Getting your team to actually use Zendesk AI

getting your team to use zendesk ai

Zendesk AI can triage tickets, draft replies, QA every conversation and resolve enquiries on its own. None of that matters if your agents don’t trust it or don’t know how to use it. We sat down with Annalise Wright, Customer Success Manager at SuccessCX, to talk through what makes a Zendesk rollout stick once it’s live.

Annalise came to customer success from education SaaS and early childhood education, and she’s studying a secondary education degree. That background shapes how she trains teams, and it shows up in nearly every answer below.

Key takeaways

  • Train agents before go-live, not after. They’re the ones in Zendesk all day.
  • A champion from inside the team does more for adoption than any memo.
  • Keep training short: Annalise cites studies suggesting attention drops after 20 to 25 minutes.
  • Zendesk QA can review every ticket, not the 5% most teams manage by hand.
  • One customer moved from 8% to 47% automated resolution by switching to Zendesk’s advanced AI.

Why do Zendesk rollouts stall after go-live?

Most rollouts stall because of change management, not software. Annalise puts it simply: if your team isn’t across how to use a new platform, there will be reluctance to adopt it. And you need adoption, because that’s how you show ROI and prove the decision to move was a good one.

That reluctance usually isn’t stubbornness. Annalise describes herself as very pro-tech, and she still resisted new platforms back when she worked in childcare. The reason was time. “I was reluctant because it felt like more work,” she says. A new tool lands on top of a day that’s already full, so people need a reason to make room for it.

Who should you train first: admins or agents?

Train the agents, and do it before you go live. Annalise sees two main roles in any Zendesk rollout: admins, who update the system, and agents, who work in it all day responding to customers. Admins don’t change the system every day. Agents are in it every single day, and if they don’t understand it, you’ll only get so far.

Yet agent training is the part that most often gets skipped. Annalise sees teams focus on the admins, go live, then train the agents afterwards. “You’ve already missed hours, days of productivity,” she says.

Her recommendation is as many in-person or interactive agent training sessions as your team needs, all before go-live. She also suggests bringing a few agents into the testing phase. They’ll spot problems with views and triggers early, and you can fix them on the spot before the whole team is affected.

How do you get reluctant agents on board?

Start with honest communication, then put a champion inside the team. Annalise recommends telling agents plainly: I understand why you might be hesitant, and here’s what this platform can do for you. Then listen to their feedback and be open to changing things. Paul adds one more step: spell out what’s in it for each person, not just the business.

The champion matters most. Annalise doesn’t mean a team leader speaking on the team’s behalf. She means an agent who works the queue, ideally someone technically savvy or genuinely excited about the change. She says they’ll find the edge cases and talk through the real flow, with contextual understanding a senior team leader or someone running ops just doesn’t have.

Timing matters too. Late in the year is peak season for e-commerce, and for most teams it’s their busiest stretch. Annalise doesn’t say delay the change. She says plan it, and tell the team: we’re transitioning in a month, here’s the transition plan, and here are the enablement sessions we’ll go through. Then give people time to respond, and actually listen.

What kind of training actually sticks?

The training that sticks is short, interactive and at least partly face-to-face. Annalise uses a mix. Every team should have something asynchronous to come back to, like a learning management system (LMS) with interactive questions and fill-in-the-blank activities, especially through the first month or two after go-live.

She’s also a big believer in being in the room. People ask the small questions in person that never come up online, sometimes because they don’t feel psychologically safe enough to type them into a forum. So she recommends at least one face-to-face session for every rollout.

Length is the other lever. “Studies suggest that people lose interest after maybe 20, 25 minutes,” she says, so she breaks training into small chunks that are relevant to everyone in the room. In her sessions:

  1. Every participant follows along on their own screen.
  2. A second person from her team handles the small questions while she runs the session.
  3. The group works through a real ticket together and submits comments on it.
  4. If that means two shorter sessions instead of one long one, so be it.

And she keeps it a little fun. Once, while showing a team what a ticket looks like, she attached a cat meme to demonstrate attachments. A small laugh halfway through a long session is what people remember.

How do you keep training going as Zendesk keeps changing?

Make learning continuous, because the product won’t stand still. Annalise points out that the Copilot side of Zendesk has had update after update over the last six months. That’s good news, but it’s hard to keep up with, and agents won’t get value from features they’ve never been shown.

Ongoing learning doesn’t have to be heavy. It can be an LMS subscription that keeps content current. It can be a monthly admin review of what’s new, followed by a 15-minute lunch-and-learn for the agents it affects.

Staff turnover adds pressure. Paul notes that contact centres and customer service teams tend to have high churn, so knowledge comes and goes with people. That’s why SuccessCX builds enablement into longer engagements. In 6 to 12 month projects, every working session doubles as training, because the platform changes along the way and the team learns it together.

What does Zendesk Copilot do for agents?

Zendesk Copilot gives agents guidance inside the ticket so new starters can work like experienced ones. Annalise’s favourite feature for new agents is auto assist procedures: your standard operating procedures turned into step-by-step guides that appear in the ticket. Copilot suggests a response, prompts the agent through each step, and moves on when they click done.

A step can involve another system. For example, an agent might need to check whether a discount code is valid elsewhere, then return and mark it complete. Annalise also highlights:

  • Conversation summaries at the top of the ticket, so agents don’t scroll back to rebuild context.
  • Recommended knowledge articles agents can share with customers.
  • Topic detection, so agents can tell at a glance what a ticket is about and whether they’re the right person for it.

