Amazon India - Design Automation

Client: Amazon

Year: 2018 - 2019

Banner X / Designer Tool has recorded very low approval rates for automated banners in India. For instance, the approval rates in August 2018 was four-percentage (4%) in India, compared for eleven-percentage (11%) in the EU/NA and sixteen-percentage (16%) in JP. To improve the adoption rate for Banner X – automated creative, D1 had proposed a test between Automated page and Designer created event pages for three categories: Home and Kitchen, Large appliances and Consumables. Automated graphic will bring in –Faster TAT in HVG and production time of HVG graphic will be reduces to 25%. Automation in content grid event and BAU pages reduces manual design dependencies which takes 4 times longer ETA to turn the design around, and manual design support is costlier Vs automated graphics as it requires more designers to support to produce graphics that comes on level-1/2 onward pages that have lesser and lesser traffic.

Challenges and limitations with the available tools and process:

Designer Tool:

  • Lacks region-specific creatives

  • Incompatibility with image composition for categories like OHL/Consumables

  • Lacks region-specific guideline integration (superscript of symbols like % and ₹)

  • Lacks options to add bank offers/trust icons, etc

  • Challenges with integrating vernacular / Hindi version of mobile experiences

 

Banner X:

  • is not a design tool

  • relies on algorithms and rules to arrange content into a banner

  • pre-determined and standardised design output

  • lacks region-specific templates and use-cases

  • lacks image composition features and custom colour selections

 

The existing BannerX engine is automated in such a way that that tool creates graphics according to a series of standardized inputs and passes them through a set of laters that determine layout and presentation. However, lack of region-specific templates and use-cases along with its lack of image composition features and custom colour selection alienates marketers from adopting BannerX and pushes them to work with the designers. In India, there is a reliance on designers and design teams to create more “bling” as the automated banners do not satisfy most users (see Appendix I for example banner).

Automated vs designer created event pages | Jan ART 2019

Test during Jan-ART to check store performance between – (A) Manually designed store experience, Vs (B) Store experience created with automated graphics to determine whether or not automated graphics hamper store performance. The basis on the test results will extend the automated experience on to other ART events, category and sub-category pages.

Experience

We’ve tested two experience for three categories:
C (Designer created) and T (Automated) customer Experience

Metrics

Following are the key Web-lab metrics C (Designer created) and T (Automated):

  • 0.13% (p= 0.59) Conversion

  • +0.08% (p= 0.915) OPS

  • +0.68% (p=0.176) units
     

Launch criteria was that OPS be flat or increasing. Since OPS increased but p-value was inconclusive (Large appliances page affected the entire web-lab negatively due to wrong scheduling)

Key learnings

  • Last-minute inputs from multiple business and category teams delayed production of both experiences and made of mismatch of background colours.

  • Automation tool has limitation of incorporating longer category/product names. Will reach out to the tech. team to explore the options to add three lines of text for future events

  • Single ASIN tiles are preferred over ASIN collage as ASIN collages have to be created using the designer tool and the same has to be uploaded on the media centre which is time-consuming.

  • Brand logos/brand farms had legibility issues as the tool reduced the same beyond minimum size.

Next steps

Since p-value was inconclusive the test extended on Holi event pages and May ART event pages to see if automation is successful.

 

  • Test on Holi event pages both PC and mobile.

  • All May – ART event pages (content grid) will be tested against the designer created experience

Holi Store 19 (Automated Vs manual designed store experience)

D1 ran a test during Holi store to check store tiles performance between – Manually designed tiles Vs automated tile graphics. This test is to determine whether or not automated graphic experience hamper store performance. Basis on the test results we will extend the automated experience on to other ART events, category and sub-category pages. The objective of this experiment is to reduce dependency on design team for graphics which can be automated using Freeway tool. The graphics tested for automation were only tiles which had no logo units.

Metrics

  • Abandonment rate: +7.55% probable increase

  • N2GL: -0.37% inconclusive

  • Units: +0.44%, probable increase

  • Units conditional: +0.39%, probable increase

  • Log OPS: -0.14%, probable decrease

  • Conversion: 0.05% inconclusive

Reomendation

The web lab has not met the launch criteria of a) Negative impact or no change in Abandonment rate and b) No negative impact on any business Metrics (OPS, Conversion).We recommend not launching this experience. However, this experiment needs to be repeated and iterated on other merchandised pages where design element is required for a longer duration.

Features added basis on A.in requirement

​Line Breaks in Automated Banners for all Freeway Marketplaces

The Content Creation (CC) team has launched the ability to add Line Breaks in Automated Banners for all Freeway marketplaces. Previously, this feature was only enabled for China and Japan.

Banner X 3 lines headline

Basis on A.in event requirements, on March Content creation team launched the feature to define headlines or subtext on 3 lines which enabled the SM’s / Marketers to improve the usage of automation tool. Earlier the usage was limited due to only usage of two lines text.

Accents / Bling elements

Accents can be illustrations, textures, or any other subtle background elements that appear within banners. In BannerX, accents are meant to be subtle elements that can appear behind text without risking legibility. The contrast between the background color and accent should be minimal as the ASIN or other imagery features in the banner should still be the primary focus.

© 2020 by PolkaDot Creative