In Part 1, Setting up Answer Bot, we covered the initial stages of creating your first bot, from choosing the channel to host your Answer Bot, to the persona your bot will use to converse with customers.

In Part 2, we’ll cover how to tweak the very wording of your bot, so that not only the persona is on point, but also the language and meaning of every word your Answer Bot produces.

We’ll also cover how to program the bot on what it should do when something inevitably gets lost in translation between your very human customers and your Open AI-powered bot!

With your Answer Bot properly fine tuned, customers will be able to access the answers they need more quickly and at a much lower cost to your team!

Fine-tuning bot behavior

Within your new bot’s configuration page, switch over from the Settings tab to the Behavior tab.

‍

Expanding the options on the Behavior page, the first section we see allows us to tell Answer Bot how it should greet customers at the start of interactions.

‍

There are two main options here, Send a greeting and suggest answers (the default) or Start with an answer.

A note on bot answers

Whilst a guide to creating bot “answers” will be a whole article series of its own , it’s worth taking a moment to understand the concept now. In the Zendesk bot builder, an answer is a pre-programed flow (a bit like a decision tree, but used by the bot to guide its conversation) that you can create to help customers get the help they need on any specific topic.

Unlike the core AI function of Answer Bot that we talked about in Part 1, which tries to understand your customer’s intent and then matches it to knowledge stored in your Guide, answers rely on you (rather than AI) to provide the intelligent solutions your customers need. This could be by building a bot flow that pulls in shipping information via API, routes VIP customers straight to live chat based on their email address, or collects as much customer information as possible before allowing customers to request a callback, to name but a few options.

For now, we’ll focus on building our bot based on Zendesk’s AI features pulling in useful data from our Guide, as this is a considerably quicker route to getting started, with a much easier learning curve!

Say hello

As you can see, Zendesk has already suggested a default greeting. Crucially, it’s worth noting the “Generate variations” box that’s checked below.

This will harness the power behind Chat GPT so that your Answer Bot says similar versions of your greeting every time, without repeating itself too often.

For example, if you set the Message to “Welcome to Salto - how can I help you today?”, your bot will occasionally start with this line, but will also come up with lines such as “You’re through to Salto support, what would you like help with?” or “Thanks for contacting Salto, what can I do to help?”

These sorts of variations keep your interactions fresh and help prevent your customers rolling their eyes at yet another faceless bot.

If you have set up answers already, you can also prompt Answer Bot to offer a few suggestions as to which flows they might like to jump into to get support. My suggestion is that unless you’ve spent a sizable chunk of time learning how to build answer flows (something we’ll cover in another article),  leave this section blank.

Instead, ensure you have a broad range of Guide articles set up, as this is the core source of information for any Answer Bot. It’s worth reiterating this point; an Answer Bot without a whole host of Guide articles from which it can pull answers will be about as useful as an librarian with no books!

The final two options allow you to fine tune both what your bot should say when it doesn’t understand what the customer wants, and also what to say when it does understand the customer but doesn’t have the resources (Guide articles, answers) to actually help.

Clarifying questions

The set up for this section is very similar to the greeting configuration. I recommend choosing a base phrase for your Answer Bot to write when they can’t understand the customer’s intent, and then selecting the box to allow Answer Bot to generate variations of the same meaning, using slightly different phrasing.

Just make sure not to forget to include an actual request for the customer to rephrase their question. Having a bot respond “I’m sorry, I don’t understand”, but not then follow up with a request to try again to resolve the situation is one of the most infuriating experiences you can subject your customers to!

The last straw

If your Answer Bot understands your customer’s intent, but is still unable to help, we need to tell it what to say (and do) as to not leave your customer in the dark.

This section might look similar to the previous two, in that you choose a phrase and Zendesk will rephrase it as and when, but there is a crucial difference here.

Remember, this is the part of the flow where the bot is unable to help your customer. That means it was unable to find any relevant Guide articles or answers to solve your customer’s query.

In order to keep your customer happy, we need to take decisive action here. If you have generated your own answers already, you can use the option at the bottom to suggest a few of your most frequently used answer flows to see if any of them are what the customer is after. This will appear as simple buttons within the chat.

But what to do if you haven’t built any answers? Well, here Zendesk has stepped in to help (and has made a pretty clear suggestion as to how they think you should keep your customer happy).

Whenever you create a new bot, Zendesk builds one answer for you; Talk to a human. In short, this Answer allows your customer to live chat to an agent during your opening hours, or to send a ticket when your team are off. Answer Bot accepts that it can’t help your customer on its own and passes the baton to a real human. Even better, it will collect your customer’s name and email address and populate the ticket with them, saving your agents time asking and allow them to get on with solving the issue right away.

Although you’re undoubtedly reading this as you’d like to set up an Answer Bot that reduces the proportion of customers who require expensive human support, it’s important to be aware of the limitations of the technology and accept when it’s time to get some good old fashioned human-powered customer support to sort things out.

For that reason, I suggest leaving the settings as they are, but changing the message to offer the option of connecting to the team. I’d go for something like this.

With your bot configured like this, you always know that no matter how complicated a customer query is, there’s always going to be an option that provides them with the right level of support.

Summary

Answer Bot is an excellent solution for fielding low-complexity customer queries, allowing your team to focus on the more complex or urgent queries. It allows you to respond to customers in the same tone and voice as your human agents, ensuring a CX consistent with the rest of your brand.

With your bot now properly set up, you can not only solve your customer’s queries in a flash, but also ensure that customers who do need to get through to an agent can still do so. That really is the best of both worlds.