AI chatbot pricing,
from the vendor that ranked itself fourth.

Not the most advanced AI on this list and not close to the cheapest at scale. It is first because it is the only vendor here whose pricing page answers the question a buyer actually loses sleep over, which is what happens when the bot gets it wrong.

Where Help Scout genuinely wins

The fairest AI meter published by anyone in this field, including us. A conversation the customer escalates is not billed at all, one conversation is charged once however many questions the AI answered inside it, and you can set a hard monthly cap that switches the AI off rather than letting the bill run.

The long version

Every vendor on this page will tell you their AI resolves most of your tickets. Only one of them has written down what happens when it does not, and that is the entire reason Help Scout is at the top of this list rather than somewhere in the middle where its feature set would put it.

Read their billing FAQ and it says, in their own words, that if a customer asks for more help "the conversation won't count as a resolution and you won't be charged". Then it says you are charged once per conversation "even if AI Answers responds to multiple questions". Then it offers a monthly cap that switches the AI off when you hit it. Three sentences, and between them they remove almost every way an AI support bill surprises you.

Compare that with what the rest of the field publishes. Two vendors here bill you the moment the AI opens its mouth, whether or not it helped. One bills you twice when the AI succeeds and once when it fails, so success is the expensive outcome. One does not publish a rate at all. Against that, $0.75 per genuinely resolved conversation with a cap on top is not just a price, it is a position.

The catch is the seat line, and it is a real one. Pro is $75 a user with a floor of ten users, which is $9,000 a year before the AI has answered anything. That floor is not a trick. It is what a helpdesk costs when the plan assumes ten humans read the tickets. If you have three people, Standard is $75 a month and Help Scout is cheap. If you have three people and you were hoping the AI would mean you never need ten, you are paying for an assumption you disagree with.

The other limit is what the AI can read. Help Scout describes its AI knowledge as your knowledge base, web sources and custom instructions. If the answer to your most common question lives in a Confluence page or a SharePoint folder rather than in your public help centre, Help Scout's AI has no published route to it, and the honest version of that sentence is that we could not find one rather than that it definitely does not exist.

Pick it if the thing that worries you about AI support is the bill behaving badly. Nobody else here has made you that promise in writing.

Help Scout vs Outlearn, in full including the rows where they beat us.

Starting pricehelpscout.com/pricing. Standard $25, Plus $45, Pro $75 per user per month at annual billing, which their page discounts by 16 percent. Pro is listed as unlimited users with a minimum of 10, so it bills at least $750 a month.Checked 2026-08-09

$25 per user/mo on Standard, annual billing. Free plan for up to 5 users.

AI meterhelpscout.com/pricing FAQ, verbatim: "If a customer requests more help with natural language or by selecting the 'I still need help' button, the conversation won't count as a resolution and you won't be charged." And: "You'll only be charged for one resolution per conversation, even if AI Answers responds to multiple questions."Checked 2026-08-09

$0.75 per resolution, and only per resolution

Cost per resolved conversationStandard at $25 x 10 seats is $250, plus 1,000 resolutions at $0.75 is $750. On Pro the seat line becomes $750 and the total is $1,500.Checked 2026-08-09

$1.00 at 1,000 resolved conversations a month

Handoffhelpscout.com/pricing, where escalation is defined in the AI Answers billing FAQ as the customer selecting an "I still need help" button.Checked 2026-08-09

Native. The AI sits in front of a shared inbox that the same company builds, so an escalation is a conversation moving down the page rather than an integration.

SAML / SSOhelpscout.com/pricing, Security section, where SSO/SAML is listed as an add-on on Plus and Pro.Checked 2026-08-09

Yes, as a paid add-on on Plus and Pro

Knowledge sources

Their AI reads your Help Scout Docs, web sources and custom instructions. They do not enumerate third-party repositories anywhere we could find, so a policy that lives in Confluence is out of reach unless you copy it in.

Trial

15-day trial, no card. Plus three months of unlimited AI Answers resolutions free.

The most complete product in this comparison and the one with the clearest pricing philosophy. It loses the top slot on one detail buried in its own definition of an outcome, and on the fact that Fin needs Intercom seats underneath it.

Where Intercom Fin genuinely wins

It is the only vendor here that publishes a written definition of the outcome it charges for. Three clauses, on the pricing page, in plain English. Everybody else either bills a unit they define loosely or does not publish a rate.

The long version

Fin is the product the rest of this list is measured against, and Intercom has done the single most useful thing any vendor here has done for buyers: they wrote down what they charge for. Their pricing page defines an outcome in three clauses. The customer confirms the issue is resolved, or they do not ask for more help after Fin responds, or Fin completes a workflow including handoffs. That is more precision than ten of the eleven other vendors on this page offer.

It is also where the interesting problem is. Read the third clause again. A handoff completes a workflow, and a completed workflow is an outcome, and an outcome is $0.99. So on the axis this whole page is organised around, Intercom has built the one meter that charges you when the AI gives up. Not because they are hiding it. It is printed on the pricing page, which is more than most of this field manages, and there is a reasonable argument that routing a conversation correctly is work worth paying for. But if you are budgeting on the assumption that failures are free, that clause is the one that gets you.

The other thing to know is that Fin is not sold alone. It sits on Intercom seats, and the seat you need depends on what else you want. Essential is $29. SSO does not appear until Expert at $132, which for ten people is $15,840 a year before Fin has answered a single question. Teams that need SAML for a security review should price Expert, not Essential, and the honest total is much larger than the headline.

