So, you’ve decided to dive into the world of chatbots. You know what makes a good chatbot great, but how do you get started?

Two weeks ago, we shared two key differentiators that make all the difference when designing a great customer service chatbot – deep understanding of your customer, and a flexible algorithm that can handle real world interaction.

This week, we’re going in-depth to bring you four best practices in chatbot design once you’ve made the decision to build one.

1. Settle on a concrete goal before you start

Getting off to a good start will eliminate some huge potential problems ahead. There’s nothing worse than embarking on a project with a fuzzy idea of what you want at the end – so make sure you set a realistic goal and create project firm boundaries around your chatbot project.If you can imagine it, there’s a bot for it. You may be familiar with Kyber, the organizer bot that streamlines your team management on Slack, or with Eno, Capital One’s customer service bot.

If you want to optimize your own operations or save money, you may consider creating a bot focused on a single department, like HR. Or, if you want to provide more personalized customer service, you can create a bot to help users manage their orders or answer questions.

Source: There is a bot for that

Setting project boundaries off the bat will help you find real world examples to familiarize yourself with (check out there-is-a-bot-for-that for some ideas), and answer some logistical questions – where will you host it? What information do you need to get started? Which departments will need to collaborate on the project? These answers will determine your next steps. And a note about ROI – don’t be afraid to track ROI for an AI project. There is no secret formula, and if done right, returns will come quicker than you would ever imagine. If you’re not sure where to start, we’ve put together guidelines for you to follow.

2. Don’t just pass it off to IT – collaborate throughout

Cross-departmental projects are never easy, but this is a must in chatbot development. Once you’ve decided how your bot will be used and where, you’ll need to work closely with the target department, and supporting departments to ensure a smooth deployment and a quality end-product.

If you’re building a customer service chatbot, you’ll need to talk to HR and Marketing or you risk ending up with something that doesn’t fit your branding or your company culture. And asking for a list of common customer questions isn’t enough input from customer service department – your bot needs to be flexible enough to handle customers who will want to backtrack, ask follow up questions, go off on tangents, abbreviate, misspell, or speak in colloquialism – this is a nightmare if you’re working off a decision tree built from a list.

Take the time to collect more data than you think you’ll need from more sources than you think are necessary. If you’ve selected the right algorithm for your project (click here for a refresher on how to choose), your bot will learn as it interacts with customers. But it needs to be working well when you roll it out, or you’ll end up repelling customers, and it will never get the chance to learn and overcome these barriers.

3. Make it as user-friendly as possible

This goes both ways. If it’s customer-facing, it obviously needs to be customer-friendly, but it also needs to integrate seamlessly with your team and existing systems.Making a customer-friendly bot requires in-depth knowledge of your customers. Not just the information that they tend to ask for, but how they ask as well. Some design tips culled from the most successful chatbots are as follows:

• You should aim to make your bot warm and friendly, but not to the point where it distracts from task at hand.
• Your copy should be given high priority and looked over by marketing in order to nail your brand’s voice.
• You should break up information with conversational touches, and send in several smaller messages rather than one huge block. This mimics human conversation without taking focus away from the answers provided.
• You should use a blend of buttons and text. Using buttons allows you to lead the user, but saves them from having to type long messages. But don’t rely on this alone – using text allows for a more authentic experience and gives customers freedom to ask follow-up questions, change the subject, or take the conversation in whatever direction they need it to go.

Kasisto Kai, a conversational bot for retail banking, strikes a balance between friendly and helpful. Source: Quartz

To ensure easy integration with your existing team and systems, make sure you’re not building the bot in a silo. All questions, answers, data and actions should flow to the same source(s) used pre-bot. And your team should be able to make use of interactions to improve their day-to-day. If your customer service bot isn’t recording or passing along analytics to your CS team, you’re missing out on a valuable opportunity to get to know your customers better.

4. Set expectations when you introduce your bot to customers

Don’t waste this upfront opportunity to set expectations around what your bot is capable of. It’s not going to be able to answer every question thrown at it – and most customers expect this. Taking the time to explain the purpose of your bot during launch will help with retention, as users won’t be disappointed when it can’t chat about their lunch, or debate politics.

Bitcoin Buddy, a simple bot, sets expectations immediately and clearly. Source: Bitcoin Buddy

At the same time, it’s important to design ways for customers to get out when they need to speak with a person. Don’t leave your customers guessing whether they were understood – best practice is to make it crystal clear right away that your bot doesn’t understand, apologize, and offer a way for customers another way to get to their answer.

To Recap:

Chatbot implementation isn’t easy, but it can be incredibly rewarding to both your business and your customers. If you’re ready to embark on the journey, remember to plan carefully and collaborate widely, don’t skimp on UX, and don’t forget to set expectations with customers around what your bot can do.

If you’re interested in learning more about how AI-based customer service  can take ticket deflection from 10% to 60% or more, read our case study here.