Email onboarding

Your customers can start a conversation with your AI Agent by writing an email to your chosen support address (new or existing, e.g., help@company.com). They can then converse back and forth with your AI Agent to get their issues resolved.

Just choosing the right email setup?

Email handoffs define how your Email AI Agent escalates customer conversations to human agents or support systems. Different handoff paths shape key email behaviors. These include support address setup, threading, and platform-specific considerations.

  1. Choose your handoff agent platform: Zendesk, Salesforce, or an email system.
  2. Use the Quick start as a starting point before diving into details.
  3. Review the table below to decide on the right Email AI Agent behaviour.
Handoff platformShared email (AI Agent + human)?Threading behaviourConsiderations
ZendeskYesZendesk Support block + Zendesk triggers → Replies stay in one threadNo extra dedicated support address needed. Keep forwarding turned ON for your shared support address so that emails continue to reach both Ada and Zendesk.
NoZendesk Ticketing block → Creates a new thread for each handoffNeeds a dedicated support address for your AI Agent. Forward only that address to Ada.
SalesforceNoSalesforce block → Separate Cases: AI Agent uses its own address; human agents reply from another mailbox → new thread for each caseSame as above: Needs a dedicated support address for your AI Agent.
Email systemNoEmail block → Separate email threads: AI Agent sends from its own address; human agents reply from their mailbox → new thread for each caseSame as above—needs dedicated mailbox.

Configuring handoffs is an essential part of the overall setup. For detailed guidance, see Email handoffs for a complete list of available Handoff options and their configuration blocks, and Handoff scenarios for recommendations on combining those methods in specific situations.

Curious about other handoff methods for Email and other channels? Check out the decision table for the full list of options.

Quick summary: Set up email

RequirementWhy you need it
Admin access to your Ada workspaceYou must be able to enable channels and generate API keys.
Domain / DNS control (TXT / MX / CNAME records)Needed to add SPF & DKIM if you map your own address for Direct Email.
Email API keyRequired only when you choose the Email API implementation method.
Test inboxLets you run end-to-end checks before going live.
Brand assets (signature, logo, legal footer)Ensures outgoing messages match your brand from day one.

Setting up email in your AI Agent consists of a few steps:

  1. Set up your AI Agent’s content. The very first thing you’ll need is to add Knowledge so your AI Agent can answer questions.

    Once Knowledge is in place, expand your AI Agent’s capabilities with:

    Next, configure availability rules for Actions, Playbooks, and Processes to implement access control when needed, while CSAT preferences let you gather customer feedback. Together, these features build on your knowledge base to make your AI Agent more personalized and robust, giving you a solid foundation before moving on to configuring the Email channel.

  2. Activate email. Activating email doesn’t mean launching it right away - you just need to turn it on to start configuring it.

  3. Set up a handoff. Handoffs to your support team work a little differently from other channels. Take some time to configure at least one handoff, so you can pass both conversations and attachments to your support team.

  4. Test. Test your configurations with your Ada-provisioned email address, and fine tune them as necessary.

  5. Implement and launch your email channel. When you’re ready to go live and have your AI Agent respond to customers, choose your method of implementation (bring your own domain, forwarding, API, or a combination of methods), and set it up.

    If you want to control which emails get passed to your AI Agent, you can use your existing email provider to create forwarding rules after you have your custom email address set up.

  6. Audit and coach. As your AI Agent has conversations over email with your customers, keep an eye on how they go, and continue to coach your AI Agent to improve its performance.

    On the Ada dashboard, you can go to the Conversations page to view all of your AI Agent’s conversations, then apply a filter so it only displays email conversations for your review.

How your AI Agent handles email conversations

Your AI Agent uses all of the content you’ve configured in Ada to automatically resolve your customers’ inquiries, and writes longer messages that are appropriate in email but not, for example, in web chat. It uses the subject of the email to add context to the customer’s inquiry, and can address multiple inquiries in a single email.

If your AI Agent can’t resolve the customer’s issue, it forwards the conversation to your support team, who can then assist using the agent platform you already have in place.

Like all conversations your AI Agent has, you can audit its email conversations, and then provide guidance to coach and improve the way it interacts with your customers, and its ability to resolve customer issues.

Messages and conversations

Email conversations work a bit differently from other channels in your AI Agent.

Over email, your AI Agent’s messages always include:

  • A salutation (a greeting and the customer’s first name)
  • A sign-off (a closing and the AI Agent’s name)

If you don’t have an AI Agent name configured, the AI Agent will default to signing off with “Ada Bot” as its name.

If you wish to make changes to the salutation or sign-off, or make other refinements to email content, use Custom Instructions with availability rules set to email only. Additionally, you can customize your emails with headers and footers.

Your AI Agent automatically replies to every email it receives, to ensure that 100% of your customer’s inquiries get a response. If your support team needs to send and receive emails to and from customers (i.e. for escalation), without your AI Agent sending replies, we recommend using a separate email address.

