Top Chatbot Companies

The five segments of the chatbot market

The fastest way to shorten a vendor shortlist is to work out which of these bands you fall into and delete the other four. The bands differ by how you buy, not just by what you pay.

Enterprise-only platforms

Enterprise-only chatbot companies sell exclusively through a sales team, publish no rates on their own website, and expect a multi-month deployment with professional services attached.

These are contact-centre platforms first and chatbots second. They integrate with telephony, workforce management and CRM, they handle voice as well as chat, and they are bought by a procurement team against a requirements document. If you cannot name the person who will own the integration project, you are not their buyer.

The tell is the pricing page. On 20 July 2026, cognigy.com/pricing and kore.ai/pricing both return HTTP 404, and ada.cx/pricing shows no rates at all. The only public numbers for any of the three sit on AWS Marketplace, where the listings are annual commitments rather than menu prices.

Examples: NiCE Cognigy, Kore.ai, Ada, SoundHound Amelia, Yellow.ai, LivePerson

Enterprise plus mid-market suites

These companies sell a helpdesk or CRM with an AI agent attached, publish per-seat plan pricing, and quote the AI component separately per resolution.

This is where most support teams between fifty and five thousand agents actually land. The chatbot is not the product, the platform is, and the agent is sold as consumption on top of the seat licence. That structure is now near universal in the band.

Examples: Salesforce Agentforce, Zendesk, Fin (formerly Intercom), Freshworks

Mid-market AI-native specialists

AI-native specialists build only the agent, with no helpdesk of their own, and sell it to companies that already have a support stack they do not want to replace.

The pitch is depth over breadth. Because they are not defending a legacy ticketing product they tend to move faster on the model side, and because they have no seats to sell they usually price on outcomes. The risk is the same one that keeps repeating in this category: independent specialists get acquired.

Examples: Sierra, Botpress, 1mind

SMB self-serve builders

SMB chatbot companies let you sign up with a card, point the bot at your website, and be live the same afternoon with no sales call at all.

Prices are on the website, plans are flat monthly, and setup is measured in hours. What you give up is voice, deep contact-centre integration and compliance certification. For a business whose support volume is in the hundreds or low thousands of conversations a month, none of those absences hurt.

Examples: Tidio, matram.ai, Chatbase, ManyChat

Ecommerce specialists

Ecommerce chatbot companies build around order data, so the bot can answer where-is-my-order, process a return and apply a discount rather than just retrieving an article.

The differentiator is not language quality, it is write access to the store. A general chatbot can tell a shopper the returns policy. A store-integrated one can start the return. That gap decides most ecommerce evaluations.

Examples: Gorgias, Zowie

Top chatbot companies at a glance

The last column is the one to read first. A company that will not publish a rate is telling you something true about how long its sales cycle is going to be.

Prices in this table were read off each vendor's own live pricing page or AWS Marketplace listing on 20 July 2026 and are reproduced with the billing basis the vendor states. They change often. Treat them as evidence of the pricing model rather than as a quote.

When none of these companies is the answer

A page that segments a market implies you belong somewhere in it. Plenty of readers do not, at least not yet. These stages escalate, and the table above only starts being useful somewhere in the middle of them.

Stage one: there is nothing here to automate

A handful of enquiries a week, all answered the same day by whoever knows the answer. Nothing waits. Nothing repeats often enough that a machine could have taken it, and the questions that do repeat could sit on a page of your own site by Friday. Buying at this stage gets you a widget to configure, a settings screen to learn, transcripts nobody reads and a monthly line on the card. Write the answers down instead. You will need them written down whatever you buy later, so the work is not wasted. It's just early.

Stage two: the friction is real but it is a content problem

Now the same handful of questions comes back every week and somebody is pasting the same paragraphs out of an old thread, slightly out of date. It feels like a software problem. Usually it is not. Look at what your team is actually retyping, then check whether that answer exists anywhere a bot could reach it. If it does not exist in writing, no vendor in any band on this page can conjure it, because a generated answer is only as good as the source material behind it. Fix the material first. Everything you buy afterwards works better and costs less.

