Relevance AI for Customer Support: Triage, Drafts and Plan Gates

A support team can run ticket triage, cited reply drafts and help-article drafts on Relevance AI, but the helpdesk you already use decides which plan you need. Zendesk tool steps work on every plan, while Zendesk triggers, the events that start an agent the moment a ticket lands, are Enterprise-only. Freshdesk, Gmail, Outlook, Slack and WhatsApp Business can start an agent on Pro or Team. Pro costs $19 a month billed annually or $29 a month billed monthly, for the whole organization, with 2,500 Actions a month. Every organization now needs a paid plan, and Relevance AI's pricing pages describe no free trial. Our read is simple. It fits drafts-first support on Freshdesk or email; Zendesk shops should price Enterprise first.

Before the first shift: triage on every new ticket

Relevance AI's support guides describe a triage agent that returns a category, priority, urgency, tags and a recommended queue, with its reasoning, and posts VIP escalations to Slack. The Marketplace template Customer Support Categorization sorts tickets into billing, technical, account, feature and bug, with P1 to P4 priority, and sits next to a Customer Query Handler template. You clone a template, then connect the channels it needs. Cloning does not copy knowledge tables, triggers, secrets or OAuth values, so plan to reconnect the helpdesk and rebuild the triggers yourself.

The guides list three moments that can start a support agent: a new ticket, a tag update and a missed SLA, with alerts when CSAT drops or an SLA is at risk. Whether you get those moments depends on the connector. Freshdesk can trigger an agent from the agent's Triggers tab, and its API Call step reaches the full Freshdesk REST API. Shared inboxes work too. Gmail triggers fire on new mail, labels, senders, subject lines or unread mail, and an Add Label step can tag the thread that started the run. Outlook's first email can take up to 30 minutes after setup, and attachments over 25 MB are skipped.

Your helpdesk decides the plan

Zendesk is the hard case. Its connector signs in by OAuth with your subdomain and an account holding Admin or Agent permissions, and Zendesk API rate limits follow your Zendesk plan. On Pro or Team, an agent can read, search and update tickets, but nothing in Zendesk can wake it up. Relevance AI documents tools that return a list as schedule triggers with deduplication, so a scheduled tool built on Search Tickets is one pattern worth a pilot. It polls rather than reacts, and each run spends an Action. Zendesk triggers need Enterprise, which is priced only through sales ^[1].

Intercom sits in an odd place. The support guides and use-case pages name it as a helpdesk where drafts land. Teams on Salesforce Service Cloud meet the Zendesk problem, because a new Salesforce trigger connection is Enterprise-only. The documented fallback works for both: any system that can send an HTTP request can start an agent through a webhook. Our integrations guide covers native connectors, webhooks and the premium triggers in detail.

Mid-morning: drafts wait, a human presses send

The response-drafting guide is cautious. Its agent drafts a reply with knowledge-base citations and a confidence indicator, inserts it as a draft in Zendesk or Intercom, and the guide advises running drafts only before any auto-sending ^[2]. The controls match that advice. Each tool on an agent has its own approval mode: Auto Run, Approval Required or Let Agent Decide. The Outlook docs suggest putting email-sending tools on approval first and moving to autopilot later, and the Agent Autonomy Limit caps consecutive tool runs without a person, retries included.

Approvals collect on the Tasks page, which has tabs for Approvals, Escalated, Errors and All, covering agents and Workforces alike. A lead can bulk approve or reject pending approvals and rerun errored tasks from the failing step. When an agent cannot resolve a question, it escalates to a person by email or chat, and the human answer can be stored for the next customer who asks. An Escalate to Manager tool hands over a whole conversation. Paid plans also send task alarms by email, Slack or Teams for failed, awaiting-approval and escalated tasks. Our guide to Relevance AI agents explains each of these settings.

Answers come from Knowledge. A small set of macros can go into the prompt on every run, while a large help center is searched with semantic, hybrid or metadata-filtered retrieval. Sources include PDFs, CSV files, website crawls, Google Drive, SharePoint, Notion and Confluence^[p1]. One knowledge base can serve every support agent in the project, so the triage agent and the drafting agent read the same articles, and an update to one source reaches both of them.

Customer-facing channels: widget, WhatsApp, phone

For self-service, an agent's chat can sit on your site through an embed code or a pop-up widget, with brand colors, a welcome message and voice. Widget parameters hide the Relevance AI logo, file uploads and tool steps. The catch is cost exposure. Turning on any embed makes the agent public at an indexable URL, and anyone with it runs the agent on your credits ^[3]. Admins can switch off public embedding and cloning per project, a control added in May 2026.

