Here on Voice Crumb, we’ve often been open to discuss the stories behind some of our favorite projects we’ve worked on.
In this article, we’ll zoom in on the process that we’ve been following across all of these and many other projects to successfully bring conversational products to life.
While every case is different, there are some steps we generally take to ensure the final product is as good as it can get. Broadly speaking, there are 5 key phases we go through:
- Use case definition
- Conversation design
- Implementation & testing
- Launch & optimization
Let’s discuss these in more detail.
Step 1: Research
Before we start jotting down ideas and putting our heads down to work on the bot itself, it’s crucial that we immerse ourselves fully in the project and brief.
If we’re working with a partner, that means we’ll need to flesh out its world and values. We’re immersing ourselves in their organization, to understand how they think and what their goals for the project are.
As we share in our report Leading with Conversation Design, before jumping into anything it is crucial to understand what the starting point in our partner’s automation journey is. Broadly speaking, there can be 3 scenarios:
- New iteration of an existing project. In this case, the Research phase will be focused on reviewing the transcripts and available reports on the performance of the bot, to improve the existing features and possibly identify new use cases to implement.
- First assistant to automate a part of the tasks that fall under an existing unit, like customer support. While you won’t have customer data on bot interactions, things like customer support transcripts can be very informative. They’ll teach us how customers talk about products, what they ask about most frequently, and so on.
- Tabula rasa. This is the case where our partners are starting off from scratch on their automation journey. FAQs can be a good starting point and further research will be even more crucial than in other cases.
In this Research phase, we’ll try to get a full picture also on the options that are out there for our partners, so that we’re able to support them in making informed decisions about their strategy. We'll look into other bot success cases that might compare in some ways to the one we’re working on.
Qualitative and quantitative research will be in order to understand our target audience and their pain points, but also, in some cases, to identify how scalable the project might be and to confirm that there are signs in the market indicating a clear need for automation.
If we’re working on an internal project, the goals and values will be naturally clearer to us, but the research phase will still be very much present.
Step 2: Use Case Definition
Based on the research we’ve conducted and with full awareness of our goals, we’ll start conceptualizing what this means for our bot. The research will hopefully point to us which areas are still rather under-automated and support us in identifying opportunity areas and use cases to prioritize.
First, we’ll try to identify what are all the possible areas and use cases that would benefit from being automated using conversational AI. If the project we’re working on is of a more creative nature, this would be the time for idea generation and workshopping with our partners.
Whether the project is creativity- or data-driven, we’ll, then, get to the prioritization of all the possible applications we’ve mapped. Extracting 3/4 criteria from our research and initial goals, we establish priorities for the use cases we’ve identified and start elaborating a CAI roadmap.
Let’s say our analyses show that 70% of conversations happen on WhatsApp and 80% of them are related to 5 recurrent themes, or questions.
Based on the criteria of frequency of use case, cost/labor savings, and admitting there is a possibility to automate these tasks, it’d make sense for the first version of the bot to be focused on WhatsApp as a channel, and these 5 themes as use cases.
So, what’s with all the other use cases and areas of opportunity we’ve identified in the initial phase?
That’s where a CAI roadmap comes into play. It would be impossible (and if even if it were possible, it would be unbeneficial) to automate all possible use cases right from the get go, in the first launch of the bot. Creating a structured roadmap helps us define where to start and what few channels and use cases to focus on in the initial phase, without forgetting where we eventually want to get.
Step 3: Conversation Design
It’s now the time to get to design. The first thing we’ll do in this stage is defining the bot’s persona, as we describe in our Road to Conversation Design report, focusing on interaction goals, character traits, name of the bot, and tone of voice.
Depending on how complex the project is and on the way we’ve set up our collaboration with the partners, we might approach this stage in different ways:
- Directly drawing out flows and drawing up a testable prototype on Voiceflow
- Defining the general architecture first, then detailing the individual blocks and flows
- Start from writing out a happy path to get a sense of the TOV and the way we want to present the bot’s feature, then, follow with the flows
For Wheely, the car repair shop assistant we developed on the Kore.ai platform, we followed this latter approach.
Based on the two basic use cases (book a car service / ask for help with car problems) and the different types of users (new / recurring) that Wheely might be welcoming, we wrote out the core dialogues for the bot. The design was adapted to suit the different informational needs and emotional states that the user might find themselves in, based on the use case and their status (new vs. recurring).
Step 4: Implementation & Testing
Once all the flows have been defined, everyone in the extended team has tried and is happy with the prototype, we move on to the implementation. We’re big believers in trying to get as much feedback as possible on the prototype, so these last steps can move on as smoothly as possible. Depending on the platform we used in the design phase and the channel we want the bot to appear in, the hand off might require more or less effort. This is also where we make sure all API calls with external systems, links, and any other interactive elements work properly.
When we feel confident about the first version of our bot, we open up the floor to beta-testing. This is where we get more feedback on how the bot performs in real life situations and we have to be prepared to find possible issues and optimize accordingly.
Step 5: Launch & Optimization
Finally, we get to the launch phase, where the bot goes live on the defined channels and real customers start interacting with it. While it might be tempting to think this moment defines the end of this process, the reality is quite different. After launch, it is crucial to keep transcripts and dashboards monitored and start iterating, as soon as you start seeing areas of potential improvement.
As the first use cases are delivered and optimized, we can go back to the CAI roadmap we’d drawn up in Step 2. At this point, we’ll consider how to move on and work on the use cases and channels that we’d planned for the next phase.
Inspired to explore customer support automation for your company? We can help.
And if you’re still hungry for knowledge, follow us on LinkedIn for weekly updates on the world of conversational AI, or check out our interview with CAI platform evaluation experts, who shared their exclusive insights on how to choose the best technology partner to support you in your Conversational AI journey.