Generative email replies with AI Agent Essentials in Zendesk
Zendesk expands generative replies to email, replacing static auto-replies with AI-crafted answers from Help Center content. Customers get faster resolutions, agents save time, and teams can evolve toward AI Agent Advanced for contextual, multi-source automation.
The goal of any customer (or employee) service team is to make sure users get a correct answer to their question quickly. Nobody likes waiting for a resolution and even though most people can understand that some questions take time to figure out and resolve, the reality is that most of those questions can be resolved with a bit of prewritten and reusable information.
Zendesk's platform is designed for this purpose. If customers email your team, submit a form, or contact you via social channels, the platform ensures that interactions are captured and routed to the right agent. This agent can either be human or AI-powered.
Even before the advent of AI, Zendesk had powerful (for the time) self service capabilities in place to make sure the system could attempt to deflect some questions. The Help Center was, and still is, at the core of this strategy.
Customers that Google for a question, or open your Help Center via a link on your website can search for a topic, and they got a list of articles that might contain an answer to their question. Similarly, AnswerBot was Zendesk's solution for customers that reach out over email or social channels. If customers interacted with your chatbot, the bot replied with a list of applicable articles. And if they emailed your support team you could replace the default "we got your message" email, with a more powerful response that contained a list of applicable articles. The customer could then read the article, and if it helped, close their ticket.
And if it didn't, well, at least the agent didn't loose time sending those links, and maybe the customer might even reply with additional context on why the support article didn't resolve their inquiry.
This strategy works, but it does create a few issues. For one, the customer needs to do all the work. They need to read a few articles and look for the relevant section that answers their question. (As always, it's better to write 3 short on topic articles, than one longer article that handles all three).
Second, even though they get suggested solutions, the static format of those emails and the links to your help center make those emails feel like automations in response to their email, instead of personal answers to their inquiry. It's a small nuance, but if you read further in this article, you'll see what I mean.
And thirdly, Answer Bot replied based on semantic search and replied with articles that contain some version of the keywords used in the customers' email. So your success rate highly depends on how good the customer phrased their email.
Now, how can we improve this experience and provide a better automation resolution for your customers?
Generative Replies
Enter Generative Replies. Powered by LLMs these new capabilities in the Zendesk platform shift away from lists of relevant articles towards generated replies personalized for each customer.
Intelligent Triage is Zendesk's way of understanding what tickets are about. The system reads every conversation and assigns an Intent (why is the customer contacting us) to each conversation. It combines that intent together with an actual understanding of the message to find a matching piece of text in your articles. It takes those paragraphs of text and turns it into a response for the customer. (This retrieval from indexed sources is often called RAG).
You can these generative responses in action across the platform. Help Center offers Quick Answers that augment search results. The classic Answer Bot got replace with a generative AI Agent (Essentials) and Agents see Quick Answers right in the Knowledge Panel next to tickets.
But until now one of the classic support channels, email, did not get those new generative capabilities. If you bought Agent Copilot you could use a combination of Intents and Triggers to send prewritten responses for certain intents. But for all other scenario's you only had the classic Answer Bot (Auto-reply) with list of article links available.
Thomas Verschoren
Generative Replies for Email
This article wouldn't exist if that hadn't changes. Announced last month and now available to all customers is the new generative replies for email, part of AI Agent Essential. It removes the old "email with links" and replaces it with personal responses to customer emails. These responses are based on your Help Center content and offer customers answers to their questions in a personal and custom tone.
Instead of the customer needing to read an article, they now get a response that explains how they can resolve their inquiry, right from the email. If the solution helps they can either ignore the email and move along with their life. This will trigger an automatic closure via automations, followed with a satisfaction survey.
If they do reply with additional feedback or requests, the ticket will arrive in your agents' inbox and they can take over this ticket. But for most single touch tickets, AI Agent Essential over email should suffice. Naturally, any knowledge gaps the system detect should be filled by your team over time to further improve its' efficiency.
Setting up generative emails
You might have never touched auto-replies, or if you. ight have already enabled AI Agents for messaging and want to enable the same for email. Regardless on where you are in your road to automation, setting up generative emails is easy.
Thomas Verschoren
It all starts with configuring your AI Agents' persona. You describe your company, you give your agent a name and avatar and you decide it's tone of voice. These will define how your AI Agent reacts and will influence the way it phrases responses. It will for example understand if it's replying to customers, guests, passengers or employees. Or it will use a very dry or very casual tone of voice.
By default your AI Agent will pull from all public Help Center articles to reply to your customers. If your user belongs to a segment that has access to restricted articles, the AI Agent will also pull from these articles.
Once you've setup your AI Agents' persona you can start adding channels. You probably already have the bot enabled on (some of) your messaging channels, but new is that you can now also choose email and webform based channels. Scroll down on the configuration page and you'll see all email addresses and web forms linked to your AI Agents' brand.