What you get for it is the deepest product here. A real inbox, tickets, a help centre Fin reads directly, workflows, and enough scale that Fin has been pointed at more support volume than anything else on this list. It reads live sources including external help centres rather than files you upload, so it does not go stale the way an upload-based agent does.

At our test volume it lands at $1.28 per resolved conversation, second highest here. That is the price of the platform, not the price of the AI, and for a team that wants one vendor to own the whole customer conversation it is a reasonable trade. For a team that already has somewhere for humans to work and wants an AI layer in front of it, it is a lot of money for the parts you already own.

Pick it if you want one vendor for everything and you would rather pay per outcome than per seat. Read the outcome definition before you model the bill.

Starting priceintercom.com/pricing. Essential $29, Advanced $85, Expert $132 per seat per month. Fin is priced separately as usage.Checked 2026-08-09

$29 per seat/mo on Essential, plus Fin usage on top

AI meterintercom.com/pricing, verbatim: "From $0.99 per Fin outcome", where "An outcome 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".Checked 2026-08-09

From $0.99 per Fin outcome

Cost per resolved conversationEssential at $29 x 10 seats is $290, plus 1,000 outcomes at the published floor of $0.99 is $990. The rate is a floor, so a real quote can be higher.Checked 2026-08-09

$1.28 at 1,000 resolved conversations a month

Handoffintercom.com/pricing, where the outcome definition includes "Fin completes a workflow (Procedure), including handoffs", and the Fin feature list includes "Hands off to agents in preferred Inbox".Checked 2026-08-09

Documented and native: their page says Fin "Hands off to agents in preferred Inbox". The wrinkle is that the same page counts a completed workflow "including handoffs" as a billable outcome.

SAML / SSOintercom.com/pricing, where "SSO & identity management" appears in the Expert plan feature list.Checked 2026-08-09

Yes, on Expert at $132 per seat

Knowledge sources

Reads your Intercom articles, public web content and external help centres, so it is connected rather than uploaded. Scoped to the Intercom world it sits inside.

Trial

14-day trial, no card, and their page says it includes Fin and the add-ons

The cheapest credible AI support agent on this page and the best-documented handoff in the field. It scores third because its knowledge is a copy you upload, which is the failure mode nobody notices until the answers have been wrong for a month.

Where Chatbase genuinely wins

Price, by a distance, and handoff documentation that is better than ours. It is the cheapest serious option in this comparison at every volume we modelled, and its escalate-to-human action is documented properly, including what data moves with the conversation.

The long version

If you want the cheapest way to put a competent AI agent in front of your customers this month, this is it, and no amount of positioning from the rest of us changes that. Four thousand message credits for $120 covers our whole test scenario with nothing to spare and nothing wasted. The next cheapest option here costs 66 percent more, and the median costs eight times as much.

It also does the thing this page is about better than we expected. We went looking for a handoff gap and did not find one. Chatbase ships escalate-to-human as a first-class action, and their documentation spells out what moves with the conversation: chat history, user details, an issue summary, into a connected helpdesk, with the human continuing the thread there. On top of that their own inbox supports agent takeover, where clicking into a live conversation pauses the AI and toggling it resumes. That is a better-documented handoff than most of the incumbents on this list, and it is the reason Chatbase outscores us on that axis rather than the other way round.

The thing that keeps it third is the upload. Chatbase learns from a copy of your knowledge rather than from the knowledge itself. Their docs list four file formats, pdf, txt, doc and docx, and every one of them is a snapshot of something that lives somewhere else and keeps changing without telling anybody. Add the website crawl and you have two ways to teach it, both of which describe your product as it was on the day you ran them.

That matters more than it sounds, because a stale agent does not throw an error. It answers confidently, in your voice, with last quarter's refund policy. Nobody files a bug because nothing looks broken. You find out when a customer quotes your own bot back at you.

The credit meter has the same shape of problem. Credits are spent when the agent replies, not when it resolves anything, so a conversation the agent fumbles for four turns and then escalates costs the same as one it nails. Unused credits do not roll over, and once you are past your plan the auto-recharge rate is $40 per 1,000 credits, which is roughly three times the effective rate inside Standard. That gap is where the angry reviews come from, and it is worth modelling your worst month rather than your average one.

Pick it if price is the constraint and your documentation does not move much. If your documentation moves weekly, you are buying a bill and a stale answer at the same time.

Chatbase vs Outlearn, in full including the rows where they beat us.

Starting pricechatbase.co/pricing. Free 50 credits, Hobby $32/mo with 700 credits, Standard $120/mo with 4,000, Pro $400/mo with 15,000. Extra credits are $40 per 1,000.Checked 2026-08-09

$32/mo on Hobby. Free plan with 50 message credits a month.

AI meterchatbase.co/pricing, where each plan states a monthly message-credit allowance and the auto-recharge add-on is priced at $40 per 1,000 message credits.Checked 2026-08-09

Message credits, spent per AI reply, no rollover

Cost per resolved conversation1,000 resolutions at 4 replies each is 4,000 credits, which fits Standard at $120 exactly. No seat charge. Cheapest total in the comparison.Checked 2026-08-09

$0.12 at 1,000 resolved conversations a month

Handoffchatbase.co/docs/user-guides/chatbot/actions/escalate-to-human, which documents the action fields and states that chat history, user details and an issue summary are transferred to the connected helpdesk.Checked 2026-08-09

Documented properly, and better than most of this field. A first-class escalate-to-human action that creates a ticket in a connected helpdesk with chat history, user details and an issue summary, plus agent takeover inside their own inbox where clicking in pauses the AI.

