See related
No related articles

Best practices for conversational assistants

Last Update: Sep 2024 • Est. Read Time: 7 MIN
To check plan availability, see the pricing page.

With Customer Assist, you can create conversational assistants that help optimize your organization's support experience, promote customer self-service, boost agent productivity, and speed up response times. This article provides tips for setting up these assistants. 

See Introduction to Customer Assist to learn more about this feature.

Note: While we recommend these best practices, your legal team is the best resource to give you compliance advice for your specific situation.

In this article

Disclose the assistant is not a person

Your assistant is an important part of the support you provide for your customers. When required by applicable law and to provide transparency to customers, it should be made clear that they aren't interacting with a human and that the person is interacting with a bot. Generally, we can call them “Brand Bot” to clarify that they are chatbots and express that they interact with an automated system. It should still feel like they are having a friendly conversation with a machine, but it must be made clear to customers that they are communicating with a bot:

  • at the beginning of any conversation or message thread,
  • after a significant lapse of time, or
  • when a chat moves from human interaction to automated experience.

Acceptable disclosures include but are not limited to: 

  • “Hello! Welcome to [Company name]. I am a bot and am not a real person, but I’ll do my best to answer your questions.”
  • “Hello! I’m the [Company Name] bot,”
  • “Hello! You are interacting with a bot,” 
  • “Hello! You are talking to a bot,” 

Note: Even where not legally required, we recommend informing users when they’re interacting with a bot as best practice, as this helps manage user expectations about their interaction with your messaging experience.

Disclose assistant messages are not sent by a person

By default, every message the assistant sends in chat will display "Bot" to indicate to customers that they are not speaking to a person. Change this default setting in your Chat settings.

Consent to record communications

As required by applicable law, you should obtain customers’ consent to record conversations and provide transparency to customers that you use Kustomer, Inc. to facilitate this chat function. Kustomer recommends you do this in your welcome message. An example message is: “By using this chat, you agree that your communications can be recorded. For more information, please refer to [company’s] privacy policy.”

By default, your chat widget will show a short message in the footer notifying the customer that they are consenting to the conversation being recorded. You can customize this consent message in the Design tab of your Chat settings.

Let's talk about strategy

Now that you're ready to create a conversational assistant, you may wonder where to even begin. Below is a list of tips to remember when planning your assistant.

  • Start Small. Start Today. There's no need to wait to build an advanced assistant or spend months planning the project.
  • Think Big. Learn Fast. Constant iteration is the key to success. Iterating can make a massive difference in CSAT and automation rates.
  • Solve your most common use cases. Analyze your contact reason data to help you determine where to start. You can view examples of common reasons to set up an assistant here.
  • Start with one use case at a time. For example, you can start by creating an assistant that provides answers for one topic and hands over additional queries to your team.
  • Reduce barriers to chat with your team. You don't need to try to answer 100% of chat questions. Instead, redirect to agents when needed. Avoid telling the customer to repeat their question.
  • An assistant should reduce friction to resolution, not increase it.  Don’t just ask to ask - know how you are going to use the data.
  • Keep it simple and effective. Do not loop your customers through too many questions or interactions.

Tips when creating menus

Using menus in your assistant is a great way to help guide your customers to the best resolution. 

  • Start with buttons. Preferably, implement multi-level navigation with 1-3 levels of categories and topics.
  • Create the menu from the customer's point of view. For example, labeling a button “My order is late” is better than labeling it “Shipping”.
  • In your welcome menu, have a fallback option where you prompt the customer with another, more general question (for example, “Something Else”).

Routing conversations and handover expectations

When it comes to handing a conversation over to an agent, make sure you clearly define timing expectations on whether the conversation is taking place during business hours or not, and whether you are going to get back to the customer via chat or email. Configure the Agent Missed Conversation volume control setting to provide alternative channels when your team can't respond in time. We recommend 5-15min for chat conversations. 

If you plan to have your chat respond outside business hours or you can’t respond promptly, you can hand over all your chat conversations to your email queue. 

Use the Update Data interaction and queue rules to route conversations to different teams. You can also use the Update Data interaction to save any helpful conversation attributes for your agents or help with routing.

