Create your custom ML model

Launched in 2023

 
 
 

Overview of AutoML

All the AutoML products out there are complex and time-consuming to build Hive AutoML simplifies this by enabling anyone to create custom models without coding knowledge, all you need is a set of data ready to upload.

With requests from multiple customers asking for this, we spent 4 months developing this product and launched on June 2023. Ten months have passed, stakeholders want to change the product direction from having the sales team onboard all users to letting users onboard the product themselves.

 
 

Before

After

 
 
 

Role

I lead the product design since 2023, from an idea to now a launched product that’s on version 30. I oversaw the design process from stating the goal of the product to implementation while working with cross functional teams (dev, PM, EM and sales team).

In this portfolio, I will focus on the biggest flow change since the product launch in 2023 - provide user the quickest and clearest way to create an AutoML model.

 
 

Core Team

Vivian Y Lead Product Designer
Mei L Product Designer
Richard L Senior Product Manager
Annie C Assistant Product Manager
Sebastian M Engineer Manager
Pam N Senior Frontend Manager
Soumil U Frontend Engineer
Lucy L Account Executive

 
 
 

Problem

Our goal originally was to push out the product ASAP and to let customers rely on sales team for onboarding. Since 2024, stakeholders decided to let customers onboard themselves.

 
 

Business Problem

  • Retention

  • Growth

Product Problem

  • Unclear Vocabulary Used
    “What’s a Snapshot?”

  • Complex Flow
    “Didnt i just do that?”

  • Relying on sale team
    “Can we get on a call to onboard our new analyst?”

 
 
 

Goal
We need to make the product easy for
any user to onboard

 

Restructured complex flow to simple

 
 

From 8 Steps to 5 Steps

 
 

How AutoML works now

For instance, Reddit wants to make sure the platform is safe for users. They will start by collecting the past interaction history data then

Step 1 - upload data to AutoML (AutoML will start training a model to make sure it identifies sensitive text or images)

Step 2 - pick a model type for customize model development

Step 3 - review model performance and distribute the model from AutoML to Reddit (this can ensure if a user starts sending sensitive text or images, the model will notice and with the support of developers, the sensitive text or image would be blurred out or blocked)

 
 
 
 
 

Impact

From launching in April to June 2024

Business Impact

  • Retention went up by 34%

  • Growth went up by 29%

Product Impact
5 user testings were conducted after launch

  • no questions asked during the create model flow

  • “it’s so easy to create a model!” - Jack W

  • there’s zero customers that asks the sales team questions regarding onboarding or create model flow

  • Time spent: 13 seconds faster

  • Decision Count: 3 counts lower

 

Take Away

  1. The design system we used in the beginning were to match Hive’s design system, but that design system is old. The frontend team had time to create new design system for AutoML. I prioritized on making the product look similar to all Hive products at the time and should’ve prioritized the product growth and goal. In the future, I would make sure to remember this lesson and have a meeting with the PM regarding both options.

  2. The PM and I need to have a goal in mind, we should be one item, meaning we need to have the same vision of the product. Because when I designed the MVP, i remember i asked the PM, are we designing for what’s best for this sprint or whats best for the goal of this product. She replied with “best for this sprint”. Which i listened, but later realized, i shouldn’t have, because since then we had to make multiple changes in the design and frontend. Instead I should’ve told her the pros and cons with each decision before just simply trusting her with the decision.


    I will always remember the two important lessons i’ve learned.

 
 
 
Next

Chatous