Use cases by system6 use cases

AI on Zendesk and other helpdesks: when the built-in agent is enough

The helpdesk is where AI support is cheapest to buy, and pricing has converged on paying for results. Intercom charges $0.99 per Fin outcome, Help Scout $0.75 per AI resolution, Freshdesk includes 500 Freddy AI Agent sessions and then charges $49 per 100 sessions, and Zendesk puts AI agents on every plan and bills per automated resolution. Gorgias, the Shopify favorite, also charges only when its agent resolves.

For questions the help center answers, or that one connected system like Shopify answers, that is a good deal, and I would start there. The builds in this hub begin where the helpdesk agent cannot reach: second-line SaaS tickets that need logs and account data, warranty claims with photos and serial numbers checked against the ERP, a review of every conversation instead of a sample, and return reasons joined to SKUs and suppliers.

In each of them the helpdesk stays the system of record. The agent reads through the API, writes its notes and replies back through it, and the support team keeps its views, macros and SLAs.

Book a 30-min scoping callAI Workflow Automation →

What to build around Zendesk and helpdesks

Building against Zendesk, Intercom and Gorgias

  • Triggers call the agent

    In Zendesk a trigger fires a webhook when a ticket is created or updated, and the agent answers through the Ticket API as an internal note or a public reply. Updates made by the agent's own user are ignored, so it never answers itself in a loop.

  • Custom fields carry the result

    The agent sets fields such as intent, order number, product and urgency. Views, routing and reports then run on its classification, and every correction a person makes becomes a test case for the next change.

  • Backfills use the export endpoints

    Zendesk's API rate limit depends on plan and add-ons, so pulling a year of historic tickets to evaluate an agent runs through the incremental export endpoints, not thousands of single-ticket reads.

  • Actions are where the products differ

    Gorgias acts natively in Shopify: tracking, cancellations, address edits, refunds and returns. Fin can take actions in external systems you connect, and Zendesk sells advanced agents that work across platforms. None of them knows your ERP, your carrier contract or your refund rules until someone connects them.

Where built-in helpdesk AI stops, and what a custom agent adds

  • Help center questions are solved

    If the answer is in the help center, a built-in agent billed per resolution beats anything I would build on time and cost. A custom agent is for everything else.

  • The per-outcome bill at volume

    At 10,000 automated resolutions a month, $0.99 each is about $9,900 a month, every month. Owning the agent turns that into model usage and maintenance, which is worth comparing once volume is high and stable.

  • Triage the team can read

    Routing by contract tier, SLA, language and product lives in code rather than in a vendor's prompt settings. The team can read every rule and change it without opening a ticket with the vendor.

  • Residency is about processing

    Check where the AI processing happens, not only where tickets are stored. For regulated support a custom agent can run its models in the EU and log every lookup it makes.

  • No official MCP server for Zendesk

    Only community and third-party Zendesk servers appear in the MCP registry. If your team wants Claude or ChatGPT to read tickets, I would put a scoped, read-only server you control in front of the API.

Frequently asked questions

How much does Zendesk AI cost compared with Intercom Fin?

The main helpdesks now price AI by result. Intercom charges $0.99 per Fin outcome, Help Scout $0.75 per resolution, Freshdesk $49 per 100 Freddy sessions after 500 included, and Zendesk bills per automated resolution on every plan, with Copilot for human agents at $50 per agent a month. Compare them on your own volume and on which questions each can actually resolve.

Can a Zendesk AI agent take actions like refunds or address changes?

Within limits. Helpdesk agents act through the integrations and actions they support, which works well for Shopify orders and popular apps. Actions in your own ERP, billing system or carrier accounts need a connection someone builds, and anything that moves money should run through code rules with limits, not the model's judgment alone.

Should we switch helpdesk to get better AI?

Rarely. Migrating means moving macros, views, reporting, integrations and team habits, while the helpdesks' AI agents converge on the same per-outcome model. If your current helpdesk's agent cannot reach the data you need, a custom agent connected to the helpdesk you already have is faster and less disruptive than a switch.

Can you build a custom AI agent on top of Zendesk?

Yes. The agent listens to Zendesk triggers, reads the ticket and customer context through the API, looks up the other systems involved, and writes back as internal notes, drafts or public replies depending on the category. Your team keeps working in Zendesk. The same pattern works on Intercom, Freshdesk, Gorgias and Help Scout.

Also browse: Customer support · SaaS · E-commerce · Shopify · all use cases