Best AI customer support agents in 2026: prices per resolution and per seat
Intercom Fin is worth considering for support teams that want a published per-outcome price and support for an existing helpdesk. It publishes a per-outcome price ($0.99), a small monthly minimum, and says it works on other helpdesks. Zendesk AI agents and HubSpot’s Customer agent also publish per-resolution prices, while Salesforce, Ada, Sierra and Decagon either quote by conversation or sell through a sales call. If what you need is a small-team assistant for email, texts, calls and Slack rather than a ticket queue, that is a different job, and OpenAssistant covers it in its own section below.
Options at a glance
We ordered the seven support-desk products against the same rubric: whether the documented capabilities cover the job, the vendor-published evidence and its limitations, price for what you get, where it works, and privacy, security and admin controls. Products that publish a price and say how a resolution is counted score better on price than products that say “get a demo”. Vendor resolution rates are vendor claims and did not move the order. This is an editorial judgment, not a score from use.
| Rank | Product | Best for | How it is priced (published) |
|---|---|---|---|
| 1 | Intercom Fin | Teams that want a published per-outcome price, on any helpdesk | $0.99 per service resolution; 50-outcome monthly minimum (about $49.50) on an existing helpdesk |
| 2 | Zendesk AI agents | Teams already on Zendesk, wide language coverage | $1.50 committed or $2.00 pay-as-you-go per automated resolution, plus agent seats |
| 3 | HubSpot Customer agent | HubSpot customers who want the lowest published unit rate | $0.50 per resolution; needs a Professional or Enterprise edition |
| 4 | Salesforce Agentforce for Service | Large teams already on Salesforce | $2 per conversation, or Flex Credits |
| 5 | Ada | Enterprise teams that want certifications and answer checks | Not stated; demo only |
| 6 | Sierra | Enterprise teams that want one agent on many channels | “Outcome-based”, no numbers published |
| 7 | Decagon | Teams that want workflow rules in plain language | Not stated; demo only |
| Pick for a different job | OpenAssistant | A small team that wants one assistant working across email, texts, calls and Slack, with approvals set per action | Free Starter plan; Pro $72 monthly or $60 annual per user (per-user pricing confirmed by 3Gen) |
Support-desk products are built for ticket volume. OpenAssistant is a general work assistant, so the specialists above win for running a support queue.
How AI support agents are priced, with a worked example
The vendor pages show four pricing models.
- Per resolution or outcome. The billable unit depends on the vendor; it can include a resolution or a configured handoff, not only a settled conversation. Fin charges $0.99 per service resolution, HubSpot $0.50 per resolution, Zendesk $1.50 or $2.00 per automated resolution, Salesforce $2 per conversation.
- Per seat. You pay for each human agent who uses the helpdesk. Zendesk Support Team is $19 per agent per month paid yearly. Intercom Essential displays a $19-per-seat monthly new-customer offer against $29 list; check eligibility and billing terms. HubSpot Service Hub Professional is from $90 per seat per month.
- Platform fee or minimum. Fin on an existing helpdesk has a monthly minimum of 50 outcomes (about $49.50). HubSpot requires a one-time $1,500 onboarding fee on Professional.
- Quote only. Ada, Sierra and Decagon publish no price, minimum or contract term on the pages we read.
The unit prices are not like for like, because each vendor counts a “resolution” differently. HubSpot counts a conversation as resolved when it was not handed to a human rep for 72 hours. Zendesk says a messaging conversation ends 2 hours after the first message, and that an LLM verifies the resolution. Fin also bills configured Procedure handoffs; its sales-qualification outcome has a separate $9.99 rate.