SAML / SSOchatbase.co/pricing, where SSO appears in the Enterprise column under "Everything in Pro +".Checked 2026-08-09

Yes, on Enterprise only

Knowledge sources

Website crawl plus uploads. Their own documentation lists four file formats: pdf, txt, doc, docx. Every one is a copy of something that lives somewhere else.

Trial

7-day trial on Hobby, Standard and Pro. Free plan with 50 credits a month.

The best answer on this page if your blocker is a security questionnaire rather than an AI capability. The AI itself is an add-on to a shared inbox, and the seat cliff at eleven people is brutal.

Where Front genuinely wins

Enterprise plumbing on a mid-tier plan. SSO and SCIM are both on Professional rather than parked behind a custom quote, and their AI conversation rate of five cents is the lowest per-conversation number published by anyone here.

The long version

Front is on this list for one reason that has nothing to do with artificial intelligence. It is the only vendor here that puts SSO and SCIM on a mid-tier plan you can buy with a credit card. Everybody else either hides SAML behind Enterprise, charges extra for it as an add-on, or does not publish it at all. If your AI project is currently stuck in a security review, Front unsticks it for $65 a seat.

The AI is genuinely cheap too, and cheap in an unusual way. Autopilot starts at five cents a conversation, which is the lowest per-conversation number published by anyone in this comparison. At our test volume that is $99 of AI on top of $250 of seats. The whole scenario lands at $349, fifth cheapest here, and most of that is the inbox rather than the agent.

The catch is the shape of the seat line. Starter is capped at ten seats in Front's own words. Our test scenario has exactly ten people, which is the last comfortable position on their pricing page. Hire one more and the team moves to Professional at $65 each, so the seat line goes from $250 to $715 for one hire. Front is not being sneaky about this, the caps are printed on the page, but it is the single most expensive eleventh employee in this comparison.

Two things are missing that matter for this page specifically. First, the meter is per conversation rather than per resolution, so a conversation Autopilot cannot answer still bills. At five cents that is a rounding error rather than a scandal, and it is worth saying that a two-cent difference in philosophy costs less here than a dollar of it costs elsewhere. Second, and less forgivable, we could not find anything on Front's public pages about what Autopilot does when it gives up. No documented escalation, no statement about transcript continuity, nothing on the pricing page. Front owns the inbox the conversation would land in, so it is very likely fine. It is just not written down, and this page scores what is written down.

The AI add-ons also multiply. Autopilot is usage. Copilot is $20 a seat. Smart QA is another $20 a seat. Smart CSAT is $10. Four meters is a lot of arithmetic for a buyer who wanted one number, and the total for a team that turns everything on is a long way above the headline.

Pick it if the inbox is the product you actually want and SSO is the box you actually need ticked. Do not pick it expecting the AI to be the reason.

Front vs Outlearn, in full including the rows where they beat us.

Starting pricefront.com/pricing, verbatim: Starter "$25/seat/mo, up to 10 seats", Professional "$65/seat/mo, up to 50 seats", Enterprise "$105/seat/mo". Annual commit; the page publishes no monthly price.Checked 2026-08-09

$25 per seat/mo on Starter, capped at 10 seats

AI meterfront.com/pricing, verbatim: Autopilot "Starting at $0.05 /conversation". Copilot and Smart QA are $20 per seat per month each, Smart CSAT is $10.Checked 2026-08-09

Autopilot from $0.05 per conversation, not per resolution

Cost per resolved conversationStarter at $25 x 10 seats is $250, exactly at their ten-seat cap. 1,984 conversations reach the AI to produce 1,000 resolutions, at $0.05 each that is $99.20. Add one person and Professional makes the seat line $715.Checked 2026-08-09

$0.35 at 1,000 resolved conversations a month

Handofffront.com/pricing, which prices Autopilot per conversation and makes no statement about escalation, transfer or handoff. Marked not published rather than absent.Checked 2026-08-09

Not published. Front's pricing page says nothing about what Autopilot does when it cannot answer, and we could not find a documented handoff mechanism on their public pages.

SAML / SSOfront.com/pricing, where Professional lists "SSO, SCIM" and Enterprise lists "SSO, SCIM, and custom roles".Checked 2026-08-09

Yes, SSO and SCIM on Professional at $65 per seat

Knowledge sources

Front's AI works over your inbox and your knowledge base inside Front. No published connector list for third-party knowledge repositories.

Trial

14-day trial with all Professional features

A capable agent with an honest meter, sold as a second subscription beside a second product. The arithmetic of running both is the problem, not the software.

Where ChatBot.com by LiveChat genuinely wins

Publishing an exclusion list. They are one of only two vendors here who state in writing which outcomes do not count as a billable resolution, and they encode their entry price in the page title so a buyer sees it in search results.

The long version

ChatBot.com does something almost nobody in this field does, which is publish a list of the outcomes it will not charge you for. Their documentation states that a resolution does not count when the answer was irrelevant, incomplete, or transferred to a human. Read that last word again. On the axis this page is organised around, they and Help Scout are the only two vendors who have written down that a handoff is free.

They also put the price in the title tag, so a buyer searching sees nineteen dollars before they click. That is a small thing and it is more transparency than eight of the eleven other vendors here manage.

