Designing Multilingual Capability for the Chatbot
To comply with the non-disclosure agreement, I have excluded and obfuscated any confidential information in this case study.
Our client is an overseas gym that wants to employ the AI Chatbot to facilitate training sessions with their customers.
After deploying the Chatbot on the client mobile app, we start to notice most of the users are speaking to the chatbot in the native language.
Here's the problem: our Chatbot was only capable of communicating in English.
After discussing with the client, we agreed that a user interview would help us understand the situation better. We then started recruiting participants based on a list of criteria.
Here are some examples of the interview questions:
Have you encountered any issues using the APP? How did you overcome them?
Did you face any difficulties when communicating with the chatbot?
Did you have any difficulties speaking English with the chatbot?
If the chatbot could speak [language], how would it help?
Here are my discoveries during the user interviews:
The majority of the chatbot users are not comfortable with speaking English
Although some may understand English, they still prefer to communicate in [language]
Some of the technical terms were too difficult to understand
The client existing mobile app (where the chatbot resides) is in the native language
Here are some considerations:
Avoid having the client rebuild the content
The proposed solution should cater to other languages
Can our information architecture support it?
I conducted a competitor analysis so we could learn from the existing solution in the market.
There are many excellent examples out there:
Creating different bots to handle different language
Using machine translation
With my limited technical knowledge, I know I would need to tap on the team's skills and expertise. I convinced the product owner to invest some time in a design sprint and how it would help us validate the ideas and reduce the risk of building the wrong feature.
Pre Design Sprint
I gathered the stakeholders which include 6 developers, 2 QAs, 1 Product Owner, 1 Sales Manager, and 1 Business Analyst. I introduced what a design sprint is and how we can use it to answer critical business questions through design, prototyping, and testing ideas with customers.
Design Sprint Day 1
I created the Persona and User Need Statement to help the team get a clearer picture of the situation, introduce the affected user profile, and the difficulties that they were facing.
The business analyst and sales manager were invited to articulate the problem space from the business angle and to fill up any gaps we had.
User Need Statement
The product owner and I started this section by mapping out the customer journey, which was not foreign to the team. (We have done it multiple times together with the team)
The team evaluates everything they learned so far and were given some time to identify the pain points and opportunities.
Each participant was asked to come up with at least 2 HMWs, followed by a dot voting session to prioritize the specific focus for the Sprint.
How might we & dot voting
The Problem Statement
How might we configure chatbot easily with minimal duplicated effort and allow it to communicate in the language being spoken by the user?
Note: In hindsight, I admit that the problem statement was too "huge". It would have been more ideal if the problem statement was more specific and focused. However, this was our first time, and we are still learning!
Design Sprint Day 2
Crazy 8s. A fast sketching exercise that challenges people to sketch eight distinct ideas in eight minutes. Of course, I didn't do 1 minute per sketch, that's too stressful.
I do also face resistance from the developers, many claimed that they couldn't sketch. And that's okay! I allowed them to write instead. My focus here is to get the team members to be comfortable with sharing their ideas and not forcing them to draw like an artist.
Solution Sketching & Lightning Demo
We continue with a solution sketching in which each member was given three frames to further illustrate their favorite idea. We then had each member do a lightning demo of their idea. Here's an example:
Solution sketching & dot voting
A dot voting was then carried out to finalizes the direction or concept to be prototyped. Next, we moved on to storyboarding.
Design Sprint Day 3
On day 3, I worked with the product owner to sketch out the idea in storyboard format. We discussed the type of scenarios that are needed to showcase the solution and the possible failure scenarios such as:
Setting up a new language for bot
View content of a specific language
Editing content of a specific language
Selecting language on the Chatbot UI
Testing content of a specific language
Design Sprint Day 4
On day 4, I started prototyping the essential pages for user testing.
Using our Taiger Design System, I was able to create the prototype very quickly.
-- Content Hidden --
On the Following Week
I tested the prototypes with our internal users and the clients.
These are some of the tasks we tested with the participants in the user testing:
Add a new language for the chatbot
Create an intent
Create intent content for both English and [Language]
Test the intent on chatbot UI, for both languages
Try asking a question in the language that has not been set up in the chatbot
User Testing Feedback
Here's what we learned:
Adding new language for the chatbot was easy and simple
When I create a new intent for a chatbot that has more than 1 language. It was not obvious that I need to create content for the other languages as well
It's a hassle to create content for the different languages
The accuracy of machine translation is a concern
The feedback was later used for further iterations.
The design sprint ended with a validated concept to improve on. We have made great progress!