Difficult to monitor data changes and the user journey
“Frequent changes in the data make it difficult to get a clear overview, and it’s unclear how users are utilizing the data.”
AI Data & Content Management SaaS Platform
Kakao i Dataverse
Building an AI-ready data platform from fragmented tools
Kakao i Dataverse is a SaaS platform for managing content and data. It combines metadata and vector search with Retrieval-Augmented Generation to produce AI responses.
- Team & duration
- Cross-functional SaaS team · 6 phases / 18 sprints
- Company
- Kakao Enterprise
Project scope & backgroundOpen details
Kakao i Dataverse is a SaaS platform for managing content and data. It combines metadata and vector search with Retrieval-Augmented Generation to produce AI responses.
Evolving from B2B2C projects, it helps non-technical administrators manage, connect, and operate AI services with minimal engineering effort.
CompanyKakao Enterprise ↗
Role / SkillsProduct ManagerProduct Designer
DeliverablesProduct Strategy RoadmapUX DesignAPI Architecture
Project dateMar 2022 - Aug 2023 · 6 phases / 18 sprints
Context
Context & Background · Market Signal
Generative AI adoption accelerated from experiment to mass usage
The rapid growth in daily active usage signaled that companies would soon need a practical way to prepare, govern, and connect proprietary data to AI services.
From 2M to 12M DAU in the measured period6×
growth in daily active users
- 2M
DAU
November 2022
- 12M
DAU
January 2023
What this changedProduct implication: move beyond isolated chatbot projects toward a reusable data and content operations platform.
Source: SimilarWeb, ARK Invest
Context & Background
Three signals pushed the service beyond isolated chatbot projects
Fragmented tools, integration demand, and Gen AI adoption made the opportunity for a shared data platform clear.
- 01
Fragmented data management tools
Eight separate data tools increased the burden of operating in-house bots.
Approximately $2.5M in annual operating costs - 02
Growing demand to connect data
Partners needed their own data to connect seamlessly with their services.
100% of Kakao i Connect partners requested integration - 03
Strong demand to adopt Gen AI
After ChatGPT, partners wanted to connect legacy chatbots with generative-AI models.
90% of partners intended to adopt a Gen AI bot
Source: Partner interviews and platform-operation analysis
Problem Statement & Define Needs
A key issue was that bots and data administrators struggled to manage large volumes of data and found it difficult to use that data in bots or services.
04 Data Dashboard & Data Flow Analytics
Provide clear visibility of data changes and statistics with UX flow.
05 Challenges in leveraging data for AI bots
“We want to improve our bot service or adopt Gen AI using data, but we’re not sure where to start.”
Bot/Service Connection & RAG AI Bot Builder
Manage service and bot in one place and build Gen AI with an easy UX.
06 Limited development skills made data management difficult
“As a data manager, I only handle statistics provided by developers, so it’s difficult to manage the data properly.”
Low-Code Data Management System
Serve data and manage search engines through a simple UX.
Role & Decisions
Ownership
The decisions behind the product direction
- Mandate
- Define the platform direction, validate it with operators and AI specialists, and lead six phases of delivery.
- My decision
- Combined data governance, analytics, low-code operations, and RAG delivery in one SaaS platform.
- Trade-off
- Balanced no-code simplicity with the governance and integration depth required by technical teams.
- What I led
- Product Manager · Product Designer. Product Strategy Roadmap, UX Design, API Architecture.
Hypothesis · HMW Approach
“How might we enable bot managers to easily manage, operate, and improve both legacy bots and new generative AI-based bots?”
