Overview

Imagine combing through 10s, 100s, or even 1000s of profiles to find the right thumbnail designer for your favorite content creator - MrBeast, Jxmyhighroller, or someone else. How many creators can you satisfy at the same time? How efficient is such a process?

This was the case for Roster, a startup that connects creators with video editors, thumbnail designers, and other professionals who already collaborate with their favorite creators.

Disclaimer: Due to my agreement with Roster, I’ve omitted confidential information about the business in my case study.

The problem

When Roster first started, manually connecting creators to skilled professionals was a feasible way to serve users. This approach added a personal touch to every recommendation and contributed to Roster's growth. Eventually, as their talent and creator network grew, Roster developed and launched an online platform to serve their audience.

With the platform online, the sales team no longer had to do the heavy lifting for the creators. However, this shift came with the drawback of losing the personal touch in recommending just the right candidate.

Giving creators the power to find the right person

While we had a conventional filter system to help creators filter through talent, we needed a more personalized filtration method. This new method would not only showcase skilled professionals but also highlight those who would be most beneficial to a creator based on their specific needs.

The goal of the work to be done was:
‍

• Make it easy for creators to find the best talent for their needs
• Create a system where creators make minimal to no adjustments before choosing a prospect.

‍

My role

I joined Roster after they were done with creating the online platform. My first assignment was to lead design for the recommendation and job request experience, but for the sake of this case study, we’ll focus only on the recommendation experience.

I worked alongside the CEO, engineers, and one other designer.

‍

Solving the problem

The manual process for matching creators to talent was a linear one which had been tried and trusted

Before I joined, the team had determined that the best way to address the issue was to create an automated matchmaking algorithm that would mimic the manual process. This algorithm would be score-based, with scores assigned according to the major criteria that determined whether a talent was a good fit for the creator.

Each criterion was assigned a different score, with some criteria weighted more heavily than others based on their priority to the users. For instance, research and conversations with creators revealed that they most of them prioritized a talent's skills over the software they use for work.

I designed some mockups, which we presented to a few users (both talents and creators) for testing. The feedback was positive, but we needed to refine the process to make it more useful. Here are some key insights we gained:
‍

1. Hardware/equipment as a criterion was inconsequential. Users were not interested in whether a talent used a MacBook or an ASUS ZenBook to write scripts. What mattered was they had the work done.
2. The pricing/budget model did not accurately reflect the true value of the talent’s work.

‍

Number one on this list was easy to fix; we simply eliminated it from the process. Number two, on the other hand, was a tougher problem to address. It required us to go back to the drawing board, even re-evaluating and tinkering with the onboarding process for users.

I’d have to omit sharing confidential information due to my non-disclosure agreement.

We had to revise the pricing model during the onboarding process, which eventually affected how talent created their profiles to accurately reflect the value of their work. After developing different models and presenting them to our users, we were able to arrive at a solution that proved effective. Only then did we return to refining the recommendation experience.

Returning to the recommendation experience, we incorporated the feedback from the changes made to the pricing model. We also added the option for creators to search primarily based on job type rather than just skills. The reason for this was that creators hired for a specific role, and each role required certain skills.

Think of it this way: there are skills every chef must have. If a restaurant is hiring, they’re hiring for the role of a chef, but that chef must possess a predefined set of skills. This means they won’t post an open job listing for a skill; they’ll post one hiring a chef with those skills.

This change was a step in the right direction, as it also influenced the pricing model we ultimately adopted.

By the end of our sprint, the criteria had evolved to include job type as well as creative style. We left creative style as an optional criterion, allowing creators to search for talent with specific creative styles, such as dark and humorous or fun and entertaining. Including these additional steps was necessary to better serve our user base.

‍

Design and execution

I began the design process with reusable components such as buttons, text inputs, and, most importantly, the multi-step component previously used for onboarding users on the system.

This decision accelerated design execution and allowed us to move to testing quickly, as the engineers didn't have to build the visuals from scratch—only the algorithm needed to power the visuals.

I created a seven-step form to collect data for all the criteria used in matching creators to talent. After gathering the data, the algorithm runs it against the profiles of each talent on the system and displays the results to the user who initiated the search. I implemented a match rating for each displayed profile to help creators see how closely a particular talent met their needs.

To determine the best way to present the results, I researched how other platforms like Toptal, Dribbble, and Upwork display profile cards. The purpose of this research was to identify the most important information users need to make quick decisions without viewing a full profile.

After studying what other industry leaders in the marketplace were doing, I experimented with different layouts, each with slight variations. Ultimately, I chose the one we moved forward with because it had the right CTAs in the optimal positions, based on my judgment. I shared the other options only with the other designer on the team; everyone else saw only the one that I deemed most preferable.

Displaying the most important information helped ensure creators were not overwhelmed with information but were instead provided with key details that helped them make decisions efficiently. This approach aimed for a "little win" when a creator could hire talent without needing to delve into their full profile.

If the user is satisfied with a recommendation, they can save it for later or immediately send a project request by clicking a button. If they need more information about the talent, they can easily click on the card to view the person's full profile. Unfortunately, it later came to my attention that some users didn't realize they could click on the card to access the full profile. To address this, we added another CTA specifically directing them to view the full profile.

When the user clicks the ‘send project request’ button, a form appears, requesting the relevant details that the talent needs to accept the project request.

Feedback/Impact/Project takeaways

Feedback

After a few months of launching the feature, we received feedback from user interviews and testing. Users indicated that the recommendation card and project request form needed adjustments to better serve creators.

They highlighted that the most important information for them was quickly seeing the job type and previous projects of the talent as opposed to the skills of the talent. They also mentioned that they often don’t have all the information needed to inform the talent at hand. As a result, we reduced the number of questions in the form, making the process quicker and easier to complete.

‍

Impact

The biggest challenge we set out to solve with this feature was to eliminate the need for a Roster representative to help a creator choose the perfect supplier, and we achieved it. This accomplishment profited us in the following areas:
‍

• Increased Efficiency: By automating the matching process, we significantly cut down on the time and effort required to connect creators with suitable talent.‍
• Scalability: With a manual process, there were only so many creators we could assist at a time, but the platform can now handle a larger volume of matches without the need for additional human resources.‍
• User Satisfaction: Creators can quickly and easily find the talent they need, improving their overall experience.‍
• Resource Optimization: We are a small team of professionals. Automating freed up Roster representatives to focus on other tasks, improving overall productivity, and encouraging innovation.‍
• Data-Driven Insights: The automated system allows us to collect and analyze data on user preferences and matching success, enabling continuous improvement of the algorithm.

‍

With recommendations, we didn't quite hit the hammer on the head of the nail initially, as creators often had to tinker around to make additional adjustments to the recommendations we shared with them before finding a candidate they were comfortable with. However, this was still faster than having us assist them manually. We later made improvements on this by interviewing users, observing them use the system, and refining the data we presented to them on the talent recommendation card.

‍

Takeaways

Building for users is a continually iterative process, as their needs and preferences are constantly evolving. This means that developing a product is never truly finished; what I've written today can become an outdated solution tomorrow, turning it into a new problem to solve.

I thoroughly enjoyed working with a talented team. You'll notice I used a lot of "we's," and that's because, despite being a small team, we deliberated extensively together while still handling our individual assignments.

Additionally, having established components allowed us to quickly ship new updates. Lastly, conducting user interviews and observing how users interacted with the product was invaluable in understanding their challenges and designing solutions that truly work for our audience.