The best AI ticketing tools, compared on what a ticket really costs

The best artificial intelligence ticketing tool depends on a single number, what a manually handled ticket costs you today. Below ten euros, no AI agent on the market really pays for itself. Above twenty, almost all of them do. For high volume and an already structured team, Zendesk and Intercom Fin remain the two references. For a European SME that wants a fixed price, Crisp offers the best simplicity-to-price ratio. To keep your current tool and add AI to it, Klark and Cobbai do that job from France.

Here are the eight tools, their prices taken from the vendors' own pages in September 2026, and above all the question nobody asks before signing.

How this comparison was built, and what Tandem actually tested

The criteria come before the names, because a ranking without stated criteria is only an opinion in disguise. Five points decided the order.

1. Real price, with the subscription and the AI billing counted separately, taken from the vendor page and dated to the month.
2. How the AI is billed, because a per-seat rate and a per-resolution rate do not compare, and one of the two explodes with volume.
3. Hosting location and vendor identity, checked on the pricing or trust page, never on a flag placed on the homepage.
4. What the agent does when it does not know, which is the real technical differentiator in this category.
5. Reversibility, meaning the ability to add AI without changing helpdesk, or to leave without losing your history.

On transparency, here is the exact situation. Tandem has deployed support agents on client engagements, on stacks built around HubSpot Service Hub, Shopify and voice components. We have also audited environments running on Intercom. We have not deployed Zendesk, Freshdesk, Gorgias, Cobbai or Zammad at a client. For those five, the opinion rests on documentation, pricing pages and public reviews, and that is stated each time. The detail of our support method lives in our review of AI use cases in business.

The eight AI ticketing tools in one table

Cost per ticket, the only calculation that settles a support project

Before comparing eight pricing tables, work out what a ticket costs you today. Take the loaded payroll of the support team, add tools and management, divide by the number of tickets handled in the month. On the engagements Tandem has run, that figure sits around 20 euros. An AI agent billed per resolution costs between 0.75 and 1.35 euros among the vendors in this comparison, and between 0.20 and 1 euro on a custom build. The gap is a factor of twenty, and it is what funds the project.

A second figure to know before choosing. At most of our clients, roughly three tickets out of ten are simple, repetitive requests, from a forgotten password to an order status. That first wave is what gets automated within a few weeks. At one online retailer we reached more than 70 % automatic handling, but only after several months of iterating on request types. I described that architecture in my newsletter issue on classic AI use cases in business, and it holds for any of the eight tools below.

1. Zendesk, the reference for high volume

Zendesk is the de facto standard for customer service at scale, and it stays that way in 2026. Its strength is not its AI agent, it is everything around it. Skills-based routing, business rules and reporting first. Then a marketplace of more than 1,800 applications, which solves most integration cases without development. You pick it when your support team passes ten agents and works across several channels.

Key features

• AI agents are included in every Suite and Support edition, and billing only triggers on resolutions considered successful.
• Skills-based routing sends each ticket to the agent with the right expertise, which matters more than AI on technical support.
• The Copilot module assists the agent while they write, and is billed separately.
• The action builder triggers multi-step operations inside Zendesk and in third-party systems, a refund or an order update for instance.
• The knowledge base feeds self-service, the AI agent and the human agents alike, with more than 80 languages supported.

Underlying model and limits

Zendesk does not disclose the language model it uses, and nothing requires it to. The vendor claims resolution of up to 80 % of complex requests on any channel, a marketing figure to read as an upper bound. The real limit sits elsewhere. Hosting data inside the European Union requires a paid module, and advanced privacy features are reserved for Enterprise editions. On a support desk handling sensitive data, that detail changes the budget.

Pricing

Taken in September 2026 from the Zendesk pricing page, on annual billing. Suite Team at 55 euros per agent per month, Suite Professional at 115 euros, Suite Enterprise on request. The Copilot module adds 50 euros per agent, the contact centre 82 euros. Support editions start lower, at 19 euros a month. AI agent resolutions are billed on usage beyond the credits included in the plan. Zendesk is the only one of the eight tools here that does not publish that unit rate.

What users say

Zendesk shows 4.3 out of 5 across more than 6,000 G2 reviews in September 2026, the largest review base in the category. The recurring compliment is functional depth and reliability at scale. The recurring complaint is total cost, which climbs quickly once modules stack up, and a heavy learning curve for a small team.