Stage three: you need a chatbot, just not from the band you have been reading about

This is the expensive one, and it does not look like a mistake while you are making it. Someone senior says the word enterprise, so the shortlist fills up with platforms that publish no rates and sell through a named account team. Weeks later you are in a security review for a product that will handle a few hundred conversations a month, and the requirements document has taken longer to write than a self-serve deployment would have taken to run. The volume never justified the procurement. If that is you, drop a band and start again. Nobody has ever regretted buying the smaller thing first.

Stage four: genuinely borderline, and nobody can settle it for you

But some cases sit on a boundary and stay there. You have the conversation volume of a suite buyer and the procurement process of a small company, or exactly the opposite. Your answers live in a database rather than in prose, so what you actually need is write access to your own systems rather than a better model. Or the product you like most is the one most likely to change hands, and you are weighing a good tool against an unknown owner. There is no clean rule for any of that. What helps is deciding which risk you would rather carry, then asking every vendor on your list the same question about it and comparing the answers instead of the demos.

Enterprise-only platforms

Six companies here, and four of them have changed owner or are mid-transaction. That is the defining feature of this band right now.

NiCE Cognigy

Cognigy was a German conversational-AI platform strong in voice automation and in European enterprises such as manufacturers and airlines. NiCE announced the acquisition on 28 July 2025 and closed it on 8 September 2025 for total final consideration of $887.3 million, the largest deal in NiCE's history. The product now sits inside NiCE CXone Mpower, which means the buyer is a contact-centre owner rather than a support manager. Its own pricing page returns a 404. The public reference points are AWS Marketplace, where a "Basic 5K pm" listing runs $43,080 for twelve months and an enterprise private-SaaS licence covering up to ten million conversations a year is listed at $1,000,000 for twelve months.

Kore.ai

Orlando-based, and the most vertical-heavy platform in the group, with prebuilt agents for banking, insurance claims, healthcare payers and telecom. It raised strategic growth investment led by AllianceBernstein Private Credit Investors in January 2026, so it remains independent while most of its peers have not. kore.ai/pricing is also a 404. Its AWS Marketplace listings price per session, ranging from $0.015 to $0.20 depending on session type, which is the most granular public unit anyone in this band offers.

Ada

Toronto-based and focused tightly on customer service rather than on being a general automation platform, which makes it a cleaner fit than Cognigy or Kore.ai if voice is not part of your problem. ada.cx/pricing shows no rates. The one public figure is an AWS Marketplace listing, "Ada Strategic" at $33,000 for twelve months including 60,000 conversations, which works out to a far higher effective unit price than the SMB tools further down this page and buys a very different level of support.

SoundHound (Amelia)

SoundHound acquired Amelia in August 2024 and now sells Amelia 7 as a voice-native enterprise agent platform. On 2 July 2026 SoundHound and LivePerson signed an amended and restated merger agreement under which LivePerson would become an indirect wholly owned subsidiary of SoundHound, subject to LivePerson stockholder approval, regulatory clearance and a related notes restructuring. If it closes, one company will own two of the six platforms in this segment.

LivePerson

The oldest company on this page and the one whose position has changed most. LivePerson restructured through 2025, ran a 1-for-15 reverse stock split in October 2025 to stay compliant with Nasdaq listing requirements, and is now the acquisition target in the SoundHound transaction above. Existing customers are well served by the platform's messaging depth. New buyers should ask directly about roadmap ownership before signing a multi-year term.

Yellow.ai

Headquartered in San Mateo with its main engineering centre in Bangalore, Yellow.ai is the strongest option in this band for organisations that need genuinely wide language coverage across voice and chat, and it has repositioned itself over the last two years from chatbot builder to agent platform. It publishes no rates.

Enterprise plus mid-market suites

The four companies most support teams will actually shortlist. All four publish something about price, and all four have moved to charging for the AI separately from the seat.

Salesforce Agentforce

Salesforce reported that Agentforce reached $1.2 billion in ARR in Q1 FY27, up 205 percent year over year, which makes it the fastest-scaling product in this category by the only figure any vendor has actually disclosed. Pricing is $2 per conversation, or $0.10 per agent action bought through Flex Credits at $500 per 100,000 credits. On 15 June 2026 Salesforce signed a definitive agreement to acquire Fin, formerly known as Intercom, for approximately $3.6 billion, with closing expected in Q4 of Salesforce's fiscal 2027. Agentforce is the right answer when your customer data already lives in Salesforce and the wrong one when it does not.