Worked example. Assume a team of 3 human agents and 1,000 conversations a month that the AI resolves. The volume and team size are our assumptions. Every price is from the vendor pages linked in the product sections.
| Product | Resolution fees | Seats (3) | Monthly total |
|---|---|---|---|
| Fin, on your existing helpdesk | 1,000 x $0.99 = $990.00 | Your current helpdesk, not counted | $990.00 |
| Fin with Intercom Essential (displayed new-customer offer) | 1,000 x $0.99 = $990.00 | 3 x $19 = $57 | $1,047.00 |
| HubSpot Customer agent | 1,000 x $0.50 = $500.00 | 3 x $90 = $270 | $770.00, plus $1,500 one-time onboarding |
| Zendesk, committed rate | (1,000 - 15 included) x $1.50 = $1,477.50 | 3 x $19 = $57 | $1,534.50 |
| Zendesk, pay-as-you-go | (1,000 - 15 included) x $2.00 = $1,970.00 | 3 x $19 = $57 | $2,027.00 |
| Salesforce, per conversation | 1,000 x $2 = $2,000 | Licence prices not stated on the pages we read | $2,000 plus licences |
The Intercom row uses the displayed $19 new-customer offer, not its $29 list price. At $29, three seats are $87 and the same example totals $1,077. Confirm offer eligibility and billing terms (Intercom pricing).
Notes on the arithmetic. Zendesk’s page lists 5 included resolutions per agent per month on Support Team, so 3 agents give 15. HubSpot’s $90 is a “from” price. Salesforce also sells Flex Credits at $500 per 100,000, and an action costs 20 credits, which is $0.10 (20 x $500 / 100,000). The pages we read do not say how many actions one conversation uses, so we do not convert the two.
The minimum matters at low volume. At 20 outcomes a month Fin’s usage would be 20 x $0.99 = $19.80, which is below the 50-outcome minimum (about $49.50). HubSpot’s cheaper unit rate does not remove the seat cost: 20 x $0.50 = $10, plus the same $270 of seats. For more on how agent vendors bill, see AI agent pricing explained.
1. Intercom Fin
Fin’s home page headline calls it “The highest performing Customer Agent”, a vendor claim we have not verified (Fin). The vendor lists voice, chat, email, Slack and social channels, and says it works with other helpdesks including Freshdesk, Salesforce and Zendesk (Fin pricing; Fin channels).
Why it ranks first. It publishes the price and the counting rules, it runs on other helpdesks, and its trust page documents escalation, approvals and pre-launch simulations. The page says Fin “escalates to a human agent rather than guessing”, lets you run simulations of customer scenarios before go-live, and supports approvals in procedures (Fin trust page). It lists SOC 2 Type II, ISO 27001, ISO 27701, ISO 27018 and ISO 42001, hosting in the US, EU or Australia, a 99.8% SLA, and SSO with Okta, Azure AD and OneLogin.
Price. Resolutions, configured Procedure handoffs, disqualifications and self-serve routing cost $0.99 each; sales qualifications cost $9.99. One billable outcome is charged per conversation. A simple pass-through to a person without an outcome is uncharged. Existing-helpdesk use lists a 50-outcome monthly minimum (about $49.50) and no extra seat or setup fees. Voice requires a custom sales quote; a 14-day trial is listed (Fin pricing). Intercom Essential displays a $19-per-seat monthly new-customer offer against a $29 list price; confirm eligibility and billing terms (Intercom pricing).
Vendor claim. An average resolution rate of 76% across 12,000+ customers (Fin). We have not verified it.
Watch out for.
- A configured Procedure handoff is billable; a simple pass-through without an outcome is not. Confirm which workflow your team uses.
- Fin uses anonymized customer data for fine-tuning, with an opt-out. Third-party model providers are restricted from training on your data (Fin trust page).
- Fin sales qualifications cost $9.99 each; use the $0.99 rate only for the outcome categories listed above. The resolution-only example excludes qualifications and Voice.
2. Zendesk AI agents
Zendesk calls its product “self-improving AI agents built for resolution” and says they work across social, web, mobile, voice and email, support 80 languages, and “work on any platform” (Zendesk AI agents).