Each email thread is treated as a separate conversation in Ada, so it’s possible for a customer to have multiple simultaneous email threads with the AI agent. After 72 hours of inactivity from the customer in an email thread, your AI Agent considers the conversation to be over. If a customer sends a message after this period of inactivity, your AI Agent treats that inquiry as a new question, and create a new conversation for it. Context from the previous conversation does not carry over into the new conversation.

After 72 hours of inactivity, your AI Agent will analyze the conversation to determine whether it was automatically resolved.

In rare cases, email providers may fail to include the reference headers that Ada relies on to link incoming emails to active conversation. If this happens:

  1. The AI Agent will attempt to match the incoming email to an existing conversation using the “From” address and the subject line.
  2. If no match is found, the AI Agent will create a new conversation.

Changes to a subject line mid-conversation will not disrupt the thread. The AI Agent will continue responding in the same email thread, ensuring conversation continuity.

Your Email AI Agent supports replies to up to 25 total recipients across the To and CC fields (excluding the original sender). If the limit is exceeded, recipients in the To field are prioritized and excess CCs may be dropped. For full details on participant limits and how multi-party emails are handled, see Multiple participant conversations.

If the original sender of an email inquiry BCC’s someone in their email to the AI Agent, the AI Agent will only send a reply to the original sender.

If the BCC’d user replies all to the original sender’s message in the thread, the conversation becomes a multiple participant conversation and continues in the same thread.

Your AI Agent is able to respond to forwarded emails. Forwarded content, however, is inaccessible to the AI Agent and it will ask for clarification on the content the end user is referencing.

As a best practice, we recommend that you do not send mass marketing or newsletter emails from the support email address configured with your AI Agent. If you have and an end user replies to this email:

  • The AI Agent will respond, but it will lack the context about the content of the mass email and ask for clarification.
  • Each end user reply will generate a new and separate conversation thread with the AI Agent.

Your AI Agent can detect automatic replies from End Users and knows not to reply to those emails. This is so that automatic reply loops are prevented between your AI Agent and End Users.

If an End User sends a follow-up, non-automated email after an Auto-Reply, the AI Agent will engage with that new message.

Auto-Reply messages from the End User will not appear in the conversation view, and if no further genuine messages from the End User are received, the conversation will close according to our 72 hour inactivity window settings.

Your AI Agent will not reply to emails that include the following headers:

  • Auto-Submitted with values auto-generated or auto-replied or auto-notified
  • X-Autoreply with the value yes
  • Precedence with the value auto_reply

If the first message we receive in a conversation from an End User is an automatic reply, our system will consider this “spam” and we will not create a conversation from it in our back end or in the conversation view. This is to prevent inaccurate billing and resolution reporting for spam-like incoming messages.

  • Some mail providers do not adhere to providing these automatic reply headers. As a result, your AI Agent might still receive some automatic replies from end users.
  • In the event that the mail provider sends an automatic reply message that omits any of the headers, Ada uses rate limiting rules to prevent auto reply loops with the AI Agent.

When your AI Agent replies to end users, it includes instructions to receiving mail servers to prevent automatic replies from being sent in response. These instructions work best with Microsoft Outlook and are not guaranteed to prevent automatic replies from other mail providers from being sent to your AI Agent.

An email conversation includes every message exchanged in an email thread between the AI Agent and an end user (and human agent if the conversation is handed off). Conversations follow these rules:

  • Duration: A conversation continues until the end user has not responded for 72 hours or until 150 total messages have been exchanged, whichever comes first.
  • Inactivity closure: If there is no end-user reply for 72 hours and the conversation has fewer than 150 messages, it is automatically marked as closed. Any further emails in that same thread start a new conversation.
  • Message-limit closure: If a conversation reaches 150 messages (for example, due to an auto-reply loop), it remains open until 72 hours after the 150th message is exchanged, then closes. This limit is an intentional safeguard to prevent potential long-running loops with end users. Most email conversations never reach this limit.
  • Data storage and security
    • To learn about how Ada protects your data, see our Trust Center.
  • Spam
    • To have the most control over your email security, we recommend using our bring your own domain feature. That way, you can filter out spam using your mail provider’s existing settings, before forwarding them to Ada.
    • Ada doesn’t currently provide spam filtering.

Limitations

  • Email is currently not HIPAA eligible.

  • The List block, and Capture block are currently not supported for the Email channel.

  • Because all email conversations start with customer messages, content in your Greeting doesn’t appear in email conversations.

  • Your AI Agent can only reply from one email address. For example, if you have both billing@company.com and help@company.com, you will need to decide which email address is associated with your AI Agent.

    You should also make sure that the email address associated with your AI Agent isn’t used for anything else, like escalated conversations or handoffs using the Email block, so you don’t unintentionally send customers back to your AI Agent.