Whichever ticketing channel you choose will enable the AI Agents' responses on that channel. Customers who submit a ticket via a webform or email your team will get a generated response instead of a generic confirmation email after the ticket is received.
For webforms I'd like to see a way to show these results inline on the webform page, similar to the suggested articles we have now. Submitting a form and receiving a suggestion feels slower than showing the response directly on the page you press submit on. But maybe that's just me.
Triggers
For Messaging channels the AI Agent is enabled by Zendesk setting it as the first responder for that specific channel. But for email we need to take existing triggers into account. You probably already have triggers that notify customers of received tickets. Or you might have triggers that make use of auto-replies, the legacy Answer Bot or a myriad of other notifications.
Thomas Verschoren
Luckily Zendesk takes these into account. When you enable ticketing channels the system will show a list of conflicting channels. You can open these and either deactivate them, or tweak them to exclude the brands, forms and email addresses you want the AI Agent to handle.
The system will also create a set of new triggers that contain a new "Generative reply" action, as well as two new automations that create a remind and solve automation flow. Customers that get a generative reply will receive a reminder email after a few days. And if they still don't reply we can assume their issue is resolved and the system automatically closes the ticket.
Thomas Verschoren
If you want you can dive into these triggers and automations to tweak durations or to exclude CSAT for those emails e.a.
One thing I did immediately is change the new trigger to add a pending status to these tickets. Our AI Agent replied, so we don't need to see these tickets in our inbox.
Experience
Once enabled the experience for both customers and agents is pretty straightforward. When a customer emails your team they get an email with a response to their inquiry. The email clearly notes it's written by an AI Agent, and even contains the agents' name.
That same response also shows up in Agent Workspace so your agents can also see the full conversation in the ticket. Should the customer reply for further help, they get the full context.
However, for scenarios where there's no answer available, there's a fallback to the default "We got your email" notification. These do not show up in the agents' inbox, and your team can handle these tickets just like any other ticket.
There's two small things that can be improved to the experience though. For one, in all my tests I saw Agent Copilot offer suggestions to agents after the AI Agent responded. Since ticket conversations are always a back and forth between customer and agents, I would expect our Copilot to keep quiet until the customer replied to the AI Agent comment.
Secondly, and this is a nitpick, but Zendesk sends out the AI Agent response immediately. Sending a support email and getting a 2 paragraph response from a bot in seconds is actually a weird experience for customers. Delaying that reply with a few minutes, like we can with AI Agents Advanced, actually creates the illusion of work and effort, and I've seen and heard from customers that their users trust the answers more if they were sent with a small delay.
And a final tweak I'd recommend to make is to go into the trigger and change the subject from "Request received" to "Re: {{ticket.title}}". Getting an email titled "we received your request" that then contains an actual solution to your question is a bit weird. That original title invokes the idea of it being just a confirmation message that should be deleted.
Making the experience even better
Generative emails for Zendesk Suite are a welcome addition to the platform. It leverages existing knowledge sources and gives customers quicker and more precise answers to their questions.
But not every question can be resolved with just knowledge. Sometimes you need context and data that lives outside of your Help Center.
For one, some answers might be available on your website or website, and not part of your Help Center.
If you have different products you might want to know what product the customer is talking about before you send them update instructions. Or, if a customer asks about Order Statuses, you need to ask for an order number first, and need to connect to your order systems to retrieve that status.
All these more advanced options are available as part of AI Agents Advanced. Where the AI Agent Essential offering we discussed in this article is limited to generative replies based on your Help Center, AI Agents Advanced can index your website, website or other sources.
Thomas Verschoren
Additionally, you can define use cases with dialogue flows to build complex email threads that handle scenarios where you need to ask context (order number, product type) or pull in data from external sources (order status, customer type,...)
Once you start using generative emails you will encounter these scenario's where just generative does not work. Upgrading to AI Agent Advanced is a way to further automate your conversations and remove workload from your team.
You might also encounter scenario's where a fully automated flow is not ideal. Complex customer use cases. Scenarios that are exceptions on processes. This is where you can look at Agent Copilot. It helps agents writing complex answers, provides context via Actions and integrations and assists your team resolving these unique questions.
Thomas Verschoren
Conclusion
As I wrote in my article on resolutions and licensing earlier this month, I really like how Zendesk is incorporating AI into the platform to make existing features more powerful. Where previously generative replies were locked behind add-ons, now every Zendesk customer gets a taste of automation right within the platform.
The Help Center gets Quick Answers, channels get automated with generative replies, and Agents get Quick Answers right in the context panel. I wish they'd put the writing tools (expand, rewrite, summarize) into the basic Suite too someday.
And for customers that need more automation power, there's AI Agents Advanced with Agentic Procedures. There's Agent Copilot with procedures and instructions, and there's QA, for those that need additional insights.
An AI-powered resolution platform indeed.
