The Atura Process
At Atura we follow a simple yet thorough process when designing our bots. We like to keep the process as streamline as possible, avoiding verbose and drawn-out spec docs. Instead, we prefer reader-friendly and lightweight documentation, visualising concepts and language flows through diagrams wherever possible.
In this blog series, we'll be describing how we do what we do in the order that we do it. We've broken our process down into six steps:
- Understanding the context
- Determining core intents
- Utterance generation and intent mapping
- Creating the bot persona
- Designing conversation flows and dialogue
- Building, testing, improving!
Let's start at the beginning:
Step 1: Understanding the Context
Where will our bot live?
Before we can dive in to the fun stuff – designing and developing the bot – it's crucial we first analyse and understand the client's current system and processes. We need to find out what it is the bot will be doing exactly – which processes will be taken over by an AI assistant, and how will an assistant assimilate into the client's existing structure? What kind of functionality are we talking about here?
Our aim is to design a bot that helps clients reach maximum efficiency. Our bots should smooth out critical pain points in their system, such as tedious or counter-intuitive admin processes, frustrating user experience scenarios – or anything causing customer loss or discontentment. These pain points, along with insights into the client's domain lifecycle, help us sketch up a plan for our bot's functionality.
What will our bot do?
Research and experience show that bots are most successful when you begin with a narrow scope of functionality that you can do very well, and then later broaden the scope from this steady starting point. We incorporate this strategy in our approach to bot development. We break down core domain functionality into subsets according to the natural lifecycle of that domain.
In asset management, for example, the client lifecycle could be broken down into the following subsets: advice, onboarding, viewing tax certificates, balances, and statements; and ending with some core functionality like making additions, switches, and withdrawals. Based on this, we can map out the starting point of our bot as well as the later instalments to follow, all in alignment with the natural domain lifecycle:
Once we have a good understanding of this overall context and development plan, we then move on to the specifics of our bot solution. This involves fleshing out details such as the client's business culture, brand and client base. We can then create specific user stories, where we attempt to consider every type of bot user, piece of bot functionality, and bot scenario.
Our user stories inform us throughout the design phase, setting the context for the remainder of the design process. Check out our next post where we outline the following step: determining core user intents.