Case Study

Case Study

Case Study

Automated Content Moderation System

Automated Content Moderation System

Automated Content Moderation System

Custom Ink is an online retailer that provides customized apparel and accessories for groups, businesses, and events. They offer a platform where customers can design their own products, using their vast library of graphics and fonts, or you can upload your own.

Even with a relatively accurate review algorithm, with hundreds of thousands of varied content coming through their platform, there was still a chance vulgar or stolen content could make it past automation.

Manual intervention remained necessary, highlighting the need for a new, more effective tool to streamline the process and enhance profitability.

Custom Ink is an online retailer that provides customized apparel and accessories for groups, businesses, and events. They offer a platform where customers can design their own products, using their vast library of graphics and fonts, or you can upload your own.

Even with a relatively accurate review algorithm, with hundreds of thousands of varied content coming through their platform, there was still a chance vulgar or stolen content could make it past automation.

Manual intervention remained necessary, highlighting the need for a new, more effective tool to streamline the process and enhance profitability.

Maximizing Review Speed

Challenge

Challenge

In order to utilize the content review tool for more order types and eventually automate the entire review workflow, the tool needed a complete overhaul. After shadowing users to identify their pain points and tedious tasks impeding their workflow, I needed to redesign the tool to achieve quick review.

In order to utilize the content review tool for more order types and eventually automate the entire review workflow, the tool needed a complete overhaul. After shadowing users to identify their pain points and tedious tasks impeding their workflow, I needed to redesign the tool to achieve quick review.

In order to utilize the content review tool for more order types and eventually automate the entire review workflow, the tool needed a complete overhaul. After shadowing users to identify their pain points and tedious tasks impeding their workflow, I needed to redesign the tool to achieve quick review.

User and Business Goals

  • Any team member should be able to use this tool without much training (in just a day)

  • Team members should be staying within the tool as much as possible without relying on other tools such as the ones for our existing sales teams.

  • Team members should be able to make a decision on order content as quickly and easily as possible by showing all relevant information up front.

  • Save 5.5 FTEs ($400k savings) after streamlining the workflow.

Maximizing Review Speed

Meet the Team

Meet the Team

Lead UX Designer (Me!)

1 Product Manager

1 Project Manager

3 Software Engineers

Lead UX Designer (Me!)

1 Product Manager

1 Project Manager

3 Software Engineers

Before 👎

Before 👎

  • In the prior tool, team members weren't able to make quick and easy decisions without proper context clues and basic task functionality.

After 👍

  • 155+ orders an hour (goal was 65)

  • 99.9% accuracy in content review

  • Successfully saved 5.5 Full Time Employees (FTE)

  • Saved the company $400k in the first 6 months

After 👍

  • 155+ orders an hour (goal was 65)

  • 99.9% accuracy in content review

  • Successfully saved 5.5 Full Time Employees (FTE)

  • Saved the company $400k in the first 6 months

What Slowed Users Down?

What Slowed Users Down?

The prior skeleton tool illuminated some clear problem areas. After conducting remote interviews and shadowing sessions with five operations team members, I gleaned a few issues that the previous tool lacked:

The prior skeleton tool illuminated some clear problem areas. After conducting remote interviews and shadowing sessions with five operations team members, I gleaned a few issues that the previous tool lacked:

1) Zoom Functionality

Reps were manually zooming into the screen or squinting. When selecting Google Lens image search, upload showed up large and zoomed in which is helpful but this should happen in the tool itself.

2) Ability to Leave Concise Notes

For next rep that gets the order if an upload/text is denied, investigative notes will ensure the next rep isn’t doing double the work.

3) Updated Training Expectations

Taking way too much time to look into orders for issues, need to update expectations on these “deep dives”. Reps are also checking different legacy tools to confirm previous/linked/sister orders (they should not need to be previously trained in operations roles).

4) Relevant Button Text - Uploads

The generic button text not being the file name (Image #1) is slowing down the content review process. Inkers typically find context clues in file name if it’s stolen/screenshotted. Also unable to differentiate between clip art and uploads.

5) Queue Visibility

Not being able to see how many orders are in the queue slows team members down since they have to check in with manager on whether it’s a glitch vs. actual no claim (clicking “Claim Next” repeatedly).

6) Ability to See Both Proofs

Without clicking back and forth arrows would save clicks and make review quicker.

First Attempt

For the first round of designs, I focused on automating certain tasks, reducing clicks, and showing all relevant information at quick glance. The visual below with annotations was used in weekly meetings with my PM team and engineers, then a UX team critique.

Changes to focus on after feedback: Adding clarity to the "Queue", having a "Claim Next Order" button instead of an automatic toggle, and call-to-action (CTA) buttons all have equal importance and use, but this design added confusing hierarchy.

For the first round of designs, I focused on automating certain tasks, reducing clicks, and showing all relevant information at quick glance. The visual below with annotations was used in weekly meetings with my PM team and engineers, then a UX team critique.

