If you are evaluating AI for your support team this year, you probably already run some. A bot in the help desk answers password questions, deflects a share of chat, and reports a healthy number on its dashboard. The queue has not moved much, because the tickets that eat agent hours were never questions. “Refund my last charge and cancel my plan” needs someone to open the billing system, check eligibility, issue the refund, update the customer record and write back. An AI that can only reply turns that ticket into a polite handoff.
That gap is the most useful way to read this market going into 2027. The products below look alike in a demo, where each one answers a help-center question in fluent English. They differ in what happens after the answer, in who builds and maintains the automation, and in what you get billed for. Those three things decide whether a tool fits your operation, so this guide covers them first and the products second.
What actually separates these tools
Answering versus acting
Every product here can read a ticket and write a reply from your help center. Fewer can complete the request: find the charge in Stripe, check it against your refund policy, issue the refund, update the CRM and close the ticket. Look at your own ticket mix before you watch any demo. If most of your volume is “how do I” questions, a good answering agent on a well-kept knowledge base will carry you a long way. If most of it is refunds, cancellations, billing disputes and order changes, ask each vendor to show a write action in a system like yours, and ask who builds that integration.
Who builds it and who keeps it working
Self-serve tools hand you a builder. That is fast and cheap when someone on your team owns it, tunes it every week and can get engineering time for API work. Enterprise platforms assign a deployment team and expect a project that runs for weeks or months. Managed services build and operate the automation for you and give you reporting. The trade is control for effort: the less you configure yourself, the more you depend on the vendor’s speed when a policy changes.
A related question is where the AI lives. Some agents only work inside their vendor’s help desk. Others sit on top of the one you already run, which matters a lot if a migration is off the table.
What you are billed for
Per resolution, per outcome, per engaged conversation and per automated interaction sound alike and are not. A vendor that bills per conversation gets paid when the AI replies, whether or not the customer got what they needed. Per-resolution vendors each define “resolved” themselves, often as “the customer did not come back within 72 hours”. Before you compare rates, get the definition in writing and ask what happens to the charge when a ticket reopens. Then add what sits outside the meter: seats, platform fees, implementation.
The tools
The eight products are ordered by how much of the work the vendor takes on, from the most vendor-run to the most self-serve. The order is not a ranking. Prices are the vendors’ published rates as of early October 2026.
Soportio
Best for: high-volume support teams in e-commerce, SaaS and subscription apps, and mobile games whose queue is mostly refunds, cancellations, billing and order requests, and who do not want to build the automation themselves.
Soportio is an AI resolution engine for high-volume customer support, delivered as a fully managed service: it executes refunds, cancellations, billing and order requests in your existing systems and closes the ticket. For each ticket it applies your business rules, acts through APIs in your billing, CRM and help desk, and either closes the ticket or escalates it to your team with the context attached. It works on top of Zendesk, Intercom, Freshdesk or Zoho Desk, connects to Stripe, Paddle, Solidgate, Salesforce or any in-house system with an API, and handles tickets across email, chat, web forms and messaging channels such as WhatsApp, Instagram and Telegram. Soportio’s team builds each workflow, validates it on real tickets and maintains it after launch. Financial or sensitive actions can be held for human approval. Data is exchanged only through secured APIs over encrypted protocols, with no third-party connectors or intermediaries that can access it. The company says a first workflow can go live about two weeks after you provide access.
The trade-off is less hands-on control in exchange for less work on your side. You set the rules and approval steps, Soportio’s team builds and updates the workflows, and you track results in analytics. There is no self-serve builder or trial, and terms are agreed case by case. It is built for repetitive, high-volume requests, so a small team with mostly one-off technical questions will get less from it. Soportio cites up to 90% of tickets automated; the share for your team depends on your ticket mix.
