You can replace Zendesk's AI agent and keep Zendesk. Depending on your plan, Zendesk's messaging platform lets a third-party AI agent answer customers first, then pass the conversation to your team in the Agent Workspace, where it becomes a ticket. Sofa Club, a UK sofa retailer, replaced Zendesk AI with Gem CX and kept Zendesk. It now resolves 80% of its web chat and WhatsApp conversations without the team.
Can you replace Zendesk's AI agent without leaving Zendesk?
Yes. Zendesk's own developer documentation describes a "switchboard" that routes each messaging conversation to one system at a time: Zendesk's AI agent, a third-party bot, a custom integration, or the Agent Workspace where your agents work (Zendesk developer docs, checked November 2026). One system is active, the others wait. When the AI agent passes control, the conversation lands with the team.
We call it the agent swap: the AI that answers first changes, the desk stays. Tickets, macros, triggers, views, reports and your team's habits stay where they are. What changes is who replies before a person does, and what that AI can read.
For an online store, changing helpdesk to get a better AI agent is usually the wrong move. The migration often costs months of the team's attention (macros, triggers, reporting, training), and in our view the AI you would get on the new platform often has the same limits as the one you left: it is configured by your team, inside a product built for thousands of companies.
What are the options, side by side?
Three families keep Zendesk as the team's desk. Prices are the public ones, checked on each vendor's site in November 2026.
| Zendesk's AI agents | A vendor's AI agent on Zendesk (example: Fin) | Gem CX | |
|---|---|---|---|
| Who sets it up | Your team, in Zendesk's admin (procedures, actions), sometimes with Zendesk's team | Your team, in the vendor's tools, with the vendor's deployment team if you want it | Gem CX builds and configures it, your team signs off the rules |
| How it is billed | Per automated resolution, with an allowance included in Suite plans | $0.99 per outcome, on top of your helpdesk | Flat monthly fee: €990 up to 5,000 conversations, €1,490 up to 12,000 |
| Commitment | Annual billing usually cheaper than monthly, prepaid resolutions cheaper than overage | Pay per outcome, with spending limits you set | Monthly plan, one month's notice, one-time setup from €1,500 to €5,000 |
| Who tunes it day to day | Your team | Your team | Gem CX, every week |
Zendesk's AI agents
Zendesk's AI agents live inside the helpdesk: no extra vendor, no extra contract, and they can answer from your help center. Zendesk bills them per automated resolution, a conversation an LLM checks was resolved without a human, and every Suite plan includes an allowance (Zendesk pricing, automated resolutions, checked November 2026). Since May 2026, Zendesk groups resolutions into tiers and only "verified" ones draw on the allowance, while accounts on the older platform stay on it until they move. Zendesk's pricing page for France shows €1.50 per resolution bought in advance and €2.00 as overage. With tiers, the allowance becomes a pool of money and the price of each tier goes through Zendesk sales, so it's worth getting it in writing before comparing.
A vendor's AI agent that runs on Zendesk
Some AI agents are sold separately and plug into the helpdesk you already have. Intercom's Fin is one of the best known. Intercom sells Fin for Zendesk at $0.99 per outcome, with a minimum of 50 outcomes a month (Fin pricing, checked November 2026). Nothing is charged when a conversation is simply passed to your team, but a procedure set up to end in a handoff counts as an outcome. Your team builds and tunes it in Intercom's tools instead of Zendesk's, and Intercom offers deployment services.
Gem CX
Gem CX (gem-cx.com) builds the AI system for you on your catalogue, your orders and your rules, plugs it into your helpdesk, and runs and improves it every week. Billing is a flat monthly fee, with no charge per resolution and no annual prepayment (pricing, how it compares). Custom development and heavy integrations are billed separately, as dedicated time from €1,000 a month. The limit is the other side of the same coin: your team does not hold the controls of the AI, Gem CX does, and changes go through us.
What does it cost at 5,000 conversations a month?
Take 5,000 conversations a month, with the AI resolving half of them, an assumption for the sake of the sums. Those 2,500 billed resolutions come to €3,750 to €5,000 a month at Zendesk's listed rates, before the small included allowance, and about $2,475 at Fin. Gem CX stays at €990 a month, or €1,290 with email replies, after a one-time setup of €1,500 to €5,000. The order reverses at low volume. At Zendesk's listed rates, a flat €990 becomes the cheaper option above about 500 to 660 resolutions a month. Against Fin, it takes about 1,100 outcomes a month at the November 2026 exchange rate. Below those volumes, paying per outcome costs less, and the one-time setup pushes both thresholds higher. For a third-party agent, add any extra monthly active user packs from Zendesk. These amounts date from November 2026, and Zendesk's tier pricing may give a different result.
Sofa Club: from Zendesk AI to Gem CX
Sofa Club sells sofas online in the United Kingdom and gets thousands of questions a month on web chat and WhatsApp. The team first rolled out Zendesk's AI agent, over three weeks, with Zendesk's technical team. The result fell short, with decision trees to configure and a ceiling set by the platform. That was the setup at the time.
Gem CX built a system connected to the catalogue and the orders. Put head to head after one week, it was the one the team kept. In production since August 2026, it answers on web chat and WhatsApp, advises customers up to the order, and hands the team what needs a person.