Research & Strategy
User Research
Stakeholder interviews, proof-of-concept projects, and defined needs
Five stakeholder interviews
- Kakao AI Lab LLM model researcher
- Kakao Search Engine lead engineer
- Kakao AICC service product owner
- NH Investment chatbot manager
- Hey Kakao customer-service agent
Three proof-of-concept projects
- Integrated generative responses into the Hey Kakao and Kakao Work stock bots
- Implemented generative-response functionality in AICC SaaS
- Implemented AI-driven answers for the NH Investment & Securities stock bot
Four defined user needs
- Codeless data and content storage with real-time integration
- A private repository that keeps company data out of model training
- Real-time analytics for flow, drop-off, and conversion monitoring
- Role-based CRUD permissions for different team members
User Testing · POC Projects
Stock knowledge search and generative-AI bot
Testing with real end users is critical in B2B SaaS. Because access depends on partner adoption, we used POC projects to bring partners onboard and validate the product with users.
We continuously collected quantitative and qualitative feedback from a stock-knowledge search and generative-AI bot.
- Time on Task
- Success Rate
- Process Satisfaction
- Data transaction stability: volume and speed
- Number of service connections
POC setup and user feedback
Users requested CSV bulk upload, clearer tooltips for non-developers, and simpler deployment after an easy metadata setup.
Supporting detail2 additional sections
XFN Collaboration · Communication Guidelines
Clear communication guidelines kept goals and decisions visible
Share visibly
Share issues and decision points in the project channel with @all mentions.
Meet without delay
Request a follow-up meeting immediately when further discussion is needed.
Close the loop
Review and respond to questions or requests before the next daily scrum.
XFN Collaboration · Sprint Operating Cycle
A repeatable cadence aligned goals, surfaced issues, and closed each sprint
- 01
Sprint Kickoff Meetup
Align the sprint goal and make priorities visible to the full team.
- 02
Module Daily Scrum
Share module progress, raise issues early, and assign follow-up actions.
- 03
Team Weekly Scrum
Review cross-module progress and resolve dependencies before review.
- 04
Sprint Review
Review the sprint process and set the next sprint goal.
Review findings feed directly into the next sprint kickoff.
Product Evidence
Service Flow
A REST API that connects Dataverse directly to a customer's service
Customers can connect channels to reusable search, content-management, and AI-learning capabilities without rebuilding the underlying platform.
- 01
Service channels
- KakaoTalk
- Voice assistant
- Jira
- Slack
- LINE
- Custom channels
- 02
Kakao i Dataverse
- Metadata collection
- Structured search
- Vector search
- OCR
- STT
- Content authority
- Test playground
- Response adjustment
- Prompt engineering
- 03
LLM processing
- Meta
- Kakao Brain
- OpenAI
Outcomes
Kakao i Dataverse
A low-code SaaS platform for storing, managing, and integrating data and content, with a path to generative AI.
Integrated Dashboard
Track all data and content in a single view with real-time status cards and one-click access to detail.
Tooltips help non-developers understand important features. Lists and status indicators expose sync expiration, connected bots and services, expiration dates, and current status.
UX Flow-based Data Analytics
A Sankey chart visualizes how users access and interact with data step by step, revealing usage patterns and engagement drop-off points.
Hover states reveal summary statistics without leaving the page. Line, pie, bar, and other chart types make trends in large datasets easier to understand.
Search Index & Metadata Management
Create custom search indexes and rules to operate a proprietary search service.
Upload keyword and morphological data in bulk, then create search rules by selecting representative datasets and adding bubbles.
Generative AI Management
Connect multiple bots to data on one page using an App Key. Separate test and production deployments support safer integration testing.
Stored data can connect to structured search, vector search, and LLMs. Prompt testing previews answers before launch, while annotation supports ongoing generative-bot management.
Impact
Data & KPIs
Behavior, transactions, and connected services measured sustainability
We defined KPIs around user behavior, data transactions, and connected services so the platform could be improved and operated sustainably.
Ten partners represented mobility, finance, retail, public-sector, and IoT domains, including Hyundai Motors, NH Investment, GS Retail, Daiso, Sejong Self-Governing City, Samsung, and Komex.
User-behavior metrics
Button clicks, flow success, time on task, and real Voice of Customer feedback
User-flow success
Time on task
Monthly data transactions
Partners across multiple domains
Connected bots and services
60 Hey Kakao platform bots, 10+ partner bots, and 10+ partner services