Placement Recommender Application

problem

How might we make the Insights & Analytics Portal more innovative while encouraging new opportunities?

OVERVIEW

Placement Recommender was envisioned as an application in the Insights & Analytics Portal for affiliate advertisers running paid placements, offering insights into the estimated incremental sales that would be generated. This would become Rakuten’s first instance of an app using generative AI design to help establish opportunities between publishers and advertisers. Given the users’ budget and other dependencies, the app provides opportunities for optimal paid placements and estimated return on KPI metrics.

The release of this application allows advertisers to make data-driven decisions on which publishers are more likely to drive incremental value through bringing in new customers and additional sales. Publishers also enjoy better paid placement partnership opportunities which overall leads to higher earnings on both ends. 

_

My process started with gathering initial requirements from the product manager. I created a user flow diagram to better understand the complexities of the task at hand.

EARLY WIREFRAMES

To begin my low-fidelity wireframes, I identify the need for a request form and a way to view and compare the generated result recommendations. As part of design research, I looked into various competitor sources of how requested results could be displayed including options like vertically stacked lists, tables and infinite scrolling.

After reviewing with cross functional team members, we ultimately agreed on introducing a new pattern to our design library, to use a carousel scroll in conjunction with our existing card component. This pattern would efficiently display results by allowing quick scannability, KPI comparison, and applying sorting logic on up to 10 cards per single request. It would also allow stacking of previous requests for users to refer back to as needed. I aligned with engineering and product on the specifications of the carousel; given that we all saw the need for future use of such a component, the team allocated resources to have it built into our design system.

ITERATIONS

Moving towards mid-fidelity, I worked with the stakeholders such as Client and Data Science teams to understand and populate designs with details such as form field inputs, publisher information and metrics data provided in cards. Previously, these teams manually analyzed relevant data and metrics to suggest paid placement opportunities for their advertiser clients.

RESEARCH & TESTING

I continued to iterate on mocks by conducting user testing and interviews with the members of the Client team who regularly created paid placements for our large or full-service clients. One such testing performed is depicted above, where I concluded on proceeding with the option to separate results from the request form, in anticipation of additional form fields in the future. In other layout options, results would become hidden and require extensive scrolling as form options grew.

I also continued conversations with engineering team and replaced the loading spinner with a skeleton load due to expected load times of up to several seconds. The intention of the skeleton load was to reduce perceived loading times and retain user attention.

MVP

Our MVP release was finalized with:

• A request view with options for AI-generated vs. user input publishers, budget and other optional parameters

• A historical results view of previous placement recommendations generated

POST-MVP RELEASE IMPROVEMENTS

Upon release, we received feedback from clients regarding the need to view vertical and network information in the results.

I conducted a usability test with the options above. With the overall response from testing and a focus on keeping scannability the utmost priority, I proceeded with using the first option shown which utilized icons with further details on hover. This option keep the result cards a respectable height while accommodating the new information.

As iterations continue, Placement Recommender has future improvements on the horizon including:

• Removing old requests

• Filtering on results

• Further sort options

• Improving mobile views

In 2023, a year since the application’s release to production, data shows Placement Recommender has been used to generate 2,000 requests and facilitate opportunities between advertisers and publishers.