My take on Zendesk

I have not deployed it on an engagement, so I stick to what can be verified. What I take from its pricing is that it asks for an act of faith the other seven do not. You commit to a per-resolution model without knowing the price of a resolution. On a desk handling a few hundred tickets a month, that blind spot is enough to make the budget impossible to defend internally. Ask for the unit rate and the counting basis in writing before signing.

2. Intercom Fin, the most complete at autonomous resolution

Intercom made a sharper bet than its competitors. Fin, its AI agent, is built natively into the helpdesk rather than bolted on, and the vendor bills on outcome rather than usage. The product targets software vendors and online products first, the ones whose customers use chat rather than the phone. On that ground it is the hardest to beat.

Key features

• Fin answers in conversation rather than in a decision tree, and hands over to a human as soon as it leaves its knowledge perimeter.
• The unified inbox mixes human conversations and Fin conversations, with no tool switch for the agent.
• Procedures define what the agent is allowed to do, and what it must escalate.
• The knowledge base is built from the help centre, internal articles and past conversations.
• Every conversation Fin resolves feeds the next one, which makes the resolution rate improve over the first months.

Underlying model and limits

Intercom does not publish the models behind Fin, and the vendor is American. The limit lies in the counting. A resolution is billed once per conversation whatever the number of questions asked, which favours the customer. But four different events trigger billing, resolution, procedure handoff, disqualification and lead qualification. That last one costs ten times more than the others. Read the definition before estimating an invoice, and plan for a monthly minimum of 50 billed events.

Pricing

Taken in September 2026 from the Intercom pricing page. Three editions per seat per month, Essential at 29 dollars or around 25 euros, Advanced at 85 dollars or around 78 euros, Expert at 132 dollars or around 120 euros. Fin is billed 0.99 dollar per event, around 90 cents, on every edition. Lead qualification rises to 9.99 dollars. Unused events do not roll over to the next month.

What users say

Intercom shows 4.5 out of 5 across roughly 3,300 G2 reviews in September 2026. The recurring compliment is the quality of the chat experience and the resolution rate Fin actually reaches. The recurring complaint is the invoice, judged unpredictable from month to month, and a seat-plus-resolution stack that penalises large teams with low volume.

My take on Intercom

We came across Intercom in the stack of a SaaS vendor Tandem audited, alongside HubSpot and Aircall. What I take from it fits in one sentence. Fin is excellent on questions whose answer is written down somewhere, and it becomes a cost line as soon as your requests need data pulled from another system. The right move before signing is to sort three months of tickets into two piles, the ones whose answer exists in your knowledge base and the rest. The first pile is the only one Fin will handle.

3. Freshdesk, the best features-to-price ratio

Freshdesk is the tool you pick when you want Zendesk-level coverage without its entry ticket. At 17 euros per agent, the Growth edition already covers multichannel, automations and a knowledge base. The Freddy range adds AI in two distinct pieces, a customer-facing agent and a copilot for the human agent.

Key features

• The Freddy agent answers autonomously on chat, email and the customer portal, with 500 sessions included on each of the three plans.
• The Freddy copilot suggests the reply to the agent, summarises the thread and adjusts tone, and is billed per agent.
• Scenarios automate recurring sequences of actions, which replaces a good share of manual triage.
• The customer portal and knowledge base can be split per brand, useful when one team serves several banners.
• Reporting is included from the Pro edition, where several competitors reserve it for the top tiers.

Underlying model and limits

Freshworks does not publish the detail of the models behind Freddy. This offer counts in sessions rather than resolutions, which is not the same thing. A session opens as soon as the agent engages the conversation, whether or not it succeeds. You therefore also pay for exchanges that end up with a human, unlike the Intercom model. On a desk where the agent escalates often, that nuance raises the real invoice. European hosting exists as an option, to be confirmed in writing for your region.

Pricing

Taken in September 2026 from the Freshdesk pricing page, on annual billing. Growth at 17 euros per agent per month, Pro at 52 euros, Enterprise at 84 euros. All three plans include 500 Freddy agent sessions, after which extra sessions cost 45 euros per 100. The Freddy copilot is billed 29 euros per agent per month, on annual commitment only.

