Creating Image Ads Effortlessly
Launching image ads has been challenging for marketers due to limited resources and creative expertise. My focus is to address this issue by pioneering new capabilities and leveraging smart solutions.
Role
Sole UX lead
Areas
Strategy
Design
Research
Year
Sep 2022 - Dec 2023
What’s launched
Smart image recommendation: Leverage intelligent capabilities, we automatically recommend relevant and high-performing assets for advertisers to create image ads within 1 click.
Smart image recommendation, launched Nov. 2023
Smart cropper: With the 1-click smart cropping solution, advertisers are able to get as many aspect ratios as they need after identifying the focal point.
1-click smart cropping tool, launched Mar. 2023
Multi-image capability: To further provide advertisers with more creative control, we launched the first ever multi-image capability within Amazon ads. Now advertisers can use up to 3 images to optimize their ads for the 26k+ placements.
Multi-image creative, launched Oct. 2023
Smart Cropper was further integrated to multiple ad programs including Sponsored Brands, Sponsored Display, and Creative Assets. After launching the new capabilities on advertisers’ image experience, we’ve observed:
Image ads adoption
+11%
Image ad impressions
+7%
Smart Cropper adoption
54%
Additional assets
11k
Background
Sponsored Display (SD) is an advertising solution at Amazon Ads that enables advertisers to reach relevant audiences browsing both on and off Amazon with 26k+ ad placements.
Sponsored Display ads exmaple
Sponsored Display primarily served product-only ads, which showed suboptimal performance. User research and performance data from Sponsored Brands revealed that image ads achieved a 50% higher CTR compared to product-only ads. Based on these insights, we hypothesized that increasing image ad adoption would significantly boost SD ad performance.
Product only ad (left); image ad (right)
Problem and strategy
Challenges for marketers​
​
The adoption rate for SD image ads has remained low (around 11%). Through quantitative and qualitative research, I identified several pain points:
-
The vast number of SD ad sizes makes it difficult to ensure that all image ads appear visually appealing.
-
Usability issues in the image uploading process have hindered scalability.
-
Many advertisers lack image assets due to resource limitations.
Sponsored Display image ads example
Quotes from Sponsored Display advertiser interviews 2022, 2023
Long-term vision​
​
In addition to addressing pain points, we aim to support our long-term vision of leveraging Dynamic Creative Optimization (DCO) to automatically test and optimize asset variants, delivering the best-performing ads to different audiences. Expanding our ability to collect additional assets is essential to realizing this vision.
Strategy​
​
To solve the pain points and support our vision, as sole UX lead, I partnered with Product, Engineering and Science team on establishing this 3-phase strategy:
Phase 1 - Simplify the complex, build a scalable foundation
Phase 2 - Introduce more creative control
Phase 3 - Leverage intelligent solutions for automation
Design thinking workshop led by me, 2022
Phase 1 - Simplify the complex
The existing image upload flow was restrictive and required users to manually crop images twice, creating usability challenges. To scale and unlock image potential, we introduced a smart cropping solution. With one click, advertisers automatically receive all required aspect ratios after identifying the image focal point.
Old - Sponsored Display image uploading and cropping experience
New - 1-click smart cropping tool, , launched Mar. 2023
Phase 2 - More creative control
To better support thousands of ad sizes, we expanded capability by allowing users to use multiple images instead of being limited to just one. ​However, multi-image capability is new, so understanding the real user mindset and navigating through ambiguity is key.
Ideation
Narrowing down the direction
Understanding the mental model through user testing
​
However, the preferred direction had flaws, as it disrupted existing patterns and lacked clarity on the true user mental model. Working within a tight timeline, I piloted a designer-led user testing to gather insights.
User testing prepared and led by me, Oct. 2022
Quotes from the user testing, Oct. 2022
Based on user testing results, I focused on strengthening the connection between control and preview without disrupting established behaviors.
Multi-image creative, launched Oct. 2023
Phase 3 - Automation
Despite simplified workflows and expanded capabilities, marketers still faced resource constraints in obtaining images—a longstanding challenge. To address this, we launched AI Image Generator, enabling marketers to create unlimited engaging images for free, boosting image adoption from 11% to 35%. To streamline the process further, we also introduced 1-click smart recommendations, suggesting relevant images from pre-generated, pre-saved, and detail page sources for easier use.​
AI Image Generator, launched Nov. 2023
Smart image recommendation, launched Nov. 2023
What’s next
Establishing framework​
​
As the first multi-image feature introduced in Amazon Ads, I established a framework to streamline adoption across other products and ensure consistent design patterns.
Multi-media framework
Multi-text framework
Learnings
​
While we saw some improvement in image ad adoption, we failed to boost overall ad performance due to a flawed hypothesis. Sponsored Brands image ads performed well, but Sponsored Display image ads struggled. With many supported ad sizes, the dynamic layouts for Sponsored Display weren’t effective. Although image adoption initially rose from 11% to 30% after launching Smart Image Suggestion and Image Generator, it declined again due to poor ad performance. Redesigning Sponsored Display ad templates is now a top priority.
​
Despite this, the project advanced our long-term vision of using Dynamic Creative Optimization (DCO) to test and optimize asset variants. The additional assets gathered (11k+) will enhance models for asset understanding, dynamic selection, and contextual optimization.