Seven platforms, and only two of them will tell you what they cost. A comparison of Chatbase, Intercom Fin, Zendesk, Decagon, Sierra, Gladly Sidekick and Salesforce Agentforce, written by someone who works at one of them.
Zeyad Genena, Chatbase · AI agents and autonomous tools · Published 2026-08-21 · Updated 2026-08-21
Start with the thing nobody wants to talk about
Most comparisons of this category rank platforms on capability. That is the wrong place to start in 2026, because capability has largely converged. Every serious platform here connects to your knowledge base, answers in natural language, takes some form of action in external systems, and hands off to a human when it fails. The demos look alike because the products have become alike.
What has not converged is whether the vendor will tell you the price.
Of the seven platforms below, two publish a rate you can plan against. Chatbase lists every plan on its pricing page. Intercom publishes Fin at $0.99 per outcome. Zendesk publishes seat prices but not the per-resolution number that actually determines your bill. Gladly, Decagon and Sierra publish nothing at all and send you to a demo form.
That is not a minor detail. It decides who can evaluate the product at all. If you are a support lead with a budget and no procurement department, half this market is closed to you until you have sat through a sales call. So this guide is organised by what you can actually find out.
Chatbase
Chatbase is an AI agent platform for customer experience, running across chat, email and voice. Disclosure again, since it matters here: I work there.
The argument for it is speed to live and a published price. The free plan is $0 for 50 message credits and one agent, with no credit card, which means you can build something and see it answer a real question before anyone signs anything. Hobby is $40 a month for 700 credits. Standard is $150 for 4,000 credits and is the tier most small support teams land on, because it is where Helpdesk, voice, telephony, Outbound campaigns and API access switch on. Pro is $500 for 15,000 credits. Annual billing takes 20% off. Extra agents are $25 a month each, and removing Chatbase branding from the widget costs $99 a month.
Structurally, the platform covers the whole agent lifecycle. You connect data sources, define the agent’s role with instructions and Procedures, which are plain-English standard operating procedures with approval gates, test in the Playground, deploy across channels, then optimise using analytics on topics, sentiment, resolution and confidence. Agents take real action through Actions, including Stripe subscription management and Shopify orders, and can reply with Widgets such as product galleries and booking cards rather than plain text. The native Helpdesk, launched in May 2026, gives human agents one inbox with AI-drafted replies and live takeover. Backstage lets you ask your own conversation data what customers need. Agent responses cover 95+ languages.
The differentiator I would actually defend against this field is model choice. You pick the model behind each agent from OpenAI, Anthropic, Google, xAI, Mistral and others. Every other platform here decides that for you. When a better model ships, that is the difference between a config change and a vendor migration.
On results, the numbers worth quoting are the ones on the customer pages. Jumia, listed on the NYSE, has Chatbase handling 50% of all support volume, with 80% of queries reaching Chatbase resolved without a human across eight African markets. Aplazo reports a 2.2x lift in merchant closed-won rate with no added headcount. More than 10,000 businesses run on the platform, and the G2 rating is 4.8.
Where it is the weaker choice, and this is the honest part: the credit model is cheaper than per-resolution pricing below its ceiling and gets less attractive above it, so at very high volume you should model Chatbase against Intercom’s per-outcome rate rather than assuming the flat fee wins. And the compliance posture that regulated buyers need sits on Enterprise. SOC 2 Type II and GDPR apply across the platform, and data is encrypted with TLS 1.3 in transit and AES-256 at rest, with customer data never used to train shared models. HIPAA is different: it is HIPAA-eligible on Enterprise only, through a signed BAA that auto-enables Zero Data Retention. If you are in healthcare, that is an Enterprise conversation, not a self-serve signup.
Intercom Fin
Fin is the most serious alternative here and the only other platform that publishes a headline rate: from $0.99 per Fin outcome, across all three plan tiers. Seats are $29 per month on Essential, $85 on Advanced and $132 on Expert.