What users say

Freshdesk shows 4.4 out of 5 across roughly 3,700 G2 reviews in September 2026. The recurring compliment is the ratio between price and functional coverage, and how quickly it starts up. The recurring complaint is customisation depth, judged insufficient once processes become specific, and slow response times from the vendor's own support.

My take on Freshdesk

I have not deployed it on an engagement, so I stick to what the documentation allows me to verify. The point I would watch is session counting. Run the exercise on your own figures before comparing with Intercom. If your agent has to escalate one request in two, a billed session that fails costs the same as a full resolution. The advertised 45 euros per 100 sessions then becomes 90 euros per 100 resolutions.

4. Gorgias, the e-commerce specialist

Gorgias is the only tool in this comparison that never bills per seat. You pay for a ticket volume, and the number of agents changes nothing. For an online retailer whose team grows at Christmas then shrinks, that mechanism is far fairer than a per-person licence. The tool is built for Shopify, WooCommerce, BigCommerce, Magento and PrestaShop, and the AI agent sees the order, the tracking and the catalogue.

Key features

• Volume-based billing fully frees the number of agents, up to 500 users on the higher editions.
• The AI agent accesses store data, which lets it handle an order status or a return without human involvement.
• It also recommends products during the conversation, and Gorgias measures the revenue support generates.
• Knowledge-gap detection flags the topics the agent answers poorly, on the higher editions.
• Native integrations with the main commerce platforms avoid building a connector.

Underlying model and limits

The vendor is American and does not publish the models it uses. The limit is an accounting one and deserves to be understood before signing. A conversation handled by the AI agent is counted twice, once as a ticket in your plan and once as an automated interaction. Beyond the plan, an automated interaction costs 1.50 dollars, around 1.35 euros, which makes it the most expensive resolution in this comparison. Outside online retail, the tool also loses its main appeal.

Pricing

Taken in September 2026 from the Gorgias pricing page. Four volume-based editions. Starter at 40 dollars a month for 50 tickets, Basic at 77 dollars annually for 300 tickets. Pro at 471 dollars for 2,000 tickets, Advanced at 1,227 dollars for 5,000 tickets. The helpdesk portion alone comes to 10, 50, 300 and 750 dollars depending on the edition. Overage costs 0.40 dollar per ticket on the first two editions then 0.36, and 1.50 dollars per automated interaction.

What users say

Gorgias shows 4.6 out of 5 across roughly 550 G2 reviews in September 2026, the best score in this comparison on a mid-sized review base. The recurring compliment is the Shopify integration, judged unmatched. The recurring complaint is the jump from one edition to the next, brutal in price, and overage charges that surprise at the first order peak.

My take on Gorgias

We have not deployed Gorgias, but Tandem has automated support for online retailers on equivalent Shopify stacks, so the ground is the same. The trap I would flag is double counting. A store that automates 40 % of its conversations burns through its plan faster, not slower, because every automated conversation is still a ticket. Simulate your invoice on a peak month before choosing your edition, not on an average month.

5. Crisp, the simplest option for a European SME

Crisp is built in Nantes, has existed since 2015 and claims more than 600,000 users. The company has stayed independent, without raising funds, which explains how stable its pricing has been. Its main argument fits in one word, flat rate. You pay a price per workspace rather than per seat, and you add agents without the invoice moving. For an SME of ten to thirty people, it is the clearest model in this comparison.

Key features

• The flat rate per workspace covers the whole team, with no licence per agent.
• The AI agent is hosted by Crisp, which avoids adding a third-party vendor into your processing chain.
• The inbox gathers website chat, email, social channels and the phone into a single thread.
• The knowledge base feeds both self-service and the AI agent, and translates into several languages.
• The free tier is genuinely usable to start, with a working inbox and chat.

Underlying model and limits

Crisp states on its homepage that its data stays inside the European Union and that its AI is hosted in house. No American vendor in this comparison makes that claim. The limit is functional. The AI credits bundled into the plan have a ceiling, roughly 90 automated conversations on the Mini edition and roughly 450 on Essentials. Beyond that you have to move up an edition. And on complex processes, routing and business rules stay behind what Zendesk or Freshdesk offer.

