Zendesk AI Playbook: AI Agents, Copilot and the Resolution Meter

Zendesk's AI comes in three pieces, and only one of them is a separate purchase. AI agents, the customer-facing bots, are part of every current plan and are billed per automated resolution once a small monthly allowance runs out. Copilot, the agent-side assistant, is a paid add-on for Suite Professional and higher, bought for every agent seat. Admin Copilot comes free from Suite Professional. So the real AI question for most teams is volume: how many conversations will the bot close each month, and do you want the overage to bill or the bot to pause? Plan both before launch.

What each Zendesk plan gives you in AI

The map below is the one buyers most often get wrong, because three products share the word AI and each has its own gate. AI agents follow the plan, Copilot follows the seat count, and Admin Copilot follows the tier. Suite Enterprise is quote-only and sold as Suite Enterprise + Copilot through sales, so it sits outside the table.

Zendesk's pricing page lists Support Team at $19 per agent per month billed annually or $25 billed monthly, Suite Team at $55 or $69, and Suite Professional at $115 or $149 ^[1]. Every agent, admin and suspended agent takes a paid seat, which is why AI costs scale with the team; our Zendesk pricing guide covers seats and billing rules. The split that matters here is simple. Team plans get the bot and nothing on the agent side. Suite Professional is the first plan where agents get AI help, even without the add-on.

AI agents: one offering, metered by resolution

Zendesk used to sell two tiers of bot, AI agents Essential and an AI agents Advanced add-on. In March 2026 it announced a single AI agent offering with agentic features on all Suite and Support plans: generative procedures written as plain instructions, multi-step procedures, and actions that call other systems through APIs. With that change, the Advanced add-on is no longer needed for any AI agent feature ^[2]. The older Essential agents, Answer Bot flows and zero-training agents are legacy and being retired. Existing agents are migrated. If you still run one, plan a rebuild rather than a straight copy.

The bots work on messaging, email (including API and web-form tickets) and voice. Every conversation handled only by the AI agent becomes a read-only ticket, and those tickets count toward data storage like any other. SLAs, routing, macros and most triggers ignore them until a human takes over. That is useful for reporting. It also means the AI's work does not quietly inflate your SLA numbers.

Pricing runs on an automated resolution: a request the AI agent resolved without handing it to a person. Since 18 May 2026, conversations are sorted into resolution tiers. An Assisted escalation, where the AI helped but a human finished, does not count against the allowance. A Verified resolution is one where an LLM check confirmed the customer's request was actually settled. Counting is per conversation, so one customer who returns on another device or channel can create a second conversation.

Planning the meter: allowance, overage or pause

Zendesk's pricing page lists extra resolutions at $1.50 each when bought ahead (committed), or $2.00 each pay-as-you-go once the allowance is gone. Included resolutions, capped at up to 10,000 a year, do not roll over. At the limit, an admin chooses one of two behaviors: allow overage and pay as you go, or pause, which stops the AI agents and sends requests to live agents while keeping every setting. If overage is disabled, the pause is automatic.

Here is how the numbers play out for a mid-sized team. Ten Suite Professional agents include 100 automated resolutions a month. If the bot closes 300 conversations, the 200 extra cost $400 pay-as-you-go or $300 committed, by our arithmetic. Each committed resolution is 25% cheaper, but you buy it before you know the month's volume. Commit only the volume you are confident you will use. Start on pay-as-you-go with overage allowed for one measured month, read the actual count, then decide how much to commit.

Some activity never touches the meter. Testing in Admin Center, the test widget, the conversation simulator and sandboxes are free. So are third-party bot conversations and intelligent triage auto-replies. Testing in production can consume resolutions, though. A help center generative-search conversation that the AI agent resolves also counts, and deleting an AI agent ticket does not refund the resolution it used. Suite plans include 100,000 generative searches a month on top of that.

Voice AI agents cost more than one meter

Voice AI agents became generally available in September 2026 for accounts on Zendesk voice, in 53 languages. They do not work with Talk Partner Edition and must not answer emergency calls. A resolved call is billed as an automated resolution plus telephony: inbound minutes at $0.016, AI agent connection at $0.005 a minute and recording at $0.003 a minute, which is on by default. A three-minute call therefore adds about seven cents in minute charges before the resolution itself (our arithmetic). Turning off recording trims a little. Phone automation is the most expensive channel to meter.