The appeal is that outcome pricing is easy to reason about. You pay when the AI resolves something, and the number scales with the volume you actually deflect rather than with headcount. For a team already inside Intercom’s ecosystem, adopting Fin is close to frictionless.
Two things to weigh. Intercom’s pricing page does not define what constitutes a “Fin outcome”, which means the unit that determines your entire bill is set by the vendor and not published. And the per-seat cost sits on top, so a small team on Expert is paying real money before a single conversation is resolved.
Zendesk AI agents
Zendesk publishes half of what you need. Suite Team is $55 per agent per month on annual billing and Suite Professional is $115, with Enterprise on quote. AI agents are included on every Suite and Support plan, priced on “successful outcomes”, measured in Automated Resolutions and defined as requests resolved without escalation to a human.
That definition is clearer than Intercom’s. The rate is not published, which puts Zendesk in the same position as the demo-only vendors on the number that matters most.
The case for Zendesk is that you are probably already in it. If your tickets, macros and reporting all live there, adding its AI agents is the path of least resistance, and least resistance has real value. The case against is that you are buying a support suite that added AI, rather than an agent platform, and the seat cost dominates your bill either way.
Decagon
Decagon is the reference enterprise vendor in this category and is generally where large evaluations end up alongside Sierra. It is built for complex deployments with heavy configuration and a managed relationship.
There is no public pricing. Third-party blogs circulate platform fees and median contract values, and almost all of those blogs are published by competitors, so I am not going to repeat their numbers as fact. What is verifiable is that Decagon does not publish a rate and requires a sales conversation.
If you have a procurement process, a security review and a committee, that is normal and Decagon should be on your list. If you were hoping to try something this week, it is not a candidate.
Sierra
Sierra grew quickly and takes a managed-service posture: the vendor is closely involved in building and tuning the agent rather than handing you a builder. For enterprises that want an outcome rather than a tool, that is a genuine model and some teams strongly prefer it.
Like Decagon, Sierra publishes no pricing and no independently verified resolution rate. Everything you will read about what it costs comes from someone with an interest in the answer.
Gladly Sidekick
Gladly’s distinctive idea is worth taking seriously, and it is architectural rather than marketing: the platform organises around a unified customer timeline rather than around tickets. One continuous conversation per person, across channels, instead of a queue of disconnected cases. For brands where the same customer contacts you repeatedly and context is the whole problem, that shape genuinely fits better than a ticketing model.
Sidekick is the AI agent layer on top of it. Gladly publishes no pricing for any of it. Its site routes every path to a demo request, and its pricing URL now redirects to the homepage.
I would look hard at Gladly if you are a consumer brand with high repeat contact and you have the patience for a sales cycle. The timeline concept is the most interesting non-pricing differentiator in this comparison. You just cannot evaluate the cost of it without a call.
Salesforce Agentforce and Ada
Both belong in a longer list rather than a shortlist for most teams.
Agentforce makes sense when Salesforce is already your system of record and the agent needs to act inside it. The tradeoff is implementation weight: this is a platform project with a timeline, not a signup.
Ada sits in the mid-market and has been in this category longer than most. Its published material does not give a rate you can plan against either.
Side by side
| Published price | Pricing unit | Free tier | Model choice | Time to live | |
|---|---|---|---|---|---|
| Chatbase | Full price list | Message credits | Yes, $0, no card | OpenAI, Anthropic, Google, xAI, Mistral and others | Days |
| Intercom Fin | $0.99 per outcome, seats $29 to $132 | Outcome, undefined | No | Vendor decides | Days |
| Zendesk | Seats $55 to $115, AI rate not published | Automated Resolution | No | Vendor decides | Days to weeks |
| Decagon | None | Not published | No | Vendor decides | Procurement cycle |
| Sierra | None | Not published | No | Vendor decides | Procurement cycle |
| Gladly Sidekick | None | Not published | No | Vendor decides | Sales cycle |
| Salesforce Agentforce | Partial | Per conversation | No | Vendor decides | Implementation project |
Which should you pick
A small support team that wants something live this month. Chatbase. Start on the free plan, prove the agent answers your real questions, then move to Standard at $150 when you need Helpdesk and voice. Nothing else here lets you validate before you buy.