The problem is the shape of the purchase. ChatBot.com is the agent. The inbox it hands off to is LiveChat, a different product on a different subscription from the same company. So a buyer who wants an AI agent that escalates to humans is buying two things, and the resolution allowance on the agent is small. Growth is $79 and includes 200 resolutions a month. Our test scenario needs 1,000. The other 800 arrive in auto-refill packs of fifty at $49.50, which works out at ninety-nine cents each and is the highest incremental rate published by anyone on this page.

Run the arithmetic and the agent alone costs $871 for the month, of which $792 is overage. Add the LiveChat inbox at $49 a seat and ten people take the total past $1,300. The software is not the issue. The packaging is: a 200-resolution allowance on a plan aimed at growing teams means most buyers live permanently in overage, and overage is where their rate is worst.

The knowledge story is the standard one. Website crawl, uploaded articles, their own help-centre content. No published connectors into the systems where internal answers actually live, so the same staleness problem applies here as everywhere else in the builder half of this list.

SSO exists and is Enterprise-only, described on their page as custom SSO via SAML 2.0. That is a real answer, unlike four vendors here who do not mention SAML at all, but it is a sales conversation rather than a checkbox.

Pick it if you already run LiveChat, in which case half the bill is sunk and the handoff is native. If you do not, you are buying two subscriptions to get one product.

ChatBot.com by LiveChat vs Outlearn, in full including the rows where they beat us.

Starting pricechatbot.com/pricing. Essential "$25 per user / mo" monthly and "$19 per user / mo" annual with "10 resolutions per month". Growth "$99" monthly, "$79" annual, with "200 resolutions per month". Extra resolutions are "$49.50 per 50 AI resolutions package (auto-refill)".Checked 2026-08-09

$19 per user/mo on Essential at annual billing, or $25 monthly

AI meterchatbot.com/pricing for the resolution allowances and the $49.50 per 50 auto-refill pack. The exclusion list, which states they do not bill a resolution where the answer was irrelevant, incomplete or transferred, is in their own documentation.Checked 7 August 2026

Per AI resolution, with an exclusion list published in their docs

Cost per resolved conversationOne ChatBot Growth seat at $79 annual includes 200 resolutions. The other 800 arrive in auto-refill packs of 50 at $49.50, which is 16 packs at $792. Total $871 for the agent alone. Add LiveChat at $49 a seat if you want their inbox for the handoff, which for ten people is another $490.Checked 2026-08-09

$0.87 at 1,000 resolved conversations a month

HandoffChatBot.com's own documentation on what counts as a resolution, which excludes transferred conversations. The receiving inbox is LiveChat, priced separately at livechat.com/pricing.Checked 7 August 2026

Real, and their meter respects it: a conversation transferred to a human is on the published list of things that do not count as a billable resolution. The human side of it is LiveChat, which is a separate subscription.

SAML / SSOchatbot.com/pricing, where Enterprise lists "Custom SSO" described as "Integrate Text app with your enterprise or cloud SSO via SAML 2.0".Checked 2026-08-09

Yes, custom SSO on Enterprise only, via SAML 2.0

Knowledge sources

Website crawl, uploaded articles and their own help-centre sources. No published third-party connector list.

Trial

14-day trial, no card required

The best value in the field for a small team and the vaguest AI story in it. You buy dollars of AI credits and their page translates them into an approximate number of conversations, which is where the precision stops.

Where Crisp genuinely wins

Flat pricing per workspace with no seat charge at any tier. For a team of fifteen on a budget, nothing else in this comparison comes close, and their free plan is genuinely free forever rather than a trial in disguise.

The long version

Crisp does the one thing that makes a support tool cheap, which is to stop charging for people. Forty-five dollars a month for a workspace. Ninety-five for the next tier. Two hundred and ninety-five for the top self-serve tier. Add as many agents as you like and none of those numbers move. In a comparison where two vendors charge more than $750 a month in seats alone before their AI has said anything, that is a genuinely different product decision and small teams should take it seriously.

The free plan deserves a note too, because in this field a free plan is usually a trial wearing a disguise. Crisp's says forever on their own pricing page, and the paid trial is fourteen days of full access to any plan, which means you can test Plus and then buy Mini. Nobody else here lets you do that.

Where it falls apart for the purposes of this page is precision. Crisp sells AI as bundled credits denominated in dollars. Mini includes five dollars of credits, which their page says is around ninety automated conversations. Plus includes seventy-five dollars, which is around 1,350. Divide it out and you get roughly 5.6 cents a conversation, which is a good rate, but the word doing the work in both sentences is around. If your volume lands above the bundle you cannot compute your own overage from the page, and above Plus there is no published tier at all.

The other gap is everything to do with failure and security. Their pricing page does not mention SSO or SAML on any tier, including Enterprise. It does not mention handoff, escalation or transfer either. Crisp runs its own inbox so a conversation almost certainly does reach a human, and we would be surprised if a company of their size has no SAML story. But this page scores what a buyer can read, and on both counts a buyer reads nothing.

The knowledge story is the conventional one for this category: their own help centre plus your website. If your answers live in Confluence or a Drive folder, that is a copy-and-paste job, and the copy goes stale the way copies do.

At our test volume Crisp lands at $330.50, which is fourth cheapest, and unlike the seat-priced options that number is the same whether you have four agents or forty. For a fifteen-person team it is the cheapest real option on this page.

Pick it if flat pricing is the thing you want and your AI needs are ordinary. Do not pick it if you need to model the bill precisely or answer a security questionnaire this month.