Pricing: per resolved ticket, from $0.90 at under 5,000 a month down to $0.40 at 50,000 to 99,999, with custom rates above that. No implementation fee, no seat fee, and no charge for tickets escalated to your team. Soportio counts a ticket as resolved when its automation has completed the request and closed the ticket without handing it over to your team; if the customer writes back in the same ticket within a set period, it stays one ticket and is not billed again, and only a separate new ticket is billed as a new one.
Sierra
Best for: large consumer brands that want a custom, branded agent, voice included, built with the vendor.
Sierra builds agents with you in a vendor-led deployment, and the result is tailored closely to your brand, policies and systems. It prices on outcomes, so the bill is tied to conversations the agent actually closes.
Rates are not published and contracts are enterprise-sized, with a procurement cycle to match. A bespoke build also means changes run through a joint team, so plan for how quickly a new policy can reach production.
Pricing: outcome-based, by quote.
Decagon
Best for: mid-market and enterprise teams with engineering support that want precise control over agent behavior.
Decagon is a standalone platform that sits above your help desk and covers chat, email and voice. Its distinguishing idea is that operators write the agent’s behavior as plain-English operating procedures, which makes it easier to see and change exactly what the agent does in a given case, including the actions it takes in other systems.
Setup expects engineering involvement up front. There is no pricing page and no trial, and the entry cost puts it out of reach for small teams. Below roughly ten thousand conversations a month the economics are hard to justify.
Pricing: by quote. Decagon has described both per-conversation and per-resolution options.
Ada
Best for: large enterprises whose CX team wants to run an AI agent across chat, email and voice without writing code.
Ada is an AI agent platform with a no-code builder, aimed at non-technical operators. It connects to the major help desks and runs its own layer across channels, so one team can manage the agent’s behavior everywhere customers reach you.
Ada publishes no pricing and offers no self-serve trial. Third-party reviews report implementations of two to four months, so it suits a planned program more than a quick fix. Ask which unit your contract meters, since Ada has argued publicly for conversation-based billing over per-resolution.
Pricing: by quote, on an annual contract.
Kustomer
Best for: B2C teams that want CRM and help desk in one customer record.
Kustomer organizes support around a customer timeline instead of separate tickets, so an agent, human or AI, sees orders, past conversations and account data in one view. It sells AI agents for customers and a separate copilot for reps.
Choosing it means adopting Kustomer as your help desk. Its customer-facing AI agent, Concierge, is billed per engaged conversation, meaning any conversation in which the AI replies, resolved or not. If a large share of those conversations ends in escalation, the effective cost per resolved ticket will be well above the headline rate. Seat prices are no longer published, so the platform cost comes from a quote.
Pricing: seats by quote. Concierge is $0.75 per engaged conversation at pay-as-you-go rates, and AI agents for reps are a separately quoted add-on.
Zendesk
Best for: teams already running Zendesk that want a single vendor for tickets and AI.
Zendesk’s AI agents live inside the Suite, next to routing, QA and workforce tools your team may already use. Since May 2026 only verified resolutions draw on your allowance, which is a fairer unit than most. Zendesk also completed its acquisition of Forethought in March 2026 and now sells Forethought AI agents as a separate add-on, pitched as working on other service platforms too.
The cost stacks in layers: seats, the Copilot add-on for agents, then resolutions. The per-resolution rate is the highest published one in this list. You also have two AI agent lines to choose between, and the Forethought one is priced only through sales.
Pricing: $1.50 per automated resolution on a committed volume, $2.00 pay-as-you-go, above a small allowance per seat. Suite seats run from $55 to $115 per agent per month on annual billing.
Gorgias
Best for: Shopify stores that want support and order data in one place.
Gorgias is an ecommerce help desk with an AI agent built in. Because it reads Shopify order data natively, it handles the requests that make up most of a store’s queue (order tracking, returns, cancellations) without integration work, across email, chat, SMS and social channels.
The AI agent is tied to a connected Shopify store, so merchants on other carts should look elsewhere. Billing has two meters: each automated interaction is charged at the AI rate and also counts as a ticket on your help desk plan. Model both before you commit.