Zendesk

Zendesk has built its AI agent capability by acquisition twice. It bought Ultimate in March 2024, folding the standalone product into its own AI agent platform, and it announced the acquisition of Forethought on 11 March 2026, subsequently completing it and rebranding the product as Forethought AI agents by Zendesk. Suite Team is $55 and Suite Professional $115 per agent per month billed yearly. AI agents are included on every Suite and Support plan but billed on automated resolutions, and the per-resolution rate is not published, so you have to ask. Our Zendesk chatbot pricing breakdown goes through what that consumption line actually does to a bill.

Fin (formerly Intercom)

The helpdesk is still sold as Intercom, but the company is Fin, and it is being acquired by Salesforce. Its pricing is the clearest in the industry: $0.99 per resolution with a fifty-outcome monthly minimum, $9.99 per qualification, and exactly one outcome billed per conversation regardless of how long the conversation runs. That single-outcome rule matters, because it caps the downside of a rambling exchange in a way that per-message billing does not. Fin is genuinely excellent, and the honest caveat is corporate rather than technical: you are buying into a product whose ownership is changing. See our Intercom pricing analysis for the arithmetic.

Freshworks

Freshdesk and Freshchat with the Freddy AI agent on top, priced at $19, $55 and $89 per agent per month billed annually, with no free plan and a fourteen-day trial. Freddy includes the first 500 agent sessions on every tier and then charges $49 per 100 sessions. Freshworks is consistently the cheapest of the four on seats, which matters if your team is large and your conversation volume is not.

AI-native specialists, SMB builders and ecommerce

Three smaller bands with one thing in common: none of them is trying to sell you a contact centre.

Mid-market AI-native: Sierra, Botpress, 1mind

Sierra, founded by Bret Taylor, sells customer-facing agents on an outcome-based model in which the vendor is paid when the agent resolves the issue rather than per seat or per token. It is the clearest articulation of where this market is heading, and it is sold through a sales team rather than self-serve. Botpress sits at the opposite end of the same band: a developer-first builder for teams that want to write the orchestration themselves, with published plans plus usage-based AI spend on top.

1mind is here for a specific reason. Drift, for years the default conversational-marketing chatbot, has been sunset. Clari + Salesloft named 1mind as the successor and refers departing Drift customers to it. Drift cannot be bought as a product today, and any list that still ranks it is out of date. If you are working through that migration, our Drift alternatives page covers the options in more detail than a referral partner will.

SMB: Tidio, matram.ai, Chatbase, ManyChat

Tidio is the strongest all-round SMB choice and the only one of the four with a real free plan, which includes 50 billable conversations a month. Paid plans start at $24.17 per month billed annually, and the Lyro AI agent is metered separately from live chat, with the standalone agent starting at $32.50 per month from 50 Lyro conversations. If you want live chat and an AI agent from the same vendor with a zero-cost way to test, start there.

matram.ai is second in this band, and the specific thing it is best for is answer traceability on a flat bill. Every reply cites the page it came from, strict mode confines the bot to your approved content, and plans are $29, $69 or $199 a month with unlimited team seats, so adding people never changes the price. Volume above the top tier is handled on a quote-based Enterprise plan with invoiced billing and no published rate. Training sources include URL crawl, sitemap import, PDF and DOCX upload up to 10MB, pasted text, and connectors for Notion, Google Drive, Dropbox, OneDrive, SharePoint, Box, Confluence, Gitbook, Zendesk, Freshdesk and YouTube. The honest weaknesses: there is no free tier, only a seven-day trial without a card; the integration library is smaller than Zendesk's or Fin's; there are no compliance certifications such as SOC 2 or ISO 27001; and it is text only, so voice is not on the table.

Chatbase is the fastest way to get a bot trained on a set of documents, with a free plan and paid tiers at $32, $120 and $400 a month, priced in message credits rather than conversations. Credits run out, and the auto-recharge add-on is $40 per 1,000, so model the volume before committing. ManyChat is a different animal altogether, built for social DM automation on Instagram, Facebook and WhatsApp rather than for a website widget, and it is the right pick if your audience lives in DMs.

