How do YouTube employees identify up and coming creative talent on the platform?

Uncovering the Problem

During my career at YouTube, it was extremely common for employees to need to identify specific types of creators for various projects and initiatives. Most YouTube employees were familiar with the platform’s largest creators, but needed help finding creators within specific data metrics, in niche categories, or creators that stood out for qualitative reasons.

YouTube has massive amounts of data on all of the platform’s creators, but the average employee did not have the time or specialized knowledge to access the raw data, create quantitative metrics to analyze the data, manually select creators that met qualitative criteria, and format the findings for distribution. Luckily, discovering new and interesting creators was always a passion for me, and so it became my job at YouTube to fulfill these requests.

COMING UP WITH A SOLUTION

When our team first started receiving requests for creator talent, our initial process was manual and time consuming. First, the person making the request would provide some criteria and context, such as “I’m looking for up and coming comedy creators between 100k- 500k subscribers to invite to a workshop next month in LA”. Using these criteria, we would create a script to do a data pull, using as many quantitative filters as possible (e.g. subscriber count, content category, location, last upload date, total views, watch time growth, etc). We would then go through the data to remove channels that did not fit the qualitative criteria (e.g. channels outside the target content type, managed by brands, or not in the desired location). Next, we would watch content from each of the remaining channels, allowing us to curate a short list of ten to twenty creators. As a final step, we would then distribute our findings back to the requestor in a nicely formatted and shareable spreadsheet.

BUT HOW DOES IT SCALE?

After fulfilling hundreds of requests, we could no longer keep up with demand. People loved the high-touch service we provided, but we needed a solution that could scale and be accessible to more YouTube employees. In addition to the time cost, it was not possible to search through all of the spreadsheets that we had created, the information on them went out of date quickly, and users often created duplicate or similar requests.

I decided that creating a central, searchable database would solve many of the problems with the existing service. With a self-serve database, employees could search for manually curated channels on their own, and then could come to us if they still couldn’t find what they were looking for. To create the database, we established a tag taxonomy divided into three main categories (talent, format, & topic), which we derived from the requests we had previously received. We then added all of the channels we had curated up until that point to a single spreadsheet, and used a data crowdsourcing platform to apply tags to each channel according to the taxonomy we developed. Lastly, we built a front end dashboard that also pulled in current channel data that was refreshed every day, making it so that users could search the database by up to three tags and apply quantitative data filters to narrow their search even further.

Results

We gave the database a name to help people remember where to access it, shared the tool with all of the previous users of our talent services, and provided demonstrations to new teams who we thought could benefit from using the platform. Through this effort, we were able to increase the number of employees using our talent services from several dozen to hundreds of users across all of YouTube’s global offices. At the same time we were able to significantly reduce the number of incoming manual requests, while keeping the quality of our creator recommendations at the same level. This impacted YouTube as a whole by ensuring that employees could more accurately identify the right creators for brand deals, events, product feedback, original programming, or partner support.

Personally, this project marked a transition point where my career started to shift from creative research and analysis into product design and management. I enjoyed the process of designing and building a platform to scale a popular and needed service, and the skills I learned helped me lean more in a new direction.

As an example of my shift toward product development, I was soon asked by leadership to scope out an expansion of the Talent Vault database to encompass purely quantitative search requests, as well as the qualitative searches the platform already supported. This expanded the objective of the project from presenting curated channel research into a user interface to easily access YouTube’s internal data on creators (scope creep!), and re-designing the tool from the ground up.