WhatsApp Business is a premium trigger on Pro and above. It needs a verified WhatsApp Business account in good standing, its tool steps cover messages, media, template messages, interactive messages and customer profiles, and several Business numbers can feed different agents. A provider change in March 2026 forced existing WhatsApp users to reconnect. Phone is weaker for support. Phone agents start on Team and place outbound calls; the docs describe no inbound line answering, so they suit callbacks rather than a support hotline.

Inside the team, a Slack trigger fires on a keyword or a mention of the Relevance AI bot once a workspace admin approves the app, and live status updates appear in the thread. Exclude Keywords stop a trigger from firing, and No Agent Reply lets the agent read a channel without posting. Microsoft Teams needs the Relevance AI app added to every channel where the agent should answer, its triggers fire only on new messages, and replies arrive as threaded Adaptive Cards.

Friday: help articles and the sales-led catalog

Resolved tickets feed the knowledge base. The KB-generation guide pairs an agent that drafts articles from resolved tickets, the Marketplace template Tim, Technical Documentation Generator, with advice to gate publishing behind a human editor. A Gmail Assistant template manages an inbox through conversation. On Team, the analytics dashboard shows tasks, Actions used, credits used, error rate and credits per task, which tells a lead which agent is expensive. Agent evaluations, the formal test suites that block a weak version from publishing, are Enterprise-only.

The names on Relevance AI's support pages are a different offer. Ticket Resolver, Support Triage, Reply Drafter, SLA Watchdog, Escalation Manager, KB Writer, CSAT Responder, Voice of Customer and Bug Report Triager are sales-led agents: you book a demo, and Relevance AI's experts tune them to your playbooks. They are not templates to clone. Relevance AI says Send Payments saves 40 hours a week with its agents. Visual Data Masking, in limited beta and working in the interface only, is Enterprise-only, as is PII redaction on exported events, so Pro and Team show customer details unmasked in the interface.

What a month of tickets costs in Actions

Relevance AI meters two things. An Action is one Tool run, however many steps it holds, and failed runs and retries still count ^[4]. Vendor Credits cover model and tool cost at wholesale with no markup, and 10,000 of them equal $20. Pro includes 2,500 Actions and 10,000 Vendor Credits a month. Team, at $234 a month billed annually or $349 a month billed monthly, includes 7,000 Actions and 35,000 Vendor Credits a month . Both plans cover the whole organization: Pro includes 2 Build Users and no End Users, Team 5 Build Users and 45 End Users.

Size it per ticket. If a triage flow runs two tools per ticket, one to classify and one to update the helpdesk, Pro's allowance covers about 1,250 tickets a month and Team's about 3,500, by our arithmetic; count the real tool runs in a pilot. Extra Actions come as a top-up pack at $80 per pack of 1,000, and unused top-up Actions roll over to the next billing cycle.

Above 6,500 Actions a month on monthly billing, Team billed monthly costs no more than Pro billed monthly plus Action top-ups, by our arithmetic, and it includes 25,000 more Vendor Credits a month. On annual billing the crossover is about 62,000 Actions a year (roughly 5,200 a month), by our arithmetic. Both figures assume your model spend fits Pro's 10,000 Vendor Credits a month; heavier model use lowers the crossover. Calling and meeting agents, the analytics dashboard and bulk scheduling also start on Team.

Support for the platform itself matters once it runs your queue. First-response targets are three business days on Pro, two on Team and one on Enterprise. Pro and Team support runs weekdays 9am to 5pm Sydney time, with engineers on call around the clock for outages. Data sits in the US, EU or Australia region chosen when the organization is created, and it cannot be moved later. Our buyer's checklist covers cancellation, refunds and the other account rules.

Should a support desk start on Pro, Team or a sales call?

Start on Pro if tickets arrive through Freshdesk, Gmail, Outlook or WhatsApp Business and one or two people will approve every draft, because the triggers, approval modes and knowledge search are all there; Pro allows only two users in the organization. Move to Team when more people need access, when you want the analytics dashboard or the 45 End Users, or when you run more than about 6,500 Actions a month on monthly billing (about 62,000 a year, roughly 5,200 a month, on annual billing), by our arithmetic, if your model spend fits Pro's 10,000 Vendor Credits a month.

Talk to sales if Zendesk or Salesforce must start agents on their own, if you need SSO, masking or evaluations, or if you would rather have the tuned Ticket Resolver than build your own. Enterprise is priced only through sales, so ask for a quote that names the helpdesk triggers, the security controls and the volume of Actions you expect.