Ecommerce: Gorgias and Zowie

Gorgias is an ecommerce helpdesk that prices by ticket tier rather than per agent and bills its AI Agent only when it resolves a conversation. The combination of no seat tax and no charge for unresolved AI attempts is unusually well aligned with how a small merchandising team actually works. Zowie is the other credible specialist, built for online retail brands and focused on turning support contacts into completed actions rather than deflected tickets. Both beat a general-purpose chatbot on order-aware questions, and both cost more than a general-purpose chatbot. Our ecommerce chatbot guide covers when that premium is worth paying.

Three things changing across the whole market

The segments above will look different in a year. These are the forces moving them.

AI-first has displaced rule-based as the default

Every company on this page now leads with a generative agent rather than a decision tree. Rule-based products have not disappeared, and they remain correct for regulated disclosures and fixed transactional flows, but they are no longer what anyone sells on the front page. The practical consequence for a buyer is that the question has shifted from "can it understand the question" to "what is it grounded in". A generated answer is only as good as the source material behind it, which is why grounding and citation have become the features worth interrogating in a demo. The rule-based versus AI comparison sets out where each still wins.

Per-resolution pricing is becoming the standard unit

Fin charges $0.99 per resolution. Zendesk bills automated resolutions on top of seats. Gorgias bills only when its AI Agent resolves a conversation. Sierra's whole model is outcome-based. Salesforce prices Agentforce at $2 per conversation with a $0.10 per-action alternative. Five years ago this category was sold per seat or per contact; today the dominant unit is the successful outcome.

That shift is good for buyers in one respect and bad in another. It aligns vendor incentives with your results, which per-seat pricing never did. It also makes your bill a function of your traffic, which is unforecastable in a way a seat count is not, and it hands the vendor the definition of what counts as a resolution. Read that definition carefully before signing, and if you want a flat, predictable line item instead, that is precisely the trade the SMB band offers. You can model both shapes in the chatbot ROI calculator.

Gartner named the credibility problem: agent washing

In a press release dated 25 June 2025, Gartner coined the term "agent washing" for the rebranding of existing products such as AI assistants, robotic process automation and chatbots without substantial agentic capability, and estimated that only around 130 of the thousands of vendors claiming agentic AI were genuine. The same release predicted that over 40 percent of agentic AI projects would be cancelled by the end of 2027 on cost, unclear value or inadequate risk controls.

The useful test comes from Anthropic's engineering write-up of December 2024, which draws the line precisely: agents are "systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks", while workflows are "systems where LLMs and tools are orchestrated through predefined code paths". Ask a vendor which of those two sentences describes their product. Most honest answers, including ours, are the second one. matram.ai is an AI chatbot with retrieval and deterministic escalation rules, not an autonomous agent, and we would rather say so on our own page than have you discover it in week three.

And a fourth: consolidation is now the base case

In roughly two years this category has seen Ultimate absorbed into Zendesk, Amelia bought by SoundHound, Cognigy bought by NiCE, Drift sunset into a referral to 1mind, Forethought bought by Zendesk, LivePerson agreeing to merge into SoundHound and Fin agreeing to be acquired by Salesforce. Independence is not a feature you should assume. When you evaluate a specialist, ask what happens to your contract and your data if the company is acquired, and get the answer in writing.

What getting this choice wrong actually costs

Not the subscription. The things that are already spent by the time anybody admits the choice was wrong.

Say you buy a band too high. The invoice is the least of it. What you really spend is calendar: the security questionnaire, the integration scoping, the internal sign-offs, the kickoff that slips because the person who owns the CRM is away. All of that runs while your customers keep waiting on exactly the replies they were waiting on before you started. Nothing about their experience changed for a whole quarter. No refund gives the quarter back.

Then there is the cost nobody writes down. Somebody sponsored this. They stood up in a meeting and said the bot would take the repeat questions off the team, and when it did not, the next automation proposal inside that company starts from a worse position than the first one did. It's invisible on the way in and permanent on the way out. Budget you can ask for twice. Credibility you mostly cannot, and the person who spent it is usually the same person you need on side for the second attempt.

And there is the work that did not happen. The stretch your best support person spent in vendor calls was a stretch they did not spend rewriting the help centre, which is the one piece of work that improves every option on this page at once. Buying a band too low costs differently but not less: you outgrow the tool mid-year, migrate under pressure, prepare your content a second time for a second platform, and pay for both while you do it. So the question worth asking before you sign is not whether the price works. It is what you would still be holding, and who would still back you, if you had to start this again next year.

