What is this about?

This post takes you through my journey as a conversation designer and the collective effort of my team in designing the conversational experience of “Ewa”, our chatbot for SAP EarlyWatch Alert. This will give you an overview of the design process we followed from our initial design thinking workshop with customers and SAP colleagues leading all the way up to building our 1st version.

Let’s talk about Jay, our user persona

*TQM : Technical Quality Manager

To bot or not to bot?

Let’s face it. Chatbots are taking over. Stats predict a bright future for chatbots and virtual assistants. But is that a good enough reason to jump on the chatbot bandwagon?

When we set out to design the conversational experience of Ewa, it wasn’t about following industry trends and giving in to the bot hype. It was about solving real problems.

Like any design process, it all began with user research activities.

  • Interviews with Technical Quality Managers (TQMs)
  • Design Thinking workshops with colleagues and customers.

Let’s Look at Jay’s user story


“I would like to assess the risk for customer EngineWorks, focus on the critical alerts and take recommended action.”



Here’s how Jay would go about it on the EWA Workspace:

The workflow in the EWA workspace


What if, all Jay had to do instead was..just ask?

Here’s the task reimagined with Ewa, our helpful chatbot :

Before making the bot, we had to decide what it was going to be all about. The bot was not aimed at being a one-stop solution for all things EWA. But it could help users with high-frequency use cases or at the very least be efficient with specific use cases. Based on our user research activities , we set focus on what the bot should handle:

  1. Risks & Alerts
  2. Capacity Analysis
  3. Performance Tuning
  4. Connecting user to an expert

Meet Ewa!

Once we decided on the scope, I began designing a personality for the bot which would impact the way the bot responds. Eventually, a bot persona was formed for Ewa, using the SAP Personality Framework as a reference for its traits and coming up with a list of tasks Ewa could perform.


Fake it till you make it : “Wizard of Oz” Method

What makes conversation design challenging is that one cannot always predict the various paths a conversation might take. At any stage of the conversation, there are various ways the conversation could branch out. In order to get an idea of how users might interact with Ewa, we used an age-old method named after a timeless Hollywood movie, “The Wizard of Oz”. This involved mimicking the persona of Ewa and interacting with our test subjects while keeping in mind its personality traits.


Conversation Flows and “Happy Path”:

The next step was to start mapping out the conversations and design the ‘flow’ – basically a tree of possible exchanges between the user and Ewa. Though I had already begun with an initial flow, conducting the Wizard of Oz tests helped refine the flows.

I created the flows on Mural, a tool which provides an ‘infinite canvas’ that also allowed others to collaborate on the content.

I began with a ‘Happy Path’ which is basically an optimal flow where everything works the way it’s intended to. In reality, the user is not always taken down the happy path and there are always situations where the flow breaks. In such cases the bot would have to redirect the user to a safe zone in the conversation flow instead of breaking down entirely. One way to avert this was to direct the user to a human expert instead.


Sketching the user interface

With the conversation flows in place, I started working on the user interface. Though the interaction would mainly be text-based, I explored other UI elements that could be appropriate to display content within the chatbot’s limited screen real estate such as lists, cards etc. At each stage of the conversation, I explored ways the chatbot could progress and lead the conversation, such as adding prominent actions for each item in a list, providing logical action buttons after the bot’s response which could eliminate the need for the user to type in each step of the conversation and potentially keep the user in the happy path.

Early Sketches



Final design vs practiced chat with TQM

Seeking expert human assistance


Wrapping up!

Not logged in