How do you order Coffee?

Transforming Language Learning with GenAI

by Team 5

Agenda

1. Introduction
2. Environment Analysis
3. User needs discovery
4. Feature proposal
5. A/B testing
6. Next steps

3

Duolingo is market leader with a Market Cap of 9B

Environment Analysis for Duolingo (3C)

4

Company�The company offers a platform for learning different languages through bite-sized lessons designed to feel like games for over 40 languages.

​

Freemium model that allows users to access basic features for free, with premium subscriptions available for enhanced learning experiences.

​

Customers�Targets broad audience of language learners, including students, professionals, and casual learners.

�Over 500 million registered users worldwide, spanning across different age groups, from children to adults, and diverse geographic regions.

​

Direct competition from other apps like Babbel, Rosetta Stone, Memrise and indirect from language schools and learning apps.

Competition

What our customers are saying?

5

User feedback on Reddit and Play Store frequently questions the usability of Duolingo for fluent conversations

Focusing on Customer Needs

6

Lack of Conversational Confidence

A user experience study published on Statista found that 80% of learners expressed that not being able to practice with native speakers was a significant barrier to continuing their subscription to language learning services

​

Gap in Accent and Diction Training�UGC indicates that learners find it challenging to mimic native accent and diction. ��Need for personalised solutions�Traditional one-size-fits-all solutions are not working.

​

User : A student travelling to a new country on exchange program

​

Objective: Improve Late Journey User Retention

Solutions Explored:

• Agent Based Chat�
• Peer-Peer Community Challenge �
• AI enabled voice chat�

Prioritization Rubric

• User confidence Impact�
• Retention�
• Effort to build�
• Scalability�

With high scalability and quick to market and reasonable retention - Introducing DuoLE

7

8

DuoLE

डुओले

ثنائي

দুয়োলে

Дуолей

多樂

დუოლეი

דואוליי

デュオレー

ಡ್ಯುಲೇ

ឌុយឡាយ

듀오레이

ਦੁਆਲੇ

DuoLE is an GenAI enabled voice chatbot capable of having real time conversations

• DuoLE – Duolingo Language Engine will be powered by proprietary language model and be able to have natural conversations with users on a wide range of topic depending on users' proficiency
• DuoLE will use a voice modulation framework to sound like a native speaker with special focus on accent and diction

​

9

A look into the product mock-up

10

10

Practice real world scenarios

Practice with AI

Get a prompt based on lessons progressed so far and converse

Tap here to start a conversation with Gen AI bot

Motivation Card

A motivational nudge with tangible output of using the above feature

DuoLE will be integrated with existing service to offer highly accurate voice chat

• Develop a POC with Data and AI teams�
• Test the system within the company and gather customer feedback�
• Perform integration with existing stack – reduce developer time�
• Ensure model works on any language – can be used for all present or future offerings�
• Enable integration with both mobile and desktop app

11

98%

Aim: Achieve accuracy in understanding the user input and responding ¹

1. Run 1000 hour of conversations in French

Implementing A/B tests for Continuous Improvement

• Our A/B tests will focus on 3 key areas
• Engagement
• Retention
• Monetization
• Each test will have a hypothesis and metrics to prove/ disprove the hypothesis
• We will run multiple A/B tests at the same time because running separate tests will take too much time and give isolated results
• All tests will be run for a duration of 2 weeks post which we will decide if we have enough data to evaluate the hypothesis
• The user base for all A/B tests will be 1% of customers randomised demographically

12

A/B testing with different voices

Area - Engagement

Hypothesis - Users are more receptive to certain voices than others

Variable - Use 4 different voices

Metric - Average session duration per user per day

Cohort - 1% of randomized French learners worldwide

Duration - 2 weeks

13

Young woman

18-23 year old

Middle aged woman

40-50 year old

Middle aged man

40-50 year old

Young man

18-23 year old

Q&A