Starting pricecrisp.chat/en/pricing. Free $0 "Forever", Mini "$45 / Per month, per workspace", Essentials $95, Plus $295. Enterprise is "Unique pricing".Checked 2026-08-09

$45/mo on Mini, per workspace, not per seat. Free plan forever.

AI metercrisp.chat/en/pricing, where Mini includes "$5 included AI credits" and "Around 90 automated conversations", Essentials $25 and around 450, Plus $75 and around 1350. The implied rate is roughly $0.056 per automated conversation.Checked 2026-08-09

Bundled AI credits stated in dollars, translated to an approximate conversation count

Cost per resolved conversationPlus at $295 flat includes around 1,350 automated conversations. 1,984 conversations reach the AI, so 634 are over, at their implied $0.056 that is $35.50. Total $330.50 for any number of people.Checked 2026-08-09

$0.33 at 1,000 resolved conversations a month

Handoffcrisp.chat/en/pricing, which does not mention handoff, escalation or transfer on any tier.Checked 2026-08-09

Not published. Their pricing page says nothing about what the AI does when it cannot answer. Crisp has its own inbox, so the mechanism almost certainly exists and is simply undocumented on the page a buyer reads.

SAML / SSOcrisp.chat/en/pricing, which makes no mention of SSO or SAML on any tier including Enterprise. Marked not published rather than absent.Checked 2026-08-09

Not published on their pricing page

Knowledge sources

Their own help centre plus website content. No published third-party connector list.

Trial

14 days of full access on any plan, plus a free plan with no AI credits

The cheapest complete helpdesk here and the one whose AI is hardest to reason about. Nothing about the AI's behaviour, its knowledge sources or its handoff is on the pricing page, and the page will not even show you a price in dollars.

Where Zoho Desk genuinely wins

Not metering the AI at all at ordinary support volume. Their Express tier documents up to 30 million free AI tokens a month, which at four exchanges per answer covers thousands of conversations, and the whole helpdesk for ten people costs less than one competitor's AI overage.

The long version

Zoho Desk is the answer to a question the rest of this page keeps dodging: what if the AI just came with the software. Their Express tier documents up to thirty million free AI tokens a month. At roughly twelve thousand tokens for a retrieval-augmented answer with a few exchanges in it, that covers something like two and a half thousand resolved conversations before the meter starts. Our whole test scenario fits inside the free allowance, which makes Zoho the only vendor in this comparison whose AI line is a zero.

The seats are cheap too. Seven dollars on Express, twenty-three on Professional, which is where the AI capabilities actually live. Ten people on Professional is $230 a month for a complete helpdesk with tickets, queues, assignment and SLAs. Two vendors on this page charge more than that in AI overage alone.

So why is it ninth. Because almost nothing about the AI is knowable from the outside. Start with the pricing page itself, which geolocates: we asked for it from a US context and it served Indian rupees, and the only way we got dollar figures was by reading Zoho's own underlying price file. A pricing page that will not reliably show a buyer a price in their own currency is a strange place to begin a purchase.

Then the unit. Tokens are the correct engineering unit and the wrong buying unit. Nobody knows how many tokens their support conversations consume, and the conversion depends on the model, the retrieval depth and the length of the answer. We used twelve thousand per resolution and said so, but that is our assumption applied to their allowance, not a number either of us can hold the other to. Every other vendor here bills something a buyer can count.

Then the omissions. Their pricing page does not mention SSO or SAML on any tier. Zoho ships SAML elsewhere in its product suite, so the honest phrasing is that this is missing from the page rather than missing from the product, but a buyer reading the page learns nothing. The same is true of handoff: no mention of escalation or transfer in any AI feature description, on a product that is a full ticketing system and therefore obviously has somewhere for the conversation to go.

The knowledge story is narrow. Its AI reads your Zoho knowledge base, and on the cheapest tier generative AI is bring-your-own-key, which means you supply the model and pay the model provider directly. That is a defensible design and it is another number that does not appear on the page.

Pick it if the budget is the binding constraint and you are willing to do the discovery yourself. Everything good about it is real and almost none of it is written down.

Zoho Desk vs Outlearn, in full including the rows where they beat us.

Starting priceZoho's own USD price file, desk-pricing-val.json, read because zoho.com/desk/pricing.html geolocates the currency and served us Indian rupees. Express $7, Standard $14, Professional $23, Enterprise $40 per user per month billed annually.Checked 7 August 2026

$7 per user/mo on Express, billed annually. Free plan for three users.

AI meterzoho.com/desk/pricing.html, where the Express tier states AI Agents "includes access to pre-built agents and up to 30M free tokens/month", and where generative AI is otherwise bring-your-own-key.Checked 2026-08-09

LLM tokens, with up to 30 million free per month

Cost per resolved conversationProfessional at $23 x 10 seats is $230. At roughly 12,000 tokens per resolved conversation, 1,000 resolutions is about 12 million tokens, inside the free 30 million allowance, so the AI line is zero. Third cheapest total here and the only one where the AI itself costs nothing.Checked 7 August 2026

$0.23 at 1,000 resolved conversations a month

Handoffzoho.com/desk/pricing.html, which does not mention handoff, escalation or transfer in its AI feature descriptions.Checked 2026-08-09

Not published. Their pricing page says nothing about what the AI does when it cannot answer. Zoho Desk is a full ticketing system, so the destination exists. The mechanism is undocumented on the page.

