MGS 2026 Musinsa Recap: How Amplitude AI Hit 70% Adoption
You rolled out an AI analytics tool, so why are your analysts still the only ones using it? If people don't trust the answers or don't see a reason to ask, data requests pile up on the analytics team again and results take days to arrive.
The Amplitude AI adoption rate is the share of employees who used Amplitude in the past month and actually used its AI features. Musinsa, one of Korea's largest online fashion platforms, raised that rate from the low teens to over 70% in three months (peaking at 76%) and cut analyses that once took analysts 2 to 3 days down to 15 minutes. This post recaps "Musinsa's Journey to 70% Amplitude AI Adoption," a MAIN HALL session at Modern Growth Stack (MGS) 2026 presented by speakers from Musinsa and Amplitude.
The daglo marketing team transcribed the MGS 2026 sessions with daglo and reread them to pull out what other organizations can borrow.
70% adoption came from design, not the tool
The Amplitude speaker compared AI without instructions to a pile of tools and lumber with no manual. Context, a maintained data catalog, integrations, and an internal rollout plan all need to be designed for AI to do real work.
Musinsa and its sister platform 29cm see more than 10 million monthly visitors. Data is abundant, but marketers and planners get stuck on questions like "which data shows what people click in the cart?" and end up filing a request with an analyst. Shortening that loop was the starting point. Today, people reach the AI in three main ways:
- Amplitude Chat: the AI agent inside Amplitude itself
- Claude or Codex + Amplitude MCP: connected from AI tools people already use
- Slack: tag Amplitude and get the same answers in chat
Earning trust: make verified dashboards the answer key
The key to trusted AI answers is pointing the AI at dashboards that have already been verified. Building a feature that works and getting people to believe its answers are two different problems.
First, the team registered key events and properties so the AI could understand two very different logging setups: GA4 for Musinsa and Snowflake for 29cm. Then they focused on analysis rules for each app surface, such as rankings, home, and cart. Each surface had slightly different rules and logs, and that knowledge lived only in analysts' heads.
- They listed every surface and added its dashboard link to the AI's instructions.
- When someone asks about "rankings," the AI checks that dashboard first.
- Each dashboard explains the surface, its purpose, key metrics, and caveats.
After labeling these as "official dashboards," the AI began recommending them on its own: "This question is best answered with this dashboard." Reusing existing assets also kept costs low.
Accurate isn't enough: five steps to give people a reason to ask
An accurate AI doesn't drive adoption by itself. Slack announcements and guides barely moved the numbers, so Musinsa's analysts led a hands-on, role-specific rollout.
| Step | What they did |
|---|---|
| Adopt 1. Make it visible | A pop-up with the AI guide appears as soon as people open Amplitude |
| Adopt 2. Show, don't tell | Role-specific onboarding sessions and office hours |
| Adopt 3. Remove friction | Bring the AI into Slack and Claude, where people already work |
| Expand 1. Analysts go first | Share good questions and scenarios, run a help channel |
| Expand 2. Widen the data | Add business data on top of behavioral logs |
Each round of hands-on demos pushed adoption up by about 10 points. Cost also favored expansion: loading data into Amplitude incurs query costs, but questions after that add no extra charge.
From a few questions to a product plan in 15 minutes
Working with a merchandiser, the team went from hypothesis to final proposal for a new product in 15 minutes, a task that used to take 2 to 3 days through an analyst.
- Buyer profile: Brand M sold mostly to men 19 and under and women aged 45 to 49.
- Core segment: They focused on male buyers and looked at other categories those buyers purchase.
- Category mix: Tops were already strong, so they dug into shorts.
- Product response: Links in the AI's answers open the live app or web page, alongside visitor and click counts.
- Specs: They identified missing silhouettes, fabrics, and colors.
- Timing: Historical data pointed to the best launch month.
- Report: The AI generated a PDF or wiki page.
The speakers stressed this isn't limited to Amplitude; the same approach works with other dashboard tools.
Takeaways: a checklist for your team
AI adoption rises when you redesign how people talk to data, not when you add another tool. The session closed with a question: is your organization giving people enough reasons to talk to AI?
- Do you track monthly AI adoption as a metric?
- Have you written down the rules that only your analysts know?
- Does your AI check verified dashboards before answering?
- Are you running role-specific onboarding instead of announcements?
daglo's Speaker ID is handy for recordings where knowing who said what matters, like multi-speaker sessions and Q&A. It remembers a voice once you name it and fills in that real name instead of "Speaker 1" in future transcripts, and daglo is the first in Korea to offer this to every user.
For more sessions, see our full MGS 2026 recap, and find event details on the official MGS site.
#MGS2026 #ModernGrowthStack #Musinsa #Amplitude #AIAdoption #ProductAnalytics #DataDriven #SelfServeAnalytics
Where talk becomes work.
Get started with 1,000 credits a month — no credit card required.