Pricing

Taken in September 2026 from the Crisp pricing page, displayed in dollars. A free tier, then Mini at 45 dollars a month or around 40 euros. Essentials rises to 95 dollars or around 85 euros, Plus to 295 dollars or around 270 euros. An Enterprise edition is available on request. The bundled AI credits are worth 5, 25 and 75 dollars by edition, which represents roughly 90, 450 and 1,350 automated conversations. The trial runs 14 days, with no card required.

What users say

Crisp shows 4.4 out of 5 across roughly 180 G2 reviews and 4.5 on Capterra in September 2026, a much smaller base than the American players. The recurring compliment is value for money and the responsiveness of the vendor's own support. The recurring complaint is reporting limits and AI credit ceilings, which force an edition change sooner than expected.

My take on Crisp

Crisp is one of the tools I regularly find already in place at our clients when we arrive, alongside Intercom, HubSpot and Front. What I think of it is simple. It is the best starting point for an SME that does not yet have a structured support function, because the price does not punish the team for growing. A company handling more than a thousand tickets a month will eventually hit the credit ceiling, though, and will have to either move up an edition or plug in an external agent.

6. Klark, the French AI layer that sits on your helpdesk

Klark does not sell a helpdesk, it sells the AI yours is missing. The company was created in July 2023 in the Bordeaux area by Nicolas Pellissier, formerly of Back Market, James Rebours and Yoann Chambrun. It raised 1.7 million euros in January 2025 from Kima Ventures, Super Capital and Source Ventures. Thibaud Hug de Larauze, founder of Back Market, and Thomas France, co-founder of Ledger, joined as angels. The product installs as an application inside Zendesk, Freshdesk, Gorgias, Salesforce, Front, Intercom, HubSpot or Crisp.

Key features

• Installation happens from your helpdesk's marketplace, with no integration project and no data migration.
• Klark rebuilds a knowledge base from six months to a year of conversation history, instead of asking you to write one.
• The copilot proposes a drafted reply to the agent, who validates or corrects it before sending.
• A customer-facing agent handles simple requests in conversation, and hands over as soon as it leaves its scope.
• Translation lets a French-speaking agent reply to an Italian customer without switching tools.

Underlying model and limits

Klark relies on several market models and manages risk through knowledge rather than through the model. That is the most interesting part technically, and Nicolas Pellissier explained it to me in detail in the episode below. Personal data never goes to the model, it is replaced by markers before the call and reinjected afterwards, and storage stays in Europe. When the answer does not exist in the knowledge base, the agent does not try, it escalates. The limit is scope, Klark handles neither your routing nor your reporting, so you need a helpdesk already.

His comparison has stayed with us in our workshops. A language model behaves like a student in an exam, it attempts an answer because it has nothing to lose. The danger is not the absurd answer, that one gets spotted. The danger is the plausible answer, along the lines of take your item back to your local store, for a returns process that does not exist. Klark's counter is to withhold the knowledge from the machine rather than asking it to restrain itself.

Pricing

Klark does not publish a pricing table, and bills on success, meaning when the answer it produces is actually used. The trade publication En-Contact quoted an order of magnitude of roughly 50 cents per used answer in March 2024, with 40 to 50 % of suggestions actually taken up by agents. That figure is more than two years old, it gives an order of magnitude and not a rate. The vendor claims more than 100 client brands and displays Back Market, Click & Boat, ba&sh and Voggt, along with SOC 2 certification.

What users say

Klark has no usable review volume on software marketplaces in September 2026, and saying so beats padding. The only public feedback is the testimonials published by the vendor, which mention handling time dropping from around five minutes to under three per ticket. A testimonial chosen by a vendor is not a verified review, and reads as such.

My take on Klark

I have not deployed Klark at a client. I did spend an hour with its chief executive on the precise subject of hallucination. It is the most serious conversation I have had about this category. Choosing to handle the risk through knowledge, upstream, rather than through guardrails bolted on afterwards, is the right call. The second thing I like is commercial. Billing when the answer gets used aligns vendor and customer, where a per-session rate also bills the failures.

7. Cobbai, the French AI helpdesk

Cobbai is a French company created in August 2019, still independent, which started with voice-of-the-customer analysis before becoming a full support platform. It has the particularity of covering both strategies, installing on an existing helpdesk such as Zendesk, Salesforce or Intercom, or replacing it entirely. Its three agents carry names that state their role, Front facing the customer, Companion for the human agent, Analyst for categorisation and analysis.