Pricing: $0.90 per automated interaction on annual plans, $1.00 on monthly, and $1.50 beyond the plan’s allowance, on top of a ticket-based help desk plan.
Fin (Intercom)
Best for: teams that want to switch on an AI agent themselves and see results this month.
Fin is Intercom’s AI agent. It runs inside the Intercom help desk or on top of another one, including Zendesk, Salesforce, HubSpot and Freshdesk, so you can try it without migrating. Setup is self-serve, the price is public, and multi-step requests are handled through procedures and data connectors that your team configures.
The work of building those procedures is yours, and so is maintaining them when policies change. Read the billing definition closely: Fin charges per outcome, and an outcome covers more than a confirmed resolution. It also includes a handoff made as part of a procedure.
Pricing: $0.99 per outcome, with a minimum of 50 outcomes a month. No seat fee when Fin runs on another help desk. Inside Intercom, seats are extra.
Side by side
| Tool | Best for | Runs on | Who builds it | Billed per | Published rate |
| Soportio | E-commerce, SaaS, subscription apps, mobile games | Your help desk | Vendor | Resolved ticket | $0.40 to $0.90 |
| Sierra | Bespoke enterprise agent | Your stack | Vendor with you | Outcome | Not published |
| Decagon | Engineer-led control | Your help desk | You, with engineers | Conversation or resolution | Not published |
| Ada | Enterprise, no-code | Your help desk | You, with vendor onboarding | Conversation (resolution on request) | Not published |
| Kustomer | CRM and help desk in one | Kustomer | You | Engaged conversation, plus seats | $0.75 |
| Zendesk | Existing Zendesk teams | Zendesk | You | Automated resolution, plus seats | $1.50 to $2.00 |
| Gorgias | Shopify stores | Gorgias | You | Automated interaction, plus ticket | $0.90 to $1.50 |
| Fin | Self-serve start | Intercom or your help desk | You | Outcome (plus seats on Intercom) | $0.99 |
Rates as published by each vendor, checked October 2026. Billing units differ: an engaged conversation is charged even when it is handed to a human, while a resolved ticket or automated resolution is charged only when the AI completes the request. Fin’s outcome also covers handoffs made inside a configured procedure.
The rates are not directly comparable, because each row meters a different event. Use the table to see which model you are buying, then price it against your own volume.
How to shortlist
What share of your tickets needs an action in another system? Tag last month’s tickets by request type. If the top five are informational, an answering agent such as Fin, or the one built into your current help desk, is the cheapest route. If they are refunds, cancellations and billing changes, shortlist the tools that execute actions and ask each one who builds the integration: you (Fin, Decagon) or the vendor (Sierra, Soportio).
Who on your team will own it? A support operations lead with spare hours can run Fin, Gorgias or Zendesk’s agents. Add engineering time and Decagon becomes realistic. If nobody has the capacity, an unowned self-serve tool will stall at its first-week resolution rate, and a vendor-run option is the safer choice.
Are you willing to change help desks? If not, Kustomer and Gorgias are out unless you already use them, and Zendesk’s native agents only make sense for Zendesk customers. Everything else in this list works on top of what you have.
What does the bill look like at your real volume? Take last month’s ticket count, apply each vendor’s own definition of the billable event, and add seats and platform fees. At a few thousand tickets a month, contract-based vendors rarely pay off. At tens of thousands, a few cents of difference per ticket and the resolution definition matter more than anything on the feature page.
Then run one test before signing anything. Pick your largest request type and ask each shortlisted vendor to show the full path for it, from incoming ticket to closed ticket, in a system like yours.
- Best AI Customer Service Software for 2027 - October 31, 2026
- How to Install Pandas in VS Code (pip, venv and Conda) - October 30, 2026
- Top 10 UX and Branding Strategies for SaaS Startups in 2026 - October 30, 2026