Why it ranks second. The price is published and tied to results, the trust center lists a long set of certifications, and the product sits inside the helpdesk many teams already run. The trust center names ISO 27001:2022, ISO 42001 and FedRAMP Low Impact, and says SOC 2 Type II reports are available under NDA (Zendesk trust center). A HIPAA Business Associate Agreement is available.
Price. The pricing page says it is “based on the successful outcomes they deliver”. Automated resolutions are $1.50 each committed or $2.00 pay-as-you-go. Support Team is $19 per agent per month paid yearly ($25 monthly), Suite Team $55 ($69 monthly) and Suite Professional $115 ($149 monthly). Included resolutions are 5 per agent per month on Support Team and Suite Team and 10 on Suite Professional, capped at 10,000 a year. There is a 14-day free trial (Zendesk pricing).
Vendor claim. “Resolve up to 80%+” of service issues (Zendesk pricing). We have not verified it.
Watch out for.
- The AI agents page says resolutions are priced by value tier, introduced 18 May 2026, while the pricing page shows flat $1.50 and $2.00 figures. We follow the pricing page for the numbers. The dollar value of each tier is not stated (help center).
- Voice agents require Zendesk voice, and QA requires an add-on.
- Suite Growth and Suite Enterprise prices, data-training terms for AI agents, and any price for Forethought-based agents are not stated.
3. HubSpot Customer agent
HubSpot says its Customer agent “hands off to your team with full context” and lists voice, email, chat, forms, WhatsApp and Facebook Messenger (HubSpot Customer agent). The product page now sits under the name Agent Hub.
Why it ranks third. It has the lowest published unit rate here and a clear definition of a resolution, and it fits a team already on HubSpot. It ranks below Fin and Zendesk because it needs a Professional or Enterprise edition, the page lists fewer controls for support specifically, and the pricing page cites a different resolution rate than the product page.
Price. “$0.50 per resolution”, or $0.50 per minute for voice. A resolution means the conversation was not handed to a human rep for 72 hours, and a resolution can also occur if the agent qualifies a lead. In credits it is 50 credits per resolution at $0.010 per credit (HubSpot product catalog). Service Hub seats are Starter from $7, Professional from $90 and Enterprise from $150 per seat per month, with required one-time onboarding of $1,500 on Professional and $3,500 on Enterprise. The pricing page offers 14 days of free Customer agent access when you buy at least one Service Hub Professional or Enterprise seat (HubSpot Service pricing).
Security. HubSpot’s AI pages cite SOC 2 Type 2 audits and say third-party AI providers are contractually prohibited from training on your data (HubSpot AI page).
Vendor claim. A 70% average resolution rate, with top teams at 90%; the pricing page says “65%+”. We have not verified either.
Watch out for.
- Professional or Enterprise editions are required for Customer agent, even though Agent Hub is described as included in all editions.
- Unused credits do not roll over, and some features add telephony or SMS charges.
4. Salesforce Agentforce for Service
Salesforce says Agentforce lets you “scale service without scaling headcount”, deploys across voice, web, portals, SMS and WhatsApp, and passes complete customer context to your service team on handoff (Salesforce Service AI).
Why it ranks fourth. Capability and reach are strong for a team already on Salesforce, and the Trust Layer page says prompts and responses are never stored or used to train third-party models, with zero data retention at the model provider (Salesforce trusted AI). It ranks below the first three on price clarity: three buying routes, and certifications are not stated on the pages we read.
Price. Conversations are $2 each, for customer-facing agents only. Flex Credits are $500 per 100,000 credits, 20 credits per standard action and 30 per voice action. The Agentforce add-on is $125 per user per month for unmetered employee use, which is not a customer-agent price. There is no overage penalty, and unused Flex Credits do not roll over. A Service Cloud free trial runs 30 days (Salesforce pricing).
Vendor claim. “Agentforce resolves 85% of Salesforce’s customer service requests” (Salesforce Service AI). This is Salesforce’s own number for its own support.
Watch out for.
- Flex Credits and Conversations cannot be mixed in the same org.
- The pricing page is “for information purposes only and is subject to change”.
