Chatbot Building Best Practices

Grant Patten, Digital Media & UX Specialist, CSIO | BC Broker, Dec. 2017

Chatbots are computer programs designed to simulate natural conversation with human users, and innovative companies in the insurance industry are using them more and more to improve customer service. Chatbots typically offer a first point of customer contact, providing instantaneous responses to many types of simple inquiries and connecting users to the appropriate staff member to complete a transaction. A 2017 study from Juniper Research across multiple industries estimated that the average time savings per chatbot inquiry when compared with traditional call centres is 4+ minutes, and the study predicted that chatbots will be responsible for cost savings of over $8 billion per year by 2022.

Increasingly, companies are deploying chatbots on popular social networks and messaging platforms. In particular, Facebook founder Mark Zuckerberg made waves in April 2016 when he opened Messenger to chatbot developers. By linking chatbots to Messenger – a service with approximately 1.2 billion monthly active users as of Fall 2017 – companies have immediate access to a huge user base to engage both existing and potential customers.

This article provides an overview of some lessons learned by CSIO staff in developing a chatbot of their own, including advice and best practices that brokers can use to streamline the process of implementing a chatbot in their own operation.

Define a Minimum Viable Chatbot (MVC)

Before you dive into building anything, it’s important to first have an internal discussion around what you actually want your chatbot to be and do. What does the success of Version 1.0 look like? Make a list of desired features and capabilities, identify which ones are essential to your MVC, and build to those specifications first. Your employees will find the task much more manageable, and the experience they gain will better equip them to develop subsequent iterations with more advanced capabilities.

In defining your MVC, it is useful to identify a topic that would be of interest to the greatest number of customers and build to serve that need, e.g., providing quotes. For example, at CSIO we decided to focus our Version 1.0 chatbot on CSIOnet, the network connecting over 36,000 brokers with their carrier partners to exchange policy data and eDocs daily. This focus enabled us to target our members’ highest need as quickly as possible, and provided us with the opportunity to expand our chatbot’s capabilities in the future. What does your MVC look like? The narrower the focus you define for your chatbot, the more effective it will be.

Your MVC should also define the medium it will appear on – your website, Facebook Messenger or other messaging apps, etc.

Select a bot platform

Building a chatbot generally involves leveraging a bot builder platform – and there are many to choose from. CSIO followed a competitive analysis process to select the vendor for its chatbot, analyzing platforms such as Chatfuel, Flow XO, Motion AI, ChatterOn, and Gupshup. Analysis criteria included:

  • Price
  • Artificial intelligence (AI) component
  • Branding control
  • Analytics / reporting
  • Live chat integration (so staff can take over chat at any time)

You’ll likely want to select a chatbot platform that is easy to use and does not require much custom programming. There is also the option to not use a bot builder platform at all and instead hire a programmer or team to custom code your chatbot from scratch. This approach will likely allow for more bot customization, but will definitely be more time consuming and expensive than leveraging a platform.

Establish the conversation flow

Before you start building your bot, spend some time thinking about how you expect the conversation to flow when users are interacting with your bot. How will your chatbot greet users? At what point will your chatbot branch from a generic conversation to more specific topics? You can leverage one of the many flowchart tools (such as Lucidchart or Microsoft Visio) to map out considerations, including:

  • When users may type their responses (open input) versus presenting them with pre-defined options to choose from
  • When to show search results and what those results should include for each topic
  • What the calls to action (CTAs) should be
  • Ideal resolution for each interaction
  • Error handling (when the chatbot fails to understand a user’s intent)

Write the chatbot script

After mapping out your chatbot’s core conversation paths in a conversation flowchart, a best practice is to write, revise and finalize your chatbot’s wordings (i.e., questions & responses) in a separate script document before building the bot itself. Doing so before building can save development time, ensuring your bot is built to proper specifications rather than making heavy revisions during development or even after launch, which can be complex and difficult.

As well, writing the script is an opportunity to determine what kind of personality you want your chatbot to project – serious and businesslike, or a little fun and playful? The tone and persona of your chatbot can affect the overall user experience while interacting with it, so ensure that it meets the needs and expectations of your target customers.

Build the bot

After writing the bot conversation script, we began the process of building the chatbot using the bot builder platform we had selected (Chatfuel). This was done by adding the text responses and connecting them to anticipated user queries – no coding required. We then began the process of internally testing the chatbot Version 1, as well as conducting some usability tests on the chatbot with broker members of CSIO. The feedback gained from this testing process was then incorporated into the chatbot before soft launching the bot on Facebook.

Chatbots & the broker channel

Of course, chatbots cannot serve as a replacement for insurance brokers. This technology should be regarded as a potential complement to brokers’ operations. A chatbot can be a tool to streamline initial contact with customers before handing off the interaction to a human operator to complete the transaction.

Brokers commonly report that the vast majority of customer calls – whether for personal lines or small commercial lines – are to either make very basic policy changes, or to ask a simple question requiring only basic contextual data about the caller (e.g., policy endorsements such as adding or removing a vehicle, requesting a certificate of insurance or asking a question related to current coverage). Chatbots can address inquiries like this quickly and effectively, freeing brokers to focus on more specialized and value-added tasks.

As brokers seek to transform the customer experience through digital tools, they should consider chatbots and the significant benefits they can provide to both sales and support processes. CSIO plans to deliver an educational webinar about chatbots for the P&C industry in 2018 – keep an eye on for details.