| Per resolution | $500 | |
| Per resolution | $500 | |
| Per conversation | $1,000 | |
| Per resolution | $1,275 |
Estimates only, computed from the verified rates above. A resolution is treated as a conversation for comparability; rates are annual-billing; EUR is shown as published, not converted; minimums are applied (Fin's 50-per-month floor); rows marked * carry a footnote (hover to read). The unpublished tier (Sierra, Decagon, Ada) is not modelled. Confirm current pricing on each vendor's site.
Where this is heading
Four calls for the next 18 months, written down so we can be graded. One: the per-agent AI add-on (the $50 and $29 Copilot class) migrates into base seat prices by the end of 2027, the way mobile apps and APIs did, under pressure from flat-included competitors. Two: per-resolution rates compress toward $0.30 to $0.50 as resolution becomes a commodity and vendors compete on the meter. Three: at least one more major desk follows Kayako in publishing only its AI meter and quoting its seats. Four: Gartner's 40% cancellation call lands hardest on hybrid seat-plus-meter stacks, where cost attribution is most opaque. And the wildcard worth logging: Gartner also projects GenAI cost per resolution to exceed $3 by 2030, surpassing offshore human rates, which would mean the meter's endgame is costing more than the labour it replaced. We will re-verify this table quarterly and mark what moved.
For which agent to actually pick, see our AI agents breakdown; for the reference-price vendor's cheaper rivals, the Fin AI alternatives roundup; and for the term itself, what an AI support agent is. The Kayako and meter-switch stories have their own breakdowns: Kayako, Trengo, Crisp, Intercom, and the wider Zendesk alternatives. Drag's own flat-included pricing is on the pricing page.
Frequently asked questions
How much does AI customer support cost in 2026?
Per AI-resolved ticket: $0.49 to $2.00 depending on vendor, with $0.99 (Intercom's Fin) as the market reference. Per AI conversation: roughly $0.28 to quota tiers. Or seat-included: some platforms include AI in the seat, from $18 per user.
What is per-resolution pricing?
You pay each time the AI fully resolves a conversation without a human: published rates run from $0.49 (Featurebase) through $0.89 (Fini), $0.99 (Fin, and Lorikeet on top of its $2,100 monthly plan), $1.00 (Kayako's Kay, and Parahelp, plus token charges for internal AI use) to $1.50 to $2.00 (Zendesk). Gorgias charges $0.90 per automated interaction within its AI Agent tiers and $1.50 beyond them. The bill scales with the AI's success, and the definition of resolution varies enough to make identical rates cost differently.
What is the cheapest way to add AI to customer support?
For most small teams, a seat with AI included (from $18 per user) is the lowest and most predictable total. The expensive mistake is a premium meter at moderate volume.
Is per-resolution AI pricing worth it?
It is the right risk trade when volume is uncertain: you pay only for outcomes. It becomes the wrong one as volume grows, because success raises the bill. Cost it at your busy month, not your average one.
What happened to Kayako's pricing?
Kayako's pricing page now publishes one number: $1.00 for each ticket Kay, its AI agent, resolves without a human, with no seat fees for Kay. Escalations to humans and simple deflections do not count. Kayako One, its full help desk, is custom pricing, so the seat side of the bill comes from a sales quote.
Why do vendors charge per AI resolution instead of a flat fee?
It aligns the vendor's revenue with delivered outcomes and lowers the entry price. The customer-side cost is unpredictability: the meter runs fastest precisely when the product works best.
What does Drag charge for AI?
Seats, with AI credits included. Varley, Drag’s AI (grounded drafts, automatic tagging, summaries, answers from your docs, and Copilot), is included from $18 per user, with monthly AI credits: 1,000 a seat on Plus and 2,000 on Pro. When Varley resolves a conversation on its own (Pro), it draws on those credits; hand-offs and anything you reject are free. Teams that use more top up, and credits they buy never expire. MCP doesn’t use credits, because your own assistant does the AI work. All of it runs on a full support platform (shared inbox, boards, live chat, WhatsApp, help center, MCP server). Every price is on the AI credits page.
Where is AI support pricing heading?
Toward the meter in the short term and into the seat in the long term: outcome pricing is spreading now, but the add-on layer is already compressing, rates are falling, and the analysts expect a heavy shakeout of metered projects by 2027. The table above gets re-verified quarterly.
Co-founder
Building Drag for nearly ten years: shared inboxes, boards, and now the AI and agent layer, all on Gmail, plus HeyHelp for the personal inbox. Writes the honest versions of the comparisons.



eesel AI



Featurebase
Salesforce (Agentforce)
Fini
Lorikeet