SAML / SSOzoho.com/desk/pricing.html, which makes no mention of SSO or SAML on any tier. Zoho ships SAML elsewhere in its suite, so this is an absence on the page a buyer reads rather than a confirmed absence in the product.Checked 2026-08-09

Not published on their pricing page

Knowledge sources

Your Zoho knowledge base, and little else that they publish. Generative AI is bring-your-own-key on the cheap tier.

Trial

15-day trial of any plan, no card. Free forever plan for three users.

The right pricing philosophy with the price left out. Zendesk states that you pay only for resolutions and then does not publish the rate, which makes it the one vendor on this page whose bill a buyer cannot estimate at all.

Where Zendesk AI agents genuinely wins

Ecosystem and stated philosophy. Nothing else here integrates with as much of your stack, and their pricing page commits in writing to charging only for requests the AI actually resolved.

The long version

Zendesk gets the philosophy right and then withholds the number. Their pricing page says, in their own words, that paying per automated resolution means you pay only for customer requests that were successfully resolved by the AI agent. That is the correct answer to the question this page keeps asking, and it puts Zendesk in a small group with Help Scout and Intercom.

Then you go looking for the rate and it is not there. Not in the served HTML of the pricing page, not in a footnote, not in a comparison table. We could have taken a number from a third-party roundup and put it in our table. We are not going to, because a guessed rate is worse than an empty cell, and because every other figure on this page came from the vendor's own page and that rule does not get an exception for the biggest company in the comparison.

So Zendesk is the one row in our cost table we could not complete. Seats are computable: Suite Team at $55 a seat paid yearly is $550 a month for ten people. The AI line is unknowable from public information. If you are buying Zendesk AI, the first thing to ask on the call is the per-resolution rate and the second is whether it changes at volume, because you cannot model this purchase without both.

What the money buys is the ecosystem, and it is not a small thing. Zendesk integrates with more of a real company's stack than anything else on this page, its AI reads the help centre live rather than from uploads, and it is where a very large share of support teams already work. Handoff is barely a feature question for them because the conversation is already in the right building. It is worth noting that their pricing page does not describe the escalation behaviour either, which is a documentation gap rather than a product one.

The seat line is the second highest here and there is no published monthly option for the Suite tiers, so a team that wants to try this for a quarter is signing an annual commitment or negotiating. The top tier with Copilot is talk-to-sales only.

Pick it if you already run Zendesk, or if your requirement list is long enough that only the incumbent satisfies it. Get the resolution rate in writing before you sign anything.

Starting pricezendesk.com/pricing, verbatim: "Suite Team: $55agent/month paid yearly", "Suite Professional: $115agent/month paid yearly", "Suite Enterprise + Copilot: Talk to Sales". No monthly prices published for these tiers.Checked 2026-08-09

$55 per agent/mo on Suite Team, paid yearly

AI meterzendesk.com/pricing, verbatim: "Paying per automated resolution means that you pay only for customer requests that were successfully resolved by the AI agent". The per-resolution dollar figure does not appear in the served HTML of that page.Checked 2026-08-09

Per automated resolution. The rate is not published.

Cost per resolved conversationSuite Team at $55 x 10 seats is $550. The AI line cannot be computed, because the per-automated-resolution rate is not published on their pricing page and we will not guess it. This is the only row in the table we could not complete.Checked 2026-08-09

Not computable. They do not publish a per-resolution rate.

Handoffzendesk.com/pricing, which describes resolution but not escalation. Marked not published rather than absent, since Zendesk's agent workspace is where a handed-off conversation would land.Checked 2026-08-09

The destination is the best in the category, because Zendesk is the place most support teams already work. The AI agent's escalation behaviour is not described on the pricing page.

SAML / SSOzendesk.com/pricing, which does not mention SSO or SAML on any tier. Zendesk documents SAML elsewhere, so this is an absence on the pricing page rather than a confirmed absence in the product.Checked 2026-08-09

Not stated on their pricing page

Knowledge sources

Your Zendesk help centre and connected content, read live. The deepest integration catalogue in the category by a wide margin.

Trial

14-day trial

The clearest definition of a billable unit in this comparison, attached to the least buyer-friendly definition possible. Lyro bills when it replies, not when it resolves.

Where Tidio Lyro genuinely wins

The cheapest honest way to find out. Tidio publishes exactly what triggers a billable Lyro conversation in one sentence, their free plan carries 50 billable conversations a month, and $32.50 buys 50 Lyro conversations without a sales call.

The long version

Tidio wrote the best sentence on any pricing page in this comparison, and it is the sentence that costs them the ranking. Here it is verbatim: a Lyro conversation is a customer interaction initiated through any channel integrated with Tidio that has at least one reply from our AI Agent.

That is exemplary disclosure. There is no ambiguity, no asterisk, no defined term pointing at another defined term. You know precisely what you are buying. What you are buying is a meter that fires when the AI speaks rather than when the AI helps, which is the least buyer-friendly unit available, published with total clarity. Compare it with Help Scout, whose meter fires only on a genuine resolution, or Intercom, whose fires on an outcome. Tidio's fires on an attempt.

It matters most exactly where you would least want it to. Their pricing page lists human handoff as a Lyro feature. Follow their own definition through and a handed-off conversation is a billable one, because Lyro will have replied at least once before it worked out that it could not help. So the conversations where the AI adds least are the ones you pay full price for.

The other structural issue is the ceiling. Their published Lyro ladder runs from 50 conversations at $32.50 up to 1,000 at $700, and stops. Our test scenario needs 1,984 conversations to reach the AI to produce 1,000 resolutions, which is past the end of the page. Extrapolating at their own floor rate of seventy cents puts the month at roughly $1,413, the highest figure in this comparison, and we want to be clear that this is an extrapolation from their published floor rather than a quote. The real number comes from their sales team.