Additional tips:

  • You can create notes for your agents when they receive the assistant conversation using callable workflows before handover.
  • Anonymous chats will prompt agents to verify and merge profiles based on the provided email or phone number. Make sure you verify the information before merging.

Manage your dialogs

Dialogs are a distinct unit of conversational exchange between the assistant and your customer. We recommend making small dialogs that belong to the same topic and that you keep them simple.

Tip: A good way to think about when to create new dialogs: if you think a customer can ask a question about the topic at any point of the conversation, use a new dialog.

You can use assistant rules to redirect customers to specific dialogs depending on the button they click or the current page they are on. This gives customers an incredible first impression, and one step closer to proactive support. You can also use proactive chat with assistant rules to engage with your customers at the right time.

Support multi-language interactions by using snippets in all messages inside the conversational assistant.

Ask for feedback after a final answer is given, usually with buttons such as “Helpful” and “I still need help”.

It is generally good practice to end the conversation after a customer selects “Helpful” as a final answer.

Using callable workflows with input parameters such as your customer, order, or conversation ID allows you to perform any action on behalf of the customer. You can also return information using output parameters to be displayed in a message using the Workflow: Send Response action event.

Test, iterate, and deploy

Iteration is key to your success. Over time, add new answers to your chatbot and measure its performance.

Have multiple testers, and don’t have the person updating the assistant also be a tester. Having a pair of fresh eyes test the flows will help find inconsistencies.

Use assistant rules to test specific start dialogs or assistants for specific customers. You can also use them to personally test an assistant using a specific URL or phone number.

You can also upload the chat code snippet in codepen.com to test your assistant.

Reporting on the assistant

The conversational assistant integrates seamlessly with our reporting features. Here are a few tips you can implement to better understand how your assistant is performing:

  • Evaluate your assistant’s performance with our standard Assistant report or create custom reports that filter by the Assistant attribute. When creating custom reports on your contact reason or any other saved attribute, we recommend combining it with a specific Assistant or Assistant Status conversation attribute.
  • Turn CSAT on for your assistant, and create custom reports by contact reason to understand where you need to improve your chatbot.
  • You can also use the Update Data interaction to add tags or save information you can later report on. If you use this interaction with your contact reason attribute in different dialogs, you can keep your Assistant conversations categorized in the same way as your email or other channels.

Conversation design and brand

Your assistant is an important part of the support you provide for your customers. While they aren't interacting with a human, it should still feel like they are having a friendly conversation. Your assistant is another member of the team and should have a personality of its own. What is their name? How do they represent the brand? Develop a simple personality and be consistent with it in every message.

Note: When developing a personality for your assistant, do not pretend the assistant is a human and use a human profile image as the avatar. 

When setting up your dialogs, use "we" instead of "I" and address the user by their first name, if the end user is authenticated. 

Set the expectations for the conversation flow early on. Clearly state what the assistant is designed to do to help people make the most out of the experience.

Avoid conversation dead ends. When you reach the end of a conversation tree, try to still offer options. For example, you can move the conversation to another matter, return the customer to the menu, or just say bye if it was a final answer and end the conversation. Never block customers in a loop by repeating the same question. Likewise, clearly state when the conversation is handed over to an agent.

Consider using emojis to express emotions in greetings, buttons, and menus. A first approach to the most simple emotions could be:

  • Doubt: 😶😐😑🙄🤔🤐
  • Negativity: 😱😣😕☹️😵🤧😭😔😣😓😤👎💩🤦
  • Positivity: 😀😄👌😍😉😂🤗😎😋😇👍️😙😃👌✌️😁

Don't ignore the assistant's style and personality when communicating errors! Messages such as, "Server error" or "Something went wrong, try again later" can spoil the chat experience and make the assistant look like an unreliable machine. Make sure you maintain the assistant's personality in all scenarios.

Be concise, the less text the better. Check punctuation and try to use correct and natural expressions. Always double-check your spelling. Phrases are better than paragraphs and make sure you watch your word count.

Tips for SMS, WhatsApp, and Kustomer Voice assistants

See Best practices for WhatsApp and SMS assistants for more tips on creating menus for these channels. For tips on setting up an IVR, see this article.