We wanted to help improve the descriptions that publishers created so that advertisers would be more interested in partnering with them.
- Scope: Concept Design
- Role: Concept, Research & Visual Identity
- Tools: Figma, Figjam & HTML
- Collaborators: UX Writer, Legal & Engineers
Placeholder example for where the generative AI button could sit.
Existing Product Problem
We noticed that many of the publisher’s profiles were lacking in the descriptions section with only a sentence or a few words being shown, and this would lead to advertisers not wanting to partner up with publishers.
Advertisers want to learn about publishers and what they can offer them and the first impression counts when an advertiser views their profile.
Process
Research
[videojs_video url=”https://www.gd86.co.uk/wp-content/uploads/2024/04/ai_research.mov” autoplay=”true” controls=”true” preload=”metadata” loop=”true”]
We learned that publishers found it difficult to write descriptions about themselves, more so when they were first onboarded onto the platform. I researched profile descriptions that stakeholders and advertisers valued and explored Gen AI solutions to understand familiar patterns.
In meetings with an AI systems architect and Engineering Lead, I learned about API capabilities, data limitations, prompt query costs, hallucinations, security concerns, and response times. For familiarity, I researched what icons we could use that represented AI, and what other components are now commonly associated with AI tools.
I then discussed these findings with the Product Owner to plan our approach and determine what to include in developing our AI tool.
Ideation
We planned to use AI to generate descriptions. I focused on understanding the UX for new and existing customers, examining how they would interact with the description generator with or without existing content.
As we have a variety of sign-up processes, I wondered where else our new feature could sit within these and the possibility of including a description generator to help our advertisers as well, to create more value for our customers.
With new AI tools emerging rapidly, we needed to consider interactions for generating results, and deciding between a loader or streaming to keep customers informed.
We also planned to implement limits for shortening and generating descriptions to prevent abuse of the AI solution.
First-Round Concept Review
[videojs_video url=”https://www.gd86.co.uk/wp-content/uploads/2024/04/adescript-load.mov” autoplay=”true” controls=”true” loop=”true”]
I created draft concepts and animated prototypes to show stakeholders and engineers how the new AI generator could look and function within the web app.
The stakeholders didn’t have many concerns, apart from the limitations we had discussed earlier. One thing that came up at this stage was how we would involve legal and to make sure we could inform customers that use the generator about any legal complications and about results not always being correct.
With the web app’s styling changing, I also developed a working HTML and jQuery version to guide engineers on component consistency during implementation.
Before the final release, I worked with the UX writer and stakeholders to make sure the content was right for the customer’s UX.
Outcome
We launched a feature that uses Gen AI to create descriptions from profile information. Customers can choose not to use the AI-generated content, but we advise it to make profiles more attractive to advertisers.
Key Takeaways
With the rapidly evolving AI landscape, many features were still in development and uncertain. We collaborated closely with legal to ensure compliance if AI-generated results were incorrect.
Initial concerns included long response times, with engineers reporting up to 30 seconds for results. I worked with them to reduce this time and inform customers about wait times.
Despite these challenges, we created a robust tool for customers. To ensure consistency and familiarity with popular tools like ChatGPT, Copilot, or Bard, we plan future iterations and improvements.