Copilot: the add-on versus the free allowance

Zendesk's pricing page lists Copilot at $50 per agent per month, paid yearly, on Professional plans and higher. The catch is the seat rule. Copilot cannot be bought for chosen agents; it is an account-wide add-on for all agent seats ^[3], although access to some features can be managed by group. For ten agents on that plan, adding it to every seat raises the yearly bill from $13,800 to $19,800 (our arithmetic), a $6,000 difference the whole team pays for even if only half uses it.

What the add-on buys is agent assistance at scale: auto assist that follows procedures and proposes approved actions, suggested first replies, suggested macros, suggested assignees, ticket and call summaries, writing help including rewrite in your tone, and intelligent triage used inside workflows. Triage reads topic, sentiment, language and entities on new tickets. Generative AI for voice adds transcription and summaries, with transcription at $0.01 a minute. Usage is unlimited. Copilot does not include AI agents; those come with the plan.

Without the add-on, Suite Professional still gets a capped taste. Writing tools, ticket summaries, merge suggestions and similar tickets share a pool of five uses per agent per month, capped at 500 per account ^[4]. A merge suggestion counts only when an agent confirms the merge. At the cap these features pause until the next billing cycle, with no overage charge. Intelligent triage classifications also arrive free on Professional plans and do not touch the pool, but routing or triggering on them requires the paid add-on.

Admin Copilot, QA and Forethought

Admin Copilot is included from Suite Professional and does not draw on the Copilot pool. It posts weekly insights, ranks recommendations an admin can apply with one click, and answers questions about the setup in conversation. It can also build custom objects and action flows from a plain-language description. Every change shows a summary first, needs admin approval and lands in the audit log. Enterprise accounts can test changes in a sandbox and limit access by custom role. Professional accounts rely on a snapshot and manual reversal through the audit log, so keep a human reviewing each change.

Quality assurance sits in the Workforce Engagement Bundle, which Zendesk's pricing page lists at $50 per agent per month, paid yearly. It pairs Zendesk QA with workforce management and covers all agents in the account. AutoQA scores conversations, including AI agent and voice conversations, against categories such as empathy and solution offered, and teams can add their own prompt-based categories. Zendesk says QA scores 100% of interactions. Treat that as its claim. Forethought, which Zendesk acquired in early 2026, is a separate line: standalone AI agents sold through Zendesk Sales for an annual fee, with no public price. It also works for teams that are not on Zendesk. For a Zendesk account, the built-in AI agents are the starting point.

Your data and the AI layers

Three documents set the rules, and they say different things. The AI Services Addendum says you own inputs and outputs, and that no LLM provider may use your Service Data except to provide and secure the AI Services. The Customer Agreement lets Zendesk use Service Data to provide, secure and improve its products. The trust center describes generic cross-account models trained on aggregated, sanitized Service Data without personal-data fields or attachments, plus generative AI from OpenAI zero-data-retention endpoints, Microsoft Azure, Amazon Bedrock and Google Cloud, whose providers never train on customer data.

AI agents strip personal data such as email addresses on receipt, but Zendesk says it cannot guarantee full removal. Generative features are labelled and can be switched off, and opt-outs exist. Where the law requires it, telling end users they are talking to AI is your job. Read all three layers before you switch it on.

Zendesk's marketing claims run high: AI agents automating up to 80% of interactions, 4.8 billion resolutions delivered, models pre-trained on 18 billion interactions, and 82% higher agent productivity with Copilot. These are Zendesk's figures. Your pilot numbers are the ones to plan on. For a side-by-side with Freshdesk's AI pricing, see Freshdesk vs Zendesk.

The bottom line

Zendesk's AI is strongest when you buy it in the right order. Start with the AI agents your plan already includes, set the limit behavior to overage for one measured month, and commit ahead only the resolution volume you are confident you will use. Move to Suite Professional when agents need help, since that plan unlocks Admin Copilot and the capped Copilot features. Add Copilot only when most of the team will use it, because it bills every seat. Teams on Support Team or Suite Team get the bot, but no AI help for agents.