Key features

• The Front agent holds the conversation with the customer on chat, email and self-service.
• The Companion agent assists the human agent with drafts and searches through the knowledge.
• The Analyst agent categorises, routes and surfaces trends in contact reasons, a direct legacy of the company's original business.
• Supervised categorisation on the lower editions becomes unsupervised on the top edition, useful when contact reasons shift quickly.
• The Enterprise edition opens up server location choice, with an option compliant with health data hosting requirements.

Underlying model and limits

Cobbai does not publish the models it uses and holds SOC 2 type 2 certification. The dedicated server option compliant with French health data hosting requirements is only available on the top edition. That makes it one of the rare French answers for a support desk touching medical data. The limit lies in the price structure, the most composite in this comparison. You pay a seat, then each conversation, then each full resolution. Three counters, three possible surprises.

Pricing

Taken in September 2026 from the Cobbai pricing page, on annual billing. Essential at 29 euros per seat per month for one to five seats, one AI agent and under 500 monthly conversations. Advanced at 59 euros for five to thirty seats, up to three agents and unlimited conversations. Enterprise at 99 euros from ten seats, unlimited agents and hardened security. On top of that come 0.15 euro per conversation on all three editions, and 0.75 euro per full resolution completed by the Front agent without escalation.

What users say

Cobbai has no usable review volume on software marketplaces in September 2026. The vendor does display verifiable references that matter on the French market, including Doctolib, Veolia and ManoMano. A displayed reference is not a user review, but on this segment it says something about the level of scrutiny the platform had to satisfy.

My take on Cobbai

I have not deployed it on an engagement, so I stick to what the pricing and the documentation allow me to verify. Two things strike me as notable. At 75 cents, its resolution is the cheapest published rate in this comparison, almost half what Gorgias charges beyond the plan. And the Analyst agent answers a need the others handle poorly, understanding why people contact you rather than merely replying to them. On a support desk trying to cut volume at the source, that piece matters as much as the answering agent.

8. Zammad, the open source option you host yourself

Zammad is the only tool in this comparison you can install on your own servers, because its code is open. It is the answer for organisations that cannot hand their customer exchanges to a third-party platform, public sector, healthcare, defence or sensitive industry. The vendor is German and puts two claims on its homepage, made and hosted in Germany, and sovereign AI.

Key features

• The software is open source and installs on your infrastructure, with no dependency on a cloud provider.
• The vendor-hosted edition remains an option, with servers located in Germany.
• Channels cover email, web form, SMS, chat, Telegram, WhatsApp and Facebook depending on the edition.
• Service level agreements and the knowledge base arrive from the Professional edition.
• Smart Assist adds AI as an option, with per-call billing that stays independent of the subscription.

Underlying model and limits

Zammad bills AI per technical call rather than per resolution, which changes the economics entirely. One call costs 3 cents, and a conversation consumes several. The calculation is therefore harder to anticipate, but the order of magnitude stays far below per-resolution models. The limit is clear. Zammad's AI agent does not have the maturity of Fin or the Zendesk agent, and self-hosting transfers the running cost, backups and updates to you. This is not a tool for a team without technical skills.

Pricing

Taken in September 2026 from the Zammad pricing page, on annual billing. On the hosted version, Starter costs 7 euros per agent per month up to five agents. Professional is 16 euros up to thirty-five agents, Plus 25 euros with no agent limit. Self-hosted, the software is free and only support is paid, Business at 2,999 euros a year, Enterprise at 5,999 euros, Corporation at 9,999 euros. The Smart Assist option costs 3 cents per AI call.

What users say

Zammad has no usable review volume on software marketplaces in September 2026, which is common for an open source project whose community speaks elsewhere. The public feedback available praises interface clarity and customisation freedom, and flags the absence of some features that are mature at commercial vendors.

My take on Zammad

I have not deployed it on an engagement, so I stick to what can be verified. Zammad is not chosen for its AI, it is chosen on the sovereignty question, and that decision belongs to the executive committee rather than to the support team. If that constraint is not yours, the other seven tools will cost you less in time. If it is yours, this is the only serious option on the list, and billing at 3 cents per call is by far the gentlest in this comparison.

Buy a support agent or build one?