A team already living inside Intercom. Fin. The integration advantage is real and the $0.99 rate is public. Model it against your deflection volume and check what Intercom counts as an outcome before you commit.
A team already living inside Zendesk. Zendesk AI agents, for the same reason, with the same caveat that you will need a sales conversation to learn the resolution rate.
A consumer brand with high repeat contact. Put Gladly on the list for the timeline model, and budget for a sales cycle.
An enterprise with procurement, security review and a committee. Decagon and Sierra. You are buying a relationship rather than a product, which is appropriate at that scale, and you will not shortcut the process.
Healthcare or anything with a BAA requirement. Skip the self-serve tiers everywhere. On Chatbase specifically that means Enterprise, where HIPAA eligibility comes through a signed BAA with Zero Data Retention enabled.
What to test before you sign anything
Four questions cut through most of the marketing in this category.
Can it take actions in your systems, or only answer questions? An agent that cannot change a subscription or look up an order cannot close the tickets that cost you the most.
What does failure look like? Every one of these will be wrong sometimes. The thing to evaluate is the handoff: does a human get the conversation with full context, or does the customer start again?
Which model is behind it, and can you change it? This is the question that ages best. Model quality moves every few months.
Is the compliance posture you need on your plan, or only on Enterprise? SOC 2 and GDPR are widely available. BAAs, SSO and data residency usually are not.
Answer those four honestly against your own ticket data and the shortlist tends to pick itself.
Verdict
For a team that wants an AI support agent live this month without a procurement cycle, Chatbase is the pick: it publishes its full price list, starts free, and reaches Helpdesk, voice and Outbound at $150 a month. Intercom Fin is the strongest alternative and the only other platform that publishes a headline rate, at $0.99 per outcome. If you are running an enterprise evaluation with a committee and a security review, Decagon and Sierra belong on the list, but you will not learn what they cost until you talk to sales.
Frequently asked questions
What is the difference between an AI chatbot and an AI agent?
A chatbot matches questions to answers. An agent takes actions: looking up an order, changing a subscription, booking a slot, escalating with context attached. The distinction matters commercially, because a platform that only answers questions cannot close a ticket that requires doing something.
Which of these platforms publish their pricing?
Two publish a usable rate. Chatbase lists every plan and price on its pricing page. Intercom publishes Fin at $0.99 per outcome plus seat costs. Zendesk publishes seat prices ($55 and $115 per agent per month on annual billing) but not its per-resolution rate. Gladly, Decagon and Sierra publish nothing and route you to a demo.
How much does Chatbase cost?
Free at $0 for 50 message credits and one agent, Hobby at $40 for 700 credits, Standard at $150 for 4,000 credits with Helpdesk, voice and Outbound, Pro at $500 for 15,000 credits, and Enterprise on quote. Annual billing takes 20% off. Extra agents are $25 each per month and removing Chatbase branding is $99 a month.
Is a per-resolution price better than a credit allowance?
It depends on volume and on how much you trust the vendor’s definition of a resolution. Per-resolution scales linearly and is easy to forecast, and none of these vendors publish a precise definition of what counts. A fixed credit allowance is cheaper below its ceiling and gets expensive above it. Model both against your actual ticket volume before signing.
What should I check before buying any of them?
Four things. Whether the agent can take actions in your systems or only answer questions. What the handoff to a human looks like when it fails. Which model sits behind it and whether you can change that. And whether the compliance posture you need is on your plan rather than only on Enterprise.
Sources
- Chatbase pricing (all plans and add-on costs)
- Chatbase customer results: Jumia
- Intercom pricing (Fin per outcome, seat tiers)
- Zendesk pricing (Suite tiers, automated resolutions)
- Gladly (no published pricing, demo request only)
- Decagon (no published pricing)
- Sierra (no published pricing)
- Salesforce Agentforce
- Ada
- AICPA on SOC 2 reporting
Written with AI assistance and reviewed against our editorial standards.
