Case Study Ninety.io 2023 – 2024

Knowledge Portal: 0→1 Learning Platform with AI

How I designed and launched a centralized knowledge-sharing platform from scratch — through alpha, beta, and GA — exceeding every target KPI in 90 days, then followed up with an AI companion that resolved first-session friction and drove sustained engagement.

40K+
Monthly page views within 90 days of launch
40%
Account-level adoption — nearly half of all Ninety accounts
1,000+
Collections created, averaging 300+ new per month
10%
Daily active user penetration from launch
Knowledge Portal UI showing collections list, topic progress indicators, and the Maz AI companion integration.
The Knowledge Portal — centralized collections, progress tracking, and an AI companion to guide users through first-session creation.

Overview

No central place to share what your organization knows

Ninety.io lacked a centralized solution for teams to share internal knowledge, onboarding content, and coaching resources. Before the Knowledge Portal, teams used fragmented tools — Slack, Google Docs, email — to share critical information. There was no way to ensure people had access to the right content, track engagement or completion, or curate reusable knowledge artifacts across teams.

I led the design and execution of the Knowledge Sharing Tool (KST) — an extensible learning platform built to help organizations document, assign, and track learning materials — from scratch through alpha, beta, and GA, continuously evolving it based on user feedback and behavioral data.

The Problem

Critical knowledge was scattered and untrackable

The absence of a centralized tool created three compounding problems for Ninety organizations:

Inconsistent onboarding — Managers struggled to onboard team members consistently. Without a shared system, critical knowledge lived in individual people’s heads or scattered across docs no one could find.
No engagement visibility — Coaches wanted to assign learning materials to clients post-session but had no way to track whether content was viewed, completed, or acted on.
No reusable artifacts — Users needed to filter and organize content relevant to their role. Without structure, every org reinvented the same knowledge from scratch.

“We needed a centralized tool that could organize content clearly across roles and teams, track completion and assignment, and scale with growing organizations.”

Research & Discovery

Interviews, journey maps, and flow mapping

I ran interviews with internal teams, coaches, and clients to understand knowledge-sharing pain points. Key insights validated the problem scope: managers struggled to onboard consistently, coaches wanted a way to assign post-session content, and users needed role-based filtering to find what was relevant to them.

I created user journey maps for both Admin and Learner archetypes — visualizing their motivations, friction points, and success paths. These flows clarified product requirements and gave the cross-functional team (PM, two engineers, and a second designer) a shared lens for evaluating UX tradeoffs.

I also mapped the full content creation and assignment journey — from Knowledge Share to Rock and To-Do creation — which clarified dependencies across tools and uncovered opportunities to streamline the user path and reduce redundant steps.

User testing rounds

We ran multiple rounds of moderated usability tests focused on three areas where early designs had the most ambiguity:

CTA clarity — Single dropdown vs. contextual actions. Users consistently missed the single dropdown; contextual actions performed significantly better.
Drag-and-drop — Reordering content within collections. Several participants expected keyboard-based reordering as a fallback; we added it.
Information hierarchy — Whether users understood the Folder › Subject › Topic taxonomy. Early naming confused admin and learner modes; renaming to “Collection” resolved most of the ambiguity.

Key Design Decisions

Three decisions that defined the product

Dual modes: Admin vs. Learner — Creators needed robust editing tools; learners needed a clean, focused reading experience. Giving each role a distinct interface rather than trying to serve both in one view eliminated the most common points of confusion in early testing.
Progress indicators at topic and subtopic level — Visual completion cues at multiple levels of hierarchy shifted usage patterns immediately. Managers could see accountability at a glance; learners had a clear sense of where they were. Simply adding % completion changed how people engaged with the content.
Scoped collaboration model — Only users with creator roles could edit shared Collections, preventing accidental changes. This felt restrictive in early feedback but turned out to be the right call — open editing created messy hierarchies in beta; scoped roles protected content quality at scale.

The sequencing decision that shaped the MVP: we postponed “content analytics” and “draft modes” until we validated real-world use. Scoping tightly meant we shipped faster and learned from actual behavior rather than designing for hypothetical needs.

Outcome

Every target KPI exceeded in 90 days

40K+
Monthly page views — strong, consistent user engagement from launch
10%
Daily active user penetration — a clear sign of tool discovery and habitual use
40%
Account-level adoption — nearly half of all Ninety accounts with at least one active KP user monthly
1,000+
Collections created, averaging 300+ new collections per month

AI Iteration

The barrier wasn’t the tool — it was the blank page

After launch, adoption data and follow-up user interviews revealed a specific friction point: users understood why the Knowledge Portal was useful, but many stalled at the very first step — creating a new Collection. The problem wasn’t discoverability or navigation. It was the blank page.

I led the design of a Maz AI companion that guides users step-by-step through creating their first Collection. Maz uses a conversational flow to break the process into approachable questions, helping users go from a blank page to a fully structured Collection in under 10 minutes.

By directly targeting the first-session barrier, the AI companion improved first-session success rates and increased overall engagement with the Knowledge Portal. This follow-up iteration demonstrated a pattern I’ve carried into subsequent AI work: the best place to introduce AI is at the exact moment a user is most likely to give up.

This is the same human-in-the-loop insight that shaped my work on Maz for goal-setting (Draft Rocks) and on the AI Assistant at Constant Contact: AI earns trust not by doing everything, but by showing up at the right moment with the right amount of help. The Knowledge Portal intervention was where I first validated that framing in production.

Reflection

What this taught me about 0→1 product design

Balance flexibility and structure

Giving admins too many open-ended content tools led to messy hierarchies in beta. Preset templates and scoped roles helped without removing creative control. The right amount of structure is whatever prevents the most common mistake without feeling restrictive to the users who don’t make it.

Progress tracking drives behavior

Simply adding % completion shifted usage patterns measurably. Visibility of progress isn’t just a UI nicety — it changes how people engage with content and how managers maintain accountability.

Scope MVP tightly, then iterate on what breaks

Postponing content analytics and draft modes until we had real-world validation was the right call. The AI companion iteration wouldn’t have been possible without the behavioral data we collected in those first 90 days — we needed to see exactly where users stalled before we could design the right intervention.

← Back to portfolio Discuss this work →