Implementing the right chatbot: what to look for

You might be convinced of the value of deploying chatbots to support your contact centre, but do you know how to choose the right chatbot?

In this article, we’re looking at some of the key factors surrounding chatbots, including deployment, features and integration. This should help you evaluate your options and decide which chatbot technology is most suitable for your organisation.

Ultimately, you need to consider which chatbot platform provides the greatest value, while offering the simplest integration for the most manageable cost. Broadly speaking, there are three categories of chatbot:

• Menu or button-based
• Keyword recognition
• Contextual (learns from interactions)

 

Chatbots will lead to cost savings of over $ 8 billion by 2020, primarily in banking & healthcare. (Chatbots, a Game Changer for Banking & Healthcare – 2017)

Start with your needs

First things first. Before you delve into the world of chatbots and consider your options, it’s important to define your goals so that your search is focused on the features your organisations needs, and less influenced by the features that vendors are promoting.

Work with your teams to plan how a chatbot will fit into your contact centre. Which systems does your chatbot need to integrate with? Which categories of queries do you want the chatbot to handle? What should happen when customers don’t want to use the chatbot? And how can customers exit the chatbot process – or transfer to a human agent?

As we look at some of the other features and functions of chatbots, we will explore what these elements mean for your organisation.

Gartner identified customer service as one of the most important use cases: By 2020, chatbots should take over 85% of customer service interactions.  – Top 10 Strategic Predictions by Gartner – 2017

Integration and APIs

Having explored how you want the chatbot to work in your contact centre, you should have a list of the apps and systems that will require integration. This list might include payment gateways, CRM systems, ID&V processes and customer communication tools. For example, you might want a chatbot that can confirm the customer’s identity, query your SAP CRM, and then push your customer communications management (CCM) platform to create a text message confirming the customer’s request.

Many chatbots offer APIs and are designed to be integrated with popular enterprise solutions. How much programming is required to complete the integration? And how much time and expense will this add to the deployment?

Natural language processing (NLP)

Natural language processing is the technology that voice assistants like Alexa and Siri use to understand what we’re asking, and turn those words into actions. In a chatbot, NLP allows users to type any sequence of words into the chat window. The chatbot can understand the user’s meaning and their intent (i.e. are they asking a question or making a statement).

Chatbots without NLP usually resort to giving users canned responses to choose from. Instead of typing a question, the chatbot asks users to click a button that best represents their query. As you can see, a chatbot without NLP is likely to be less capable of dealing with a wide array of customer queries, unless you use a complex menu system to gradually refine the customer’s needs.

Chatbots that learn

Chatbots will only learn if you teach them. Your chatbot can only provide the information it has been given, in ways it has been taught. So if a chatbot is giving customers their bank balance, it is only because it has been programmed to do so, and connected to the data sources it needs.

This sounds obvious, but it is important if you want to develop a chatbot that ‘learns’ from interactions – and doesn’t rely solely on manual input to develop its knowledge and skills.

If you want to create a smarter chatbot that gradually learns from interactions, you must create feedback loops (or other processes) so that evidence from experiences is fed back into the chatbot’s code. For example, chatbots can recognise popular customer choices, and then learn to prioritise those most popular selections.

Complexity

Rather like any enterprise software, you can choose from solutions that include an immense range of functionality, and can be adapted to suit your changing needs, or you can choose a simpler package that is easier to deploy, but limits what you can achieve.

Features

Which features are essential to your contact centre? The features that often vary between vendor include:

• Live testing options
• Archives – so customers and agents can access previous conversations
• Personalisation
• Third-party integrations – so you can easily connect your existing applications
• API – so your developers can build custom integrations
• Admin tools
• Reporting and analytics
• Creative tools
• Product integration for recommendations
• Sales and payment processing
• Multi-language support

Utterances, intents and entities

When you start to develop your chatbot, you’ll find yourself talking about these three concepts:

Utterances are the things your customers say or write.

Intents are your user’s goals.

Entities are things that modify the customer’s intent.

For example, a customer might ask to open a new account. ‘Open’ is the intent. ‘Bank account’ is the entity.

Or a customer might type: I want to order a pizza. In this example, ‘order’ is the intent and ‘pizza’ is the entity.

Dialogue flow design

Once again, you have a choice between more flexible chatbots with a user-friendly interface, or more complex solutions that require more technical skills to implement and optimise. Some chatbots offer drag-and-drop interfaces so you can design the dialogue flows, while others require coding skills to update and customise.

Exception handling

How does the chatbot respond when a customer does something unusual, or the chatbot simply doesn’t understand the input? Does your chatbot give you options to follow-up on abandoned chats, or to reroute customers who get frustrated, or need more personalised support?

Only 43% of consumers said they would prefer to communicate with a human. 34% said they would use a bot to connect with a human employee.  – The 2018 State of Chatbot Reports – 2018)

Dialogue datasets

If you are building a custom chatbot, you may need to consider which data to load into your solution. The dialogue dataset is essentially a database of conversations that form the basis of the chatbot’s ‘brain’. Datasets include lists of questions and answers, customer support conversations, chatroom discussions, and even dialogue from films.

In 2017, Microsoft acquired Maluuba, a company focused on creating open datasets to support machine learning and AI systems like chatbots. Maluuba’s datasets include machine reading comprehension, goal-oriented dialogue and visual question answering.

Managing changes and updates

Do you want a chatbot that you can easily update? It’s likely that you’ll want to tweak your chatbot as you learn how your customers interact with it. Some chatbot platforms offer a user-friendly visual UI so you can easily see how the chatbot is performing and also adjust how it functions.

Are you looking for the right chatbot?

If you are exploring the market and evaluating your chatbot options, we can help. Our consultants have deployed chatbots in as little as 8 to 12 weeks. And because we understand the unique pressures and opportunities of contact centres, we can ensure that you get maximum value from your investment, and a chatbot that improves the customer experience.

Read more about our chatbot solution.