- Service Cloud seat prices are not stated on the pages we read, so the total cost of a team is not shown here.
5. Ada
Ada describes “AI customer service agents that resolve, act, and continuously improve” and lists voice, email, chat, Messenger, WhatsApp, SMS, Instagram and in-app channels (Ada; Ada platform).
Why it ranks fifth. The control and security story is clear: the trust page says AI agents verify responses against approved sources, holds SOC 2 Type II and HIPAA, says customer data is never used to train models, and says zero data retention is enforced with all LLM providers (Ada trust). MCP tools stay off until you turn them on. It loses places because there is no published price.
Price. Not stated. The pricing path leads to a demo booking page that says Ada will reach out within one business day to schedule a 30-minute call (Ada demo). No minimum, contract term or trial is published.
Vendor claim. Ada’s pages give several automated-resolution figures: 84% and “over 80%” on the platform page, and “77% up from 32%” for one customer on the demo page. The figures differ and we have not verified them.
Watch out for. You cannot estimate cost without a sales call.
6. Sierra
Sierra calls itself “the leading conversational AI platform for businesses” and says one agent deploys across chat, SMS, WhatsApp, email, voice and ChatGPT (Sierra; channels).
Why it ranks sixth. Channel breadth and the security list are strong: SOC 2, HIPAA, GDPR, PCI, FedRAMP High, CCPA, CSA STAR, ISO 27001 and ISO 42001, plus “supervisory layers” around the language models (Sierra trust). It ranks below Ada because the pages we read do not describe human handoff or escalation rules, and no price is published.
Price. Described as “you pay only when the software achieves specific, valuable outcomes” (Sierra product). No number, minimum, contract term or trial is published, and the pricing page returned an error.
Watch out for. Handoff rules are not stated on the pages we read. The Sierra trust center did not render, so its contents are not covered here.
7. Decagon
Decagon’s tagline is “the AI concierge for every customer”. It unifies chat, voice and email, supports 70+ languages on voice, and transfers calls to human agents with a summary (Decagon; voice).
Why it ranks seventh. Controls are well described: Agent Operating Procedures let you define workflows in natural language (Decagon AOP), an always-on QA system reviews every conversation, a supervisor checks for hallucinated answers before sending, and zero data retention is enforced with AI providers (Decagon security). It ranks last because no price is published and no compliance certification is named on the pages we read.
Price. Not stated. The calls to action are “Get a demo”, and no minimum, contract term or trial is published.
Vendor claim. A customer result of “70% chat and voice resolution” and an 80% deflection rate. We have not verified them.
Watch out for. Ask for the certification list in writing, since none is named publicly. The Decagon trust center did not render.
OpenAssistant: a general assistant, not a support desk
Support-desk products are built for ticket volume. OpenAssistant is an AI assistant that acts for a person or small team, and the pages we read describe no support-desk or ticket-queue use case (OpenAssistant). If you run a real queue, pick from the seven above.
It is our recommendation for a small team that wants one assistant working through email, texts, calls and Slack, with approvals before it acts. The page lists calls, iMessage, SMS, WhatsApp, email and a web app, with its own inbox, phone number and calendar (OpenAssistant). Slack works in DMs and in channels once added, and the assistant only responds when invoked (OpenAssistant on Slack). It also “chases replies, reminds people of what they owe you”.
Approvals are set per action: Manual, Confirmation Window or Autonomous, and a Confirmation Window is not Manual approval (autonomy settings). The changelog says that when the assistant is on a call and needs you urgently, notes reach you in your own text chat (changelog). The trust page states a SOC 2 Type II audit and that it never trains on your data (OpenAssistant trust).
Price. The Starter plan is free: 1,000 credits at signup, then 100 bonus credits a day once used up, up to 1,000 a month. Pro is $72 per user per month, or $60 annual, at the 5,000-credit tier; Team is $120 or $100 annual (OpenAssistant pricing; per-user pricing confirmed by 3Gen). Simple tasks use about 10 to 50 credits, and plan credits do not roll over (credit terms).