None of which makes it a bad product for the buyer it is built for. A small Shopify store with fifty support conversations a month gets a free plan that covers the whole volume, ten seats included, a seven-day trial of everything, and an AI agent that understands a product catalogue. At that size Tidio is one of the easiest yes decisions on this page. It is the scaling curve that turns against you, and it turns exactly where their published pricing stops.

SSO is on Premium, which has no published price, so a security review means a sales call.

Pick it to find out cheaply whether AI support works for your shop. Re-price everything before you cross a thousand conversations a month.

Tidio Lyro vs Outlearn, in full including the rows where they beat us.

Starting pricetidio.com/pricing, annual billing. Free $0 with 50 billable conversations, Starter "$24.17/mo." with "100 Billable conversations", Growth "Starts at $49.17/mo." from 250, Plus "Starts at $300/mo. + monthly usage". All non-custom tiers include 10 seats.Checked 2026-08-09

$24.17/mo on Starter at annual billing, 10 seats included. Free plan available.

AI metertidio.com/pricing, verbatim: "A Lyro conversation is a customer interaction initiated through any channel integrated with Tidio that has at least one reply from our AI Agent." Lyro pricing "Starts at $32.50/mo. From 50 Lyro AI conversations".Checked 2026-08-09

Per Lyro conversation, triggered by at least one AI reply

Cost per resolved conversationStarter at $24.17 includes the seats. 1,984 conversations get at least one Lyro reply, and every one of those is billable. Their published ladder stops at 1,000 conversations for $700, which is $0.70 each, so 1,984 at their own floor rate is $1,388.80. Above 1,000 the ladder is sales-gated and this figure is an extrapolation at their published floor, not a quote.Checked 2026-08-09

$1.41 at 1,000 resolved conversations a month

Handofftidio.com/pricing, where "Human handoff" appears in the Lyro AI Agent feature list, and where a billable Lyro conversation is defined as one with "at least one reply from our AI Agent".Checked 2026-08-09

Listed as a Lyro feature in plain words on the pricing page. Whether a handoff is billed is not stated, and given their own definition of a billable conversation it almost certainly is, because Lyro will have replied at least once before deciding to hand over.

SAML / SSOtidio.com/pricing, where SSO appears in the Premium tier feature list and nowhere else.Checked 2026-08-09

Yes, on Premium only, which is contact-for-pricing

Knowledge sources

Their own help centre, uploaded content and website crawl, with ecommerce catalogue awareness for Shopify stores.

Trial

7-day trial of all features and plans, no card. Free plan with 50 conversations a month.

The best ecommerce AI agent here and the worst meter design in the comparison. Their own documentation confirms that a conversation the AI fully resolves bills twice, while one it hands to a human bills once.

Where Gorgias genuinely wins

Doing things rather than saying things, for ecommerce specifically. Its AI agent is wired into the commerce stack well enough to action a refund or an address change, which is a different job from answering a question, and nothing else on this page does it as natively.

The long version

Gorgias is the one vendor on this page that is genuinely good at something none of the others attempt. If you run a Shopify store, most of your support volume is not a question, it is a task: where is my order, change this address, refund this item, pause this subscription. Answering those needs commerce data and permission to act, and Gorgias is wired into that stack more natively than anything else here. On that job it beats us and it beats Intercom.

Then you read their billing documentation and the meter is the strangest design in the comparison. Two sentences, consecutive, from their own docs. You are charged both a ticket fee and an automation fee if the AI Agent responds to a ticket and does not hand over to a human. You are charged only a ticket fee if the AI Agent responds and does hand over.

Read that as a buyer. The outcome you want, the AI resolving the conversation entirely on its own, costs two units. The outcome you were trying to avoid, the AI giving up and passing it to a person, costs one. Success is priced above failure. We have looked at a lot of pricing pages for this page and that is the only meter in the field that inverts the incentive that way.

There is an argument for it, and it is worth stating fairly: the automation fee is what you pay for the automation working, and the ticket fee is what you pay for the ticket existing, and both things did happen. But the effect on a buyer's model is that improving your resolution rate increases your bill twice over, which is the opposite of every other vendor on this page.

It compounds because their definition of a billable ticket is broad. Their docs say a ticket is billable when at least one message in it is sent from your Gorgias helpdesk, whether from a human, from the AI Agent, or from an automatic rule reply. An autoresponder makes a ticket billable.

Run our scenario and it is $1,686, second highest here, and the interesting part of that number is that $1,215 of it is automated-interaction overage at $1.50 a unit. A rational buyer stays on Pro and eats the overage rather than upgrading to Advanced, because Advanced costs more in total, which is a sign the tiers are not shaped around how the product is actually used.

One process note. Their plan cards did not render into the served HTML when we checked on 9 August 2026, so the tier prices above carry our earlier verification date of 7 August. The billing rules come from their documentation and were re-read on the 9th.

Pick it if you sell on Shopify and your AI needs to act rather than answer. Model the double meter at your real resolution rate before you commit.

Gorgias vs Outlearn, in full including the rows where they beat us.