Changes to focus on after feedback: Adding clarity to the "Queue", having a "Claim Next Order" button instead of an automatic toggle, and call-to-action (CTA) buttons all have equal importance and use, but this design added confusing hierarchy.

First Attempt

Critiques, Tests, and Iterations

Critiques, Tests,
Iterations

After going through two rounds of UX team critiques and more user testing, I updated the claiming button to be a toggle for auto-claiming, and updated the layout to be a narrower view to accommodate the investigative process (reverse Google searches in a separate window). The button styles were also updated to create equal importance to all three actions.

After going through two rounds of UX team critiques and more user testing, I updated the claiming button to be a toggle for auto-claiming, and updated the layout to be a narrower view to accommodate the investigative process (reverse Google searches in a separate window). The button styles were also updated to create equal importance to all three actions.

Winner Winner

Winner Winner

After multiple rounds of designs, testing, critiques, more testing, and more designs, the final design (for now!) was ready. What did the new tool solve?

After multiple rounds of designs, testing, critiques, more testing, and more designs, the final design (for now!) was ready. What did the new tool solve?

1) Clear Customer Information

Users could quickly see the customer name, email, and address to quickly investigate whether the images or text were stolen.

2) One Click Address Search

No more copying and pasting, the address link would pull up a Google search to see if it's a business (good context clue) or random residence.

3) Clearly Outlined Text and Uploads

No more guessing if something in the preview image is customer inputted text/upload vs. our design lab clipart.

4) Clearly Labeled File Names

Uploads now showed the actual file name to help the user figure out if the image was stolen (ie. "Screenshotted" or "DisneyStolen.png"

5) Claiming Toggle

A claiming toggle was created to speed up the order process even more. Once the order was approved, denied, or canceled, a new order was automatically claimed.

6) Proof View

Being able to see both proofs (front design and back) saved users for additional clicks. The arrows only appeared if there were right and left sleeves.

7) Zoom Functionality (See Below)

Being able to hover zoom over the proof, and *bonus*, engineers added the ability to scroll zoom even further into the image for tiny items!

8) Notes Requirement (See Below)

Previously, if an order was denied or canceled, the sales rep talking to the customer had no idea why it wasn't approved. Having notes required saves the next rep from doing double the investigative work.

Ooooo, extras?!

Ooooo, extras?!

In order to keep the project within scope, I designed with the MVP (minimum viable product) in mind. Through weekly calls with engineers and the collaborative relationship we'd built over time, I learned they were happy to implement even more features, which is ALWAYS a win!

In addition to the hover zoom functionality seen below, users could use their mouse scroll to zoom in even further to completely remove the need for Google Lens to see the smallest details.

In order to keep the project within scope, I designed with the MVP (minimum viable product) in mind. Through weekly calls with engineers and the collaborative relationship we'd built over time, I learned they were happy to implement even more features, which is ALWAYS a win!

In addition to the hover zoom functionality seen below, users could use their mouse scroll to zoom in even further to completely remove the need for Google Lens to see the smallest details.

Hover Zoom Functionality

Users can now hover over the image previews and see an enlarged view of the design details.

Users can now hover over the image previews and see an enlarged view of the design details.

Required Notes

The context provided in these required notes saves the next rep from doing duplicative work.

The context provided in these required notes saves the next rep from doing duplicative work.

More Interviews, Shadowing, and Iterating

More Interviews, Shadowing, and Iterating

More Interviews, Shadowing, and Iterating

After the new tool was deployed, I continued to follow up with team members through remote shadowing and post interviews to ensure everything was working as intended, and plan for future iterations of this tool if needed.

After the new tool was deployed, I continued to follow up with team members through remote shadowing and post interviews to ensure everything was working as intended, and plan for future iterations of this tool if needed.

Reflection

Reflection

After deploying the new tool, feedback was overwhelmingly positive. The most notable sentiments from users were "You really listened to us" and "we're flying through orders!". The workflow was reinvigorated, with team members blowing numbers out of the water:

🔥 155+ orders an hour (goal was 65)

✅ 99.9% accuracy in content review

🙌 Successfully saved 5.5 Full Time Employees (FTE)

💰Saved the company $400k in the first 6 months

In the future the tool will continue using user metrics and inputs to increase machine learning accuracy and expand most orders out to automation.

This was a quick turnaround project, but if I had more time I'd create dark mode to ease eyes after long work days, and explore a widened layout once reverse Google searches aren't necessary.

After deploying the new tool, feedback was overwhelmingly positive. The most notable sentiments from users were "You really listened to us" and "we're flying through orders!". The workflow was reinvigorated, with team members blowing numbers out of the water:

🔥 155+ orders an hour (goal was 65)

✅ 99.9% accuracy in content review

🙌 Successfully saved 5.5 Full Time Employees (FTE)

💰Saved the company $400k in the first 6 months

In the future the tool will continue using user metrics and inputs to increase machine learning accuracy and expand most orders out to automation.

This was a quick turnaround project, but if I had more time I'd create dark mode to ease eyes after long work days, and explore a widened layout once reverse Google searches aren't necessary.

Back to Top

© Jimin Ngo 2025