The question comes up the moment an executive sees a per-resolution rate and does the multiplication. On support specifically, the answer is documented and it leans clearly one way. The fund a16z surveyed one hundred chief information officers in 2025. More than 90 % of respondents were testing a third-party application for customer service rather than an internal build. That is the highest proportion of any function studied.

I summarised that trade-off in a one-page chart on choosing between building and buying an AI solution. The logic that works is hybrid. You buy to move fast, you build to differentiate. On a support desk, that means plugging in a market agent within two to four weeks and measuring the real gain. Custom work comes after, only on the cases the tool cannot take. That extra layer often runs through light orchestration, such as an n8n workflow, rather than a rebuild. We map that boundary in our article on the difference between AI agents and AI workflows.

Which AI ticketing tool should you choose for your profile?

• You are starting out, under 300 tickets a month. Crisp, for its per-workspace flat rate and a free tier that is genuinely usable. The price will not move as the team grows.
• You sell online on Shopify. Gorgias, without hesitation, because its agent sees the orders. Simulate your invoice on a peak month, not an average one.
• You publish software and your customers use chat. Intercom, because Fin is the most complete at autonomous written resolution and its billing counts only once per conversation.
• You want functional coverage at the best price. Freshdesk, after first checking how many of your sessions end up with a human.
• You already have a helpdesk and do not want to change it. Klark or Cobbai, which install inside it within hours and spare you a migration.
• Your data cannot leave your premises. Self-hosted Zammad, the only serious option on this list against that criterion.
• You are past ten agents across several channels. Zendesk, for routing and reporting, while demanding the per-resolution rate in writing before signing.

One case is not covered here, the support desk that already lives entirely inside Salesforce. Its answer is Agentforce, but that belongs to the choice of system of record rather than the choice of support tool, and it deserves its own article.

One channel is also missing from this table, voice. None of the eight tools handles inbound calls properly, and that is often half the volume. On the support engagements Tandem has run, the phone sometimes accounted for 70 % of contacts. The subject has its own platforms, which we compared in our comparison of AI voice agent platforms, and its own method, described in our AI voice agent guide.

What nobody says about AI in customer service

Three blind spots come back in every comparison in this category, and they decide the outcome all the same.

Cost per resolution is going up, not down

The dominant narrative promises a steady fall in the cost of AI. Gartner says the opposite for customer service. In a press release from January 2026, it forecasts that the cost per resolution of generative AI will exceed three dollars by 2030, more than many offshore human agents in B2C. The reasons given are rising data centre costs, the end of subsidised pricing and increasingly token-hungry use cases. The practical consequence is clear. A per-resolution contract signed today will not be renegotiated downwards, so lock the term and the ceiling.

Since August 2026, your agent has to introduce itself

Article 50 of the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed of it, unless it is obvious from the context. That obligation has applied since 2 August 2026. In practice, a support agent posing as a human is no longer an option. Check that the tool you choose lets you display that notice, and that your team has not switched the label off for commercial comfort.

Automatic resolution is not the real gain

Every sales demo shows the same screen, the customer asks a simple question and the agent answers alone. That level accounts for roughly three tickets out of ten, and it is genuinely easy. The real gain sits in the other seven. There the agent pulls the order, the history and the reason, qualifies the request, then prepares a draft the human agent validates. That level is billed nowhere and demonstrated never, but it is what halves handling time.

Which AI ticketing tool should you pick in 2026?

If you were to keep one name per situation, here are three. Crisp for a European SME that wants a predictable budget and data in Europe. Intercom for an online product whose customers write rather than call. Klark or Cobbai to add AI without touching the helpdesk already in place, with French vendors you can reach.

But the tool choice is not what will decide your result. What will decide it is the quality of your knowledge base and the definition of what the agent is allowed to do when it does not know. A mediocre tool on clean knowledge beats an excellent tool on outdated documentation, every time. Start by sorting three months of tickets and writing the twenty answers that come back most often. That work depends on no vendor, and it is most of the distance.

Tandem supports customer service teams through that framing, from the cost-per-ticket calculation to deploying the agent. If you want to know what can be automated at your company before buying anything, start with an AI audit of your processes. If the framing is done and you are looking for delivery, our Autopilote offer covers building and running these agents.

This comparison is rechecked every quarter, including prices, whether the tools still exist, and screenshots. The last check was September 2026.