"In one week, it was ten times better, and really low effort from me."
Nadine Duckworth, CX Director, Sofa Club
Sofa Club still uses Zendesk. Conversations that need a person become tickets, with the conversation and what was already checked, and the team keeps working in Zendesk. The full story, and how each figure is measured, is on the Sofa Club case study.
Why does a helpdesk's AI agent hit a ceiling?
A helpdesk's AI agent hits a ceiling when a question needs data the platform does not hold. The help center answers "what is your returns policy". It does not answer "where is my sofa", "will the corner unit fit a 2.4 metre wall" or "can I still change the fabric on my order". Those need the order, the delivery and the catalogue.
On Zendesk, the AI agent reaches outside data through actions and integrations that your team configures in the platform. Each new case adds a procedure or an action to write, test and maintain, and the CX team slowly becomes the integrator of its own agent. Zendesk is moving its procedures to plain written text, and each one stays tied to a use case. A procedure can link to another flow, at which point the first one stops (Zendesk docs, checked November 2026). From what we see, that's where it gets harder, because each procedure stays siloed and it's difficult to make them work together on the same case. Honestly, that's the part teams underestimate when they sign.
The platform is also the same code for thousands of companies. A need that is specific to your store (a delivery rule, a stock source, a sale period) has to be built by your team with Zendesk's actions and flows, or wait for the vendor's roadmap. We call it the platform ceiling. A system built on your own data and maintained by someone who can change its code moves it.
When is Zendesk's own AI agent the right choice?
Zendesk's AI agent is the right choice when most of your questions are general and your volume is small. Opening hours, returns policy, product care: the help center answers them, the agent is already in your plan, and the allowance may cover your month.
It stops being enough when a good share of questions are about a specific order or a specific product, across several channels. To know which case you are in, read the conversations it hands over.
Gem CX is not the right choice for every team either. If you want your own staff to build and tune the agent day to day, a tool your team configures (Zendesk's agent or a vendor's) fits better. And connecting a system to your orders takes access to your store and sometimes to your carrier, which means a few approvals on your side.
How do you switch without leaving Zendesk?
Two conditions to check first. Messaging must be activated on your account, and third-party bots need a Sunshine Conversations licence, included from Suite Professional up, with a limited number of monthly active users. Extra packs are bought from Zendesk.
Switch one channel at a time and compare before you commit. The steps that matter:
- Read 50 conversations your current agent handed over, and note why: it could not see the order, the policy was missing, the customer wanted a person.
- Check what the replacement can read: orders, delivery status, catalogue, your help pages. If it cannot read the order, it will hand over the same conversations.
- Compare both on real questions before switching, with the new agent running behind the scenes, since Zendesk keeps only one active at a time. At Sofa Club, the comparison happened after one week.
- Keep the handover inside Zendesk: the conversation, what was checked, and the reason, so your triggers, views and reports keep working.
- Make the new agent the responder on that channel once it is ready. Zendesk keeps one system active per conversation, and an AI agent replaced as default responder goes back to draft.
Start with step 1 this week. Fifty handed-over conversations tell you more about your AI agent than its dashboard does.
FAQ
Does Sofa Club still use Zendesk?
Yes. Gem CX replaced Zendesk's AI agent, not Zendesk. Conversations that need a person become Zendesk tickets, with the conversation and what was already checked, and the team keeps working in Zendesk.
Can a third-party AI agent hand conversations over to Zendesk agents?
Yes, depending on your Zendesk plan. A third-party bot or a custom integration can be the active responder, then pass control to the Agent Workspace (Zendesk developer docs, November 2026). The conversation then reaches the team in Zendesk, where it works as usual.
How much does Zendesk's AI agent cost?
Zendesk bills its AI agents per automated resolution, a conversation resolved without a human and checked by an LLM. Suite plans include an allowance, and extra resolutions cost less when bought in advance than as overage. Zendesk's pricing page for France shows €1.50 in advance and €2.00 as overage, but the price of the newer resolution tiers goes through sales (November 2026). At 5,000 conversations a month with half resolved, that's €3,750 to €5,000 a month at the listed rates.
How long does it take to replace Zendesk's AI agent?
At Sofa Club, Gem CX was put head to head with Zendesk AI after one week, and the team kept it. On gem-cx.com, the first version goes live in about two weeks, depending on the scope and the integrations.
Sources
- Zendesk, Switchboard, developer documentation, checked November 2026.
- Zendesk, About the automated resolutions platform (prior to resolution tiers), About automated resolution tiers, Managing resolution allowances and Upgrading from automated resolutions to resolution allowances, help center, checked November 2026.
- Zendesk, About Sunshine Conversations in Zendesk Suite, Managing third-party bots in Admin Center and About monthly active users for Sunshine Conversations, help center, checked November 2026.
- Zendesk, About generative procedures for AI agents, help center, checked November 2026.
- Zendesk, pricing, checked November 2026.
- Intercom, Fin pricing, Fin for Zendesk and deployment services, checked November 2026.
- Gem CX, Sofa Club case study and pricing, November 2026.