Which segment you actually belong in

Answer this before you look at a single vendor page. It removes about eighty percent of the market from consideration in one step.

If you sit on a boundary, the cheaper decision is almost always to start one band down and move up when you outgrow it. Migrating from an SMB builder to a suite is a week of work. Migrating out of a signed enterprise term is a year. See the small business chatbot guide if that is where you land.

Frequently asked questions

Sources

• Salesforce investor relations, "Salesforce Signs Definitive Agreement to Acquire Fin", 15 June 2026 - The ~$3.6bn agreement to acquire Fin, formerly known as Intercom, expected to close in Q4 FY27, and Agentforce reaching $1.2 billion in ARR in Q1 FY27, up 205% year over year.
• Fin pricing page, read 20 July 2026 - $0.99 per outcome or resolution, a 50-outcome monthly minimum, $9.99 per qualification, and one outcome billed per conversation.
• Gartner press release, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027", 25 June 2025 - The term "agent washing", the estimate that only around 130 of thousands of vendors claiming agentic AI were genuine, and the 40% project cancellation prediction.
• Anthropic engineering, "Building effective agents", 19 December 2024 - The definitions used to test agent claims: agents as systems where LLMs dynamically direct their own processes and tool usage, workflows as systems orchestrated through predefined code paths.
• Cognigy newsroom, "NiCE Closes Acquisition of Cognigy" - NiCE closing its acquisition of Cognigy on 8 September 2025 for total final consideration of $887.3 million, and Cognigy now sitting inside NiCE CXone Mpower.
• Zendesk newsroom, "Zendesk Completes Acquisition of Forethought" - Zendesk announcing the Forethought acquisition on 11 March 2026 and completing it, with the product rebranded as Forethought AI agents by Zendesk.
• Zendesk newsroom, "Zendesk to Acquire Ultimate", March 2024 - Zendesk acquiring Ultimate in March 2024 and folding it into its own AI agent platform.
• Zendesk pricing page, read 20 July 2026 - Suite Team at US$55 and Suite Professional at US$115 per agent per month billed yearly, with AI agents included on every Suite and Support plan but billed on automated resolutions at a rate not published.
• Freshdesk pricing page, read 20 July 2026 - Growth $19, Pro $55 and Enterprise $89 per agent per month billed annually, no free plan, a 14-day trial, and Freddy AI Agent at 500 included sessions then $49 per 100 sessions.
• Tidio pricing page, read 20 July 2026 - A free plan with 50 billable conversations, Starter from $24.17 per month, and the Lyro AI agent metered separately from $32.50 per month at 50 Lyro conversations.
• Chatbase pricing page, read 20 July 2026 - A free plan with 50 message credits, Hobby $32, Standard $120 and Pro $400 per month, and auto-recharge credits at $40 per 1,000.
• Gorgias pricing page, read 20 July 2026 - AI Agent included on every plan and billed only when it resolves a conversation, with helpdesk plans priced by monthly ticket volume rather than per agent.
• SEC filing, SoundHound AI and LivePerson amended and restated merger agreement, 2 July 2026 - LivePerson becoming an indirect wholly owned subsidiary of SoundHound AI subject to stockholder approval, regulatory clearance and notes restructuring.
• LivePerson Form 10-Q, Q1 FY2026 - LivePerson's 2025 restructuring plan and the 1-for-15 reverse stock split implemented in October 2025 to comply with Nasdaq listing requirements.
• SoundHound AI newsroom, acquisition of Amelia - SoundHound acquiring Amelia in August 2024 and expanding into enterprise conversational AI.
• Kore.ai newsroom, strategic growth investment from AllianceBernstein, January 2026 - Kore.ai remaining independent, headquartered in Orlando, with growth investment led by AllianceBernstein Private Credit Investors.
• Clari + Salesloft newsroom, 1mind partnership - Drift being sunset and Clari + Salesloft referring existing Drift clients to 1mind as the successor under an exclusive agreement.
• ContactBabel, 2026 US Contact Center Decision-Makers' Guide - The average inbound call costing $7.20, 47% more than an email and 23% more than a web chat, used as the channel-cost backdrop for automation spend.