Starting pricegorgias.com/pricing, read 7 August 2026, when the plan cards reconciled to helpdesk plus included AI interactions at $0.90 each: Starter $40, Basic $77, Pro $471, Advanced $1,227 at annual billing. On 9 August 2026 the plan cards did not render into the served HTML at all, so the tier figures below carry the earlier date.Checked 7 August 2026

Helpdesk from $50/mo with the AI Agent bundled into the card price

AI meterdocs.gorgias.com, "How you're billed for using Gorgias", verbatim: "You are charged both a ticket fee + automation fee if AI Agent responds to a ticket and does not hand over the conversation to a human agent" and "You are charged only a ticket fee if AI Agent responds to a ticket and does hand over the conversation to a human agent".Checked 2026-08-09

Two meters. A ticket fee, plus an automation fee when the AI resolves without a human.

Cost per resolved conversation1,984 conversations reach the AI and every one is a billable ticket. Staying on Pro at $471 with 2,000 tickets and 190 included interactions, then paying 810 automated interactions of overage at $1.50, is $1,686. Advanced at $1,227 plus 470 of overage is $1,932, so a rational buyer stays on Pro and pays the overage. Tier figures verified 7 August 2026.Checked 7 August 2026

$1.69 at 1,000 resolved conversations a month

Handoffdocs.gorgias.com, "How you're billed for using Gorgias", which distinguishes the two cases in consecutive sentences.Checked 2026-08-09

Documented, and the meter rewards it: their docs state you are charged only a ticket fee when the AI Agent hands over, versus both fees when it does not. Handing off is literally the cheaper outcome.

SAML / SSOgorgias.com/pricing, which does not mention SSO or SAML. Marked not published rather than absent.Checked 2026-08-09

Not published on their pricing page

Knowledge sources

Your help centre and your commerce data. The commerce integration is the point: order status, subscriptions, refunds.

Trial

Free signup. Their pricing page does not state a trial length.

The clearest number and the weakest deal behind it. A dollar a ticket sounds like outcome pricing and is not: it bills the tickets that reach it, including the ones your team ends up answering, behind a 250-ticket floor and a $3,000 annual minimum.

Where Helply genuinely wins

The simplest price in the entire comparison. One dollar a ticket, no seat fees, unlimited agents. Nobody else here can state their price in four words, and for a buyer exhausted by credit tables that is worth something real.

The long version

Helply has the best price sentence on this page and we want to give it proper credit before taking it apart. One dollar a ticket. No seat fees. Unlimited agents. That is the whole pricing page, and after an afternoon of credit tables, token allowances, seat cliffs and bundled dollars of AI, reading it is a physical relief. Simplicity in pricing is a real feature and almost nobody in this category is willing to pay for it.

The problem is what the unit counts. A dollar a ticket reads like outcome pricing, and it is not. It is arrival pricing. A ticket that reaches Helply bills a dollar, and their page states no exclusion for tickets the AI does not resolve. So at a 50 percent auto-solve rate you are paying two dollars for every conversation the AI actually handles, and the other dollar buys a ticket your own team answered.

Run our scenario. To get 1,000 AI-resolved conversations at a 50.4 percent rate, 1,984 tickets have to reach the system. At a dollar each that is $1,984, the highest computable figure in this comparison, and the shape of that number is the point: you cannot reduce it by improving anything except your total ticket volume.

Then the commitment. A 250-ticket monthly floor means a small team pays $250 whether or not it has 250 tickets. A $3,000 minimum annual contract means you are signing for a year. Every other vendor on this page will take a credit card for a month, and eight of them will let you start free. Helply is the only one asking for a contract before you have measured anything.

What is not on the page matters too. Nothing about SSO. Nothing about handoff, which for a product whose whole premise is that AI handles the easy volume and humans handle the rest is a strange omission, because handoff is the seam the entire product sits on. Nothing about a trial. Their knowledge story is a self-improving knowledge base rather than a published connector list, so it is hard to know from outside what it can actually read.

To be fair to them, the seat position is genuinely good and genuinely aligned with where this category is going. Charging nothing for people is correct and we do the same. It is the ticket meter underneath it that we would not sign.

Pick it if you want one number and your ticket volume is both large and predictable. Do not pick it before you have measured your own auto-solve rate somewhere that will let you measure it for free.

Helply vs Outlearn, in full including the rows where they beat us.

Starting pricehelply.com/pricing, verbatim: "$1/ ticket", "Seat fees None", "Unlimited agents, unlimited seats, AI included.", "Minimum 250 tickets / month", "Minimum annual contract $3,000".Checked 2026-08-09

$1 per ticket. No seat fees. Minimum 250 tickets a month.

AI meterhelply.com/pricing, which prices per ticket and states no exclusion for tickets the AI does not resolve. Whether unresolved tickets are billed is not stated on the page.Checked 2026-08-09

Per ticket that reaches it, not per ticket it resolves

Cost per resolved conversation1,984 tickets reach the AI to produce 1,000 resolutions, and every one of those tickets bills a dollar. $1,984 for the month, against a $3,000 annual minimum that is irrelevant at this volume and painful at a small one.Checked 2026-08-09

$1.98 at 1,000 resolved conversations a month

Handoffhelply.com/pricing, which does not mention handoff, escalation or transfer.Checked 2026-08-09

Not published on the pricing page. Their positioning is that humans handle what the AI cannot, but there is no documented mechanism or transcript guarantee on the page a buyer reads.

SAML / SSOhelply.com/pricing, which does not mention SSO or SAML.Checked 2026-08-09

Not published on their pricing page

Knowledge sources

Your help centre and connected content. They publish a self-improving knowledge base as a product surface rather than a connector list.

Trial

Not stated on their pricing page