Chatbots give customers a ‘live chat’ experience through a computer without a human typing responses on the other side of the connection. They enable the use of conversational language instead of a traditional website form and it can be imperceptible to the user whether they are talking to a machine or a human. They can be powered by AI, but are typically underpinned by rules and decision trees similar to a traditional application adapted for the chat interface.

The benefits to businesses are clear and dual purpose; automate the simple call centre queries to save operational costs while at the same time offer instant, 24/7, 365-day availability to customers.

While chatbots are now widely used in other sectors (particularly in the retail e-commerce space), the mortgage industry has so far been slow to adopt, despite much talk of ‘robo-advice’ in recent years which has, nevertheless, failed to materialise.

There are many obvious applications for brokers, lenders and servicers which could help drive efficiency – crucial as the trend towards tighter margins does not look like abating.

The use case

Before rushing in to develop a proof of concept, spend some time identifying the use case. Building a chatbot is no different from building any other digital product so the business objectives, user goals, success metrics and business case need to be considered first. List all the points in the customer journey that can be replaced or supplemented with a chatbot.

Ask yourself:

• Is the idea a genuine fit for the technology or would it be better suited left to human interaction.
• How you will judge the success of your chatbot
• What the top queries that your customer services team have to answer every day are, and are the answers to these questions common and generic, or specific to the customer.
• If your chatbot needs to give specific information about a customer’s case (rather than generic answers from a fixed database), does the system that contains that information have an API?
• Will your chatbot enable open conversations or multiple choice answers? If the former, how will your chatbot handle out of context questions?
• How the chatbot should replicate the tone of voice of your brand.

Design the flow

 Chatbot conversations are designed in conversation flows. Start with pen and paper to visually map the conversational steps users need to follow to reach a milestone.

Later in the process you will build the detail and create a prototype that can be tested, but this initial flow design will ensure that there are no ‘dead ends’ for users.

Chose a technology framework

The good news for incumbents looking to roll out a chatbot is that the barriers to entry have become very low. There are many frameworks available at relatively little cost and so there is no need to reinvent the wheel. You can simply plug in and train an existing commercial framework.

Popular frameworks include Amazon Lex, Google Dialogflow, IBM Watson Assistant and Microsoft Bot Framework. While similar in functionality, they vary in cost per usage, in features and the level of technical knowledge required to setup.

At Dock9 we have primarily used the Microsoft Bot Framework. It offers two different routes for bot development: firstly its ‘QnA Maker’ enables you to quickly and easily create a bot based on a pre-configured list of questions and answers.

Create a Proof of Concept

Design and build a proof of concept that shows the minimum functionality required. To do this you will need map the questions and expected answers and intents from the user at each stage in more detail.

Tools such as Botframe enable non-technical people to rapidly build a demo ready for refinement.

Test, learn and refine

With your proof of concept now built and tested internally comes a pivotal point in the project: testing with real users.

The most successful fintech companies all have one thing in common: they regularly test their digital journeys with real customers (be they borrowers or brokers) and use the insights gained to optimise and drive conversions, avoiding the trap of making decisions on a “hunch”.

I have written before in Mortgage Strategy about the importance of testing and validating ideas as quickly as possible in the digital product design process. Testing concepts early with real users is especially important for chatbot projects and should ideally be undertaken within the first few weeks of kicking off a Chatbot development project.

Summary

With potential upsides so high, and barriers to entry now so low, those companies looking to improve their operational efficiency as well as improving customer experience would be well advised to explore how their company could utilise chatbots in 2020.

Mark Lusted, managing director, Dock9