Paul is clear that this takes commitment up front. Procedures have to be documented and set up properly before Copilot can return accurate, on-brand answers.

Can Copilot keep knowledge when experienced agents leave?

Yes. Knowledge Copilot is one of Zendesk’s newest features, and it addresses a problem Paul raises: when experienced agents leave, their knowledge often walks out the door with them. Knowledge Copilot reads your ticket history, including the replies your more seasoned agents send, and recommends new procedures and articles. It also suggests changes to existing articles, and you can ask it directly, for example to draft an article about your opening times.

Should you trust AI-written help articles?

Trust them as a first draft, never as a final one. “Never just generate and publish,” Annalise says, and that applies to human-written content too. In her experience AI-drafted articles are mostly good, though not always amazing. “Might get you 80% of the way,” she says, which still beats starting from zero. A person reviews and finishes the rest.

How does Zendesk AI help your customers?

It gets urgent issues seen first and gives customers faster, more consistent answers. Annalise is most excited about intelligent triage. You define a topic, such as delivery cancellation, then build routing rules around it. A cancellation can jump to the top of the queue, go to a specific team member, carry urgent priority and a tight SLA, so you don’t ship an order that has to come back.

Paul adds that some businesses face penalties or fines if they miss response times. Flagging those tickets before a human even opens them protects the service level.

Speed and consistency follow from the agent tools. Recommended replies let agents respond quicker, and auto assist procedures keep answers consistent. In regulated industries like banking and fintech, those procedures double as guardrails, so agents can’t send just anything back to a customer.

Can you really QA every ticket?

Yes, with Zendesk’s quality assurance add-on. Paul notes that in the manual days, teams could review maybe 5% of tickets at most. Zendesk QA lets you set up scorecards that run across all of your tickets, and Annalise says it’s quick to set up and not a complicated configuration.

It also reviews your bots. If you run an advanced AI agent, Zendesk QA scores its conversations too, so you see service quality for every customer, whether a person or the AI handled it.

From there, you can use AI to target the tickets that need a closer look, such as negative sentiment combined with delivery cancellations. A person reviews those and sends feedback to the agents. Annalise is firm that team leaders should still review tickets manually. “Let’s not put AI to review everything,” she says. QA simply lets them cover far more ground.

How much can a Zendesk AI agent actually resolve?

It varies, but the numbers Annalise has seen are strong. One AI agent set up with nothing but access to knowledge, no integrations and one or two specific use cases, resolved 80% of enquiries straight away. Another team moved from Zendesk’s legacy Essentials bot to the advanced AI agent and went from 8% to 47% automated resolution.

Results depend on the enquiries you get and how much you want automated. Some companies don’t want everything resolved automatically, and that’s fine. E-commerce is particularly hard to predict. As Paul puts it, the sweet spot is usually high-volume, low-complexity questions from customers who just want a quick answer.

Annalise also admits AI agents can be frustrating; she’s been frustrated by them herself. Two things reduce that:

  • Integrations. Connected to a loyalty program, an AI agent can answer a logged-in customer’s question about their membership or account credit on the spot.
  • Front-loading questions. When a query needs a person, the AI agent gathers the details first, using templates in chat or email. The ticket arrives with the right topic, urgency and SLA, and the human agent avoids rounds of back-and-forth.

Frequently asked questions

What is Zendesk Copilot?

Zendesk Copilot is an add-on that brings AI assistance into the agent workspace. It includes auto assist procedures that guide agents step by step, conversation summaries, recommended knowledge articles and topic detection. Knowledge Copilot can also draft new procedures and articles from your ticket history.

Should agents or admins be trained first on Zendesk?

Agents, before go-live. Admins update the system occasionally, but agents work in Zendesk all day. Annalise Wright recommends in-person or interactive agent training before launch, and involving some agents in testing.

How long should a Zendesk training session be?

Shorter than you think. Annalise cites studies suggesting attention drops after 20 to 25 minutes, so she splits training into small, relevant chunks and runs two short sessions rather than one long one.

Can Zendesk QA review AI agent conversations?

Yes. Zendesk QA scorecards run across all tickets, including conversations handled by an advanced AI agent, so you get one view of service quality across people and AI.

What automated resolution rate can a Zendesk AI agent reach?

It depends on your enquiries and setup. Annalise has seen 80% resolution from a knowledge-only AI agent, and one team move from 8% to 47% after switching from the legacy Essentials bot to Zendesk’s advanced AI.

Keep your agents front of mind

Annalise’s closing advice ties it together. Zendesk can automate a lot and make agents’ lives easier, but the rollout works best when you ask agents what they need and keep them involved along the way.

If you’re planning a Zendesk AI rollout, or your team isn’t getting value from the tools you already pay for, SuccessCX can help with training, Copilot setup, QA and AI agents. Book an AI Readiness Session or talk to us.

Author
Picture of Paul Bichsel
Paul Bichsel
Paul is our Team Leader and SuccessCX Director. Absolutely focused on the human elements of customer experience and dedicated to his family. He revels in nothing more than a cheeky win in a game of Uno. Paul believes ‘the best time to do something, is now’ unless it cuts into his morning coffee and wordle session.
Free Newsletter
Subscribe to newsletter
Get the best tips and tricks
Customer Experience Health Check

Understand where your organisation stands across customer experience, operational excellence, and AI readiness. This assessment will help identify practical opportunities to drive better service outcomes and greater efficiency.