Kakao Enterprise · AI Voice Assistant Platform
Hey Kakao
Rebuilding a voice assistant around the needs hidden in failed conversations
Prioritized data-backed answers and child-friendly experiences over expanding generic fallback scripts.
Hey Kakao is Kakao's voice AI assistant platform. Kakao is one of the largest IT companies in South Korea.
- Team & duration
- Cross-functional AI product team · 18 months
- Company
- Kakao Enterprise
Project scope & backgroundOpen details
Hey Kakao is Kakao's voice AI assistant platform. Kakao is one of the largest IT companies in South Korea.
The platform is available through mobile apps, AI speakers, IoT control pads, and in-car systems. It serves more than five million users and also supports B2B chatbot solutions for partners including Hyundai Motors, Samsung Electronics, and Korea Investment & Securities.
User engagement and retention were declining while fallback responses to unexpected utterances were rising rapidly. Our mission was to increase engagement and encourage more meaningful interactions.
CompanyKakao Enterprise ↗
Role / SkillsProduct ManagerLead Product Designer
DeliverablesProduct PolicyUX DesignAPI ArchitectureData Report
Project dateFeb 2022 - Aug 2023 · 18 months
Context
Context & Problem
The response accuracy rate dropped below 85%
Fallback response rate
Unexpected utterances not covered by existing scenarios
Fallback Analysis
Most failed conversations clustered around four underserved intents
The fallback volume was not random. Four recurring needs accounted for 70% of the analyzed utterances, giving the team a clear product-prioritization signal.
Daily average analyzed10,000
fallback utterances per day
- 25%
Kids content
Largest unmet intent
25% - 20%
Stock price
High-value information need
20% - 15%
Kids education
Recurring parent-and-child need
15% - 10%
Character content
Engagement opportunity
10%
What this changedProduct decision: prioritize data-backed answers and child-friendly character experiences instead of expanding generic fallback scripts.
Source: Hey Kakao conversation-log analysis
Role & Decisions
Ownership
The decisions behind the product direction
- Mandate
- Set product policy and experience priorities across conversation analysis, new assistant modes, and delivery.
- My decision
- Prioritized data-backed answers and child-friendly experiences over expanding generic fallback scripts.
- Trade-off
- Focused delivery capacity on high-value answer types and children’s journeys instead of adding more generic fallback copy.
- What I led
- Product Manager · Lead Product Designer. Product Policy, UX Design, API Architecture, Data Report.
Hypothesis · HMW Approach
“How might we ensure that users do not feel disappointed with the Hey Kakao service and can engage with it more?”
We tested simple bot responses, tracked user satisfaction, and worked to reduce re-engagement time while stabilizing DAU. The A/B test informed the user research plan, MVP, and phased roadmap.
Research & Strategy
User Research
10,000+ utterances, three user segments, and five stakeholders
10,000+ utterance data analysis
- Conversation-start utterances
- Keywords used at the point where conversations ended
- The number of interaction steps users completed with the service
Three user segments
- Parents of seven-year-old children
- Stock-trading users
- Drivers who own a car and have children
Five stakeholder interviews
- Hyundai Motors designer
- Kakao Entertainment music-streaming staff
- Kakao Enterprise AI Lab engineers
- Iconix staff from a children's-content company
- Kakao Mobility UX staff
Four defined user needs
- Personalized experiences for children's content
- Integrated stock information, financial data, and economic news
- Interaction with Kakao Friends characters
- Easy control and calling while driving
Product & Delivery
Hey Kakao 2.0 Phased Roadmap
A three-phase plan covering minimal needs through hidden needs
Phase 1 · Answer with Data
- Connect real data with bot
- Data and REST API architecture design
- Additional text information with external link
Phase 2 · Answer with GUI
Answer with visualized information
Make UI type
- Stock Price UI
- Market Index UI
- Fairytale Contents Play UI
- Reward Sticker UI for Kids
Phase 3 · New Mode, New Bot & Features
Kids mode
- Integrated kids-related bots, including fairytale, zoo, and game bots
- Kids and character voice TTS
- Embedded educational content
- Character doll with built-in Voice AI
New Bot & Features
- Integrated third-party services, including music streaming, podcasts, and news
- Drive mode
- Car infotainment built with Kakao AI
What shipped
Hey Kakao 2.0
Four launches answered the four intents the fallback analysis had surfaced: a stock bot, Voice Note, a Kakao Friends character bot, and Mobility Mode.
The two that moved engagement most are set out below — Stock Bot against the stock-price intent, and Kids Mode against kids content and kids education.
Stock Bot
Rebuilding the stock bot around price first, then everything around the price
The renewal contributed a Daily Interaction increase of more than 15%.
A price-led UI carries the answer, with a follow-up menu and voice control for reading the news and detail around it rather than asking again from scratch.
- Designed the data pipeline so US stock prices, not only Korean ones, could be wired in through a real-time API.
- Set the content-search pipeline policy that decides which related news is surfaced with a price.
- Opened a B2B2C chatbot business off the back of the stock bot.
Kids Mode
Turning kids content into something a child returns to
Improving the existing kids-content bot and adding reward and education features — praise stamps, and a listenable zoo the child can interact with.
The work brought in a user segment the assistant had not held before, and improved Daily Interaction by 5%.
- Attracted a new user segment: households with children aged seven and under.
- Improved Daily Interaction by 5%.
Multi-Channel Scalability
Supporting users' lifestyles across speakers, IoT, and mobility
One bot had to arrive well on every surface it reached, so each channel got its own rule. The speaker has no display, so after an answer it offers the next step out loud — read a stock price, then ask whether to play related Kakao news. In the car, controls were reduced to the largest, simplest buttons that would not compete with driving.
Impact
Impact
Improved response accuracy and user engagement
New bots, new features, and Kids Mode improved response accuracy and encouraged deeper engagement with Hey Kakao 2.0.
Response accuracy
Recovered from 85% after the new bots and modes shipped, covering 2,500 kids-content utterances and 3,000 stock-information utterances.
Share of monthly active users
Stock Bot accounted for 6% and Kids Mode for 4% of monthly active users.
New B2B opportunities opened
An AI-textbook proof of concept with the Gyeongnam Office of Education, plus B2B2C chatbot expansion with NH Investment & Securities, Korea Investment & Securities, and GS Retail.