Watch out for.
- It does not take part in group texts or group WhatsApp chats (ways to work), and not every country is supported for texting (phone setup).
- It does not work in Slack Connect channels with external companies (Slack help). For people outside your company, the help center points to email.
- Credits are consumed by tasks, so cost per customer conversation depends on the work done. There is no per-resolution price.
For a wider comparison of general-purpose agents, see best AI agents in 2026 and best AI agents for small business.
What to ask each vendor
- How exactly is a resolution or outcome counted, and what happens to the bill if a customer reopens the conversation or asks for a human?
- What is the minimum monthly spend, the contract term, and the overage rate at your volume?
- Does the price include voice, SMS and WhatsApp, or are those charged separately?
- Which of your customer data is used to train models, and can you opt out in writing?
- How does the agent handle prompt injection, where a customer message or a web page tries to give the agent new instructions? What can it do on your systems if it is fooled?
- Which actions (refunds, account changes) need a human approval, and can you lock that setting for the whole team?
- How do you hand off to a person, and does the person see the full conversation?
What to keep for humans
- Refunds, credits, cancellations and anything involving payment details.
- Complaints, legal threats, safety issues, and customers who ask for a person.
- Account changes that cannot be undone.
- Reviewing a sample of AI-resolved conversations each week, including the ones the customer never replied to.
- Any rule about what the AI may promise: delivery dates, discounts, policy exceptions.
Related reading
- Best AI sales agents in 2026: AI SDRs and sales assistants compared: 11x, Artisan, HubSpot, Agentforce and Clay compared on prices, approval controls and deliverability questions.
- Best enterprise AI agents in 2026: governance, pricing and platforms: Enterprise agent platforms compared on governance, audit logs and price.
- AI agent pricing in 2026: credits, seats and per-resolution fees explained: Credit plans, tasks, per-resolution fees, subscriptions and enterprise fees, with published prices and a worked example.
- How to choose an AI agent: 10 questions to ask before you buy: Ten questions on job fit, channels, approvals, price, data and exit, with a job-to-product table.
- Are AI agents safe? Prompt injection, permissions and approval settings: Real risks from prompt injection to excess access, the controls to look for and a 10-question checklist.
Frequently asked questions
What is the difference between per-resolution and per-seat pricing?
Per-resolution pricing charges you each time the AI settles a customer conversation, for example $0.99 per service resolution at Fin. Per-seat pricing charges for each human agent who uses the helpdesk, for example $19 per agent per month at Zendesk Support Team. Most vendors in this article combine both.
How much does an AI support agent cost per resolution?
The published rates are $0.50 at HubSpot, $0.99 at Fin, $1.50 or $2.00 at Zendesk and $2 per conversation at Salesforce. The vendors count a resolution differently, so compare the definitions as well as the numbers. Ada, Sierra and Decagon publish no price.
Can OpenAssistant replace a help desk?
No. The pages we read describe no support-desk or ticket-queue use case, and it is priced in credits rather than per resolution. It fits a small team that wants an assistant on email, texts, calls and Slack.
Do these vendors train on my customers’ data?
It varies. Fin uses anonymized data for fine-tuning with an opt-out. HubSpot says third-party AI providers are prohibited from training on your data. Ada says customer data is never used to train models. Decagon enforces zero data retention with AI providers. Zendesk’s data-training terms for AI agents are not stated on the pages we read.
Which of these products can I try without talking to sales?
Fin (14 days), Zendesk (14 days, though its pricing page says AI trial features may vary), HubSpot’s Customer agent (14 days, but only after you buy at least one Service Hub Professional or Enterprise seat) and Salesforce Service Cloud (30 days) list a free trial. Ada, Sierra and Decagon publish no trial.
Try OpenAssistant
If your need is an assistant for a small team across email, texts, calls and Slack, start with the free Starter plan at openassistant.us. You can see the current plans on the pricing page. For a ticket queue, start with the specialist trials above.