How to Choose an AI Support Agent (2026)
Last updated: November 9, 2026
Answer in brief: Choose an AI support agent by what it can finish, not what it can say. First, read exactly what the vendor counts and bills as a resolution. Second, confirm the agent can take the actions that close your top ticket types. Third, check the guardrails: confidence thresholds, handoff rules, identity checks and spending limits. Fourth, demand testing on your own conversations before launch. Fifth, model the cost at your real volume. Sixth, match the agent to the help desk you'll keep. Agents that clear all six tend to resolve tickets end to end, and they're the ones worth paying for.
Top pick for most teams: Intercom Fin. A public $0.99 per outcome, a 14-day trial with no card, simulation testing before launch, and it runs on Intercom, Zendesk, Salesforce, HubSpot or Freshdesk.
This guide is independent research from vendors' own pricing pages and documentation, read November 9, 2026. We did not test these products hands-on. For the ranking itself, see the best AI support agents list. Pricing worked examples are on our pricing page.
Step 1: Know what counts as resolved and billed
Every vendor's billing unit is a different definition of success. This is the single biggest driver of cost, and of how much you can trust the vendor's resolution rate.
| Agent | Billing unit | Billable when | Public price |
|---|---|---|---|
| Intercom Fin | Outcome | Customer confirms, or no request for more help within 24 hours; or a configured Procedure handoff | $0.99 |
| Zendesk AI agents | Automated resolution | Only LLM-verified resolutions draw on the allowance | $1.50 to $2.00 beyond allowance |
| Sierra | Outcome | Outcomes agreed upfront | Not public |
| Decagon | Conversation or resolution | Every conversation, or each resolved one | Not public |
| Salesforce Agentforce | Conversation or Flex Credits | Every conversation, or every action | $2; or $0.10 per standard action |
| Ada | Custom | Per contract | Not public |
| Gorgias AI Agent | Automated interaction | No human needed within 72 hours | $0.90 to $1.00 |
| HubSpot Customer Agent | Credits per resolution | Source shared or action taken, no visitor handoff within 72 hours | 50 credits (about $0.45 to $0.50) |
| Forethought | Quote, outcome-based | Per contract | Not public |
| Freshdesk Freddy AI Agent | Session | Any AI interaction: 24 hours on chat, 72 hours on email | $49 per 100 |
| Kustomer | Engaged conversation | The AI responds, attempts a response or receives a handoff | Not public |
Ask every vendor: Is a silent customer counted as resolved? What happens when a thread reopens? (Fin refunds the charge; HubSpot can bill a second resolution after 72 hours.) Are handoffs billed?
Step 2: Map the actions that close your top tickets
List your top ten ticket reasons and the system each one touches. Then check that the agent can act there. Gorgias issues refunds, cancels and edits orders, and starts returns in Shopify. Fin's Data Connectors reach Shopify, Stripe and Salesforce. Kustomer's Concierge acts in Shopify and Stripe. Agentforce works natively on Salesforce data. Freshdesk's chat workflows call external APIs and can ask the customer to confirm before a high-impact action.
Step 3: Inspect the guardrails
- Confidence and scope: Gorgias lets you choose which intents the agent owns and set its confidence threshold. HubSpot and Freshdesk hand off when confidence is low.
- Deterministic steps: Fin Procedures, Decagon AOPs and Sierra's controlled access to systems of record make sure critical steps can't drift.
- Identity: Kustomer verifies with one-time codes before protected actions. Sierra masks PII and isolates cardholder data.
- Spend: Fin offers alerts and hard limits. HubSpot lets you pause credit usage.
Step 4: Test before customers see it
Insist on evidence from your own data. Fin runs full simulations and a regression library. Decagon runs simulations, A/B experiments and Watchtower QA. Gorgias tests on historical tickets in a sandbox. Sierra runs multivariate experiments. Zendesk audits every interaction with automated QA. Forethought scores conversations with AI QA. Treat published resolution rates, such as Intercom's 76% average, as vendor claims until your pilot confirms them.
Step 5: Model the cost at your volume
Multiply expected monthly resolutions by the unit price, then add seats, base fees and allowances. At 1,000 resolutions a month, standalone Fin costs $989.50 ($49 base including 50, then $0.99 each). Agentforce Conversations cost $2,000. Freshdesk sessions cost $490, but you pay for unresolved sessions too. On Zendesk, anything beyond the allowance costs $1.50 to $2.00 per resolution.
Step 6: Fit the help desk you'll keep
Keeping your help desk? Fin, Forethought, Decagon and Sierra can work with existing stacks. Committed to one platform? Its native agent saves you an integration project: Zendesk, Agentforce, HubSpot, Gorgias, Freshdesk or Kustomer. Choosing the help desk itself? See AI customer support software. Phone-first? See best AI phone agent.
The shortlist by situation
- Most teams: Intercom Fin.
- On Zendesk: Zendesk AI agents.
- Regulated consumer brands: Sierra.
- CX-owned enterprise workflows: Decagon.
- Salesforce shops: Salesforce Agentforce.
- One reasoning engine across channels: Ada.
- Shopify brands: Gorgias AI Agent.
- HubSpot customers: HubSpot Customer Agent.
- Layer on an existing desk: Forethought.
- Budget teams: Freshdesk Freddy AI Agent.
- CRM-led B2C service: Kustomer.
FAQ
What resolution rate should I expect?
It depends on your ticket mix and the actions you connect. Vendors publish their own figures, such as Intercom's 76% average across 12,000 customers, but measure it in a pilot on your own conversations.
Is per-resolution pricing always better than per-conversation?
No. Per-resolution pricing means you don't pay for failures, while per-conversation pricing is easier to forecast. Decagon says most of its customers choose per-conversation pricing.
Do I need to switch help desks?
No. Fin, Forethought, Decagon and Sierra can work with existing stacks. The agents built into Zendesk, HubSpot, Gorgias, Freshdesk and Kustomer need that vendor's platform.
Read next
Fin vs Zendesk · Sierra vs Decagon · Fin vs Decagon · Fin vs Agentforce · Fin alternatives · Zendesk alternatives · Reviews: Intercom, Zendesk, Sierra, Decagon, Salesforce, Ada, Gorgias, HubSpot, Forethought, Freshdesk, Kustomer · The ranked list
Sources
- Fin pricing: Outcomes
- Fin Procedures guide
- Zendesk resolution tiers
- Zendesk pricing
- Sierra trust and reliability
- Decagon pricing article
- Agentforce pricing
- Ada Playbooks
- Gorgias official information
- Gorgias billing
- HubSpot Customer Agent
- HubSpot product catalog
- Forethought pricing
- Freddy AI add-ons
- Freshdesk AI Agent workflows
- Kustomer AI agents billing
- Kustomer verification
Named on this page as a provider? Join the network for a linked profile, Visit buttons and featured placement.