Q1 2021
The brand should be like a machine the design team maintains and improves. What our stakeholders request, should be the machine's output. When a request contains a problem that hasn't been solved before, the solution should be packaged in case the problem is encountered again. Every part of the request with an existing solution should be assembled from a library of solutions. And when the team updates an existing solution, all live instances should automatically update too. In my dream job, I wouldn't solve the same problem twice.
In Q1 2021, I was tasked with getting Amazon Fresh's updated brand prepared for Composer, our team's automation tool for generating visual assets. For Google ads alone, there are 20 different ad placements to design for. These ads have different aspect ratios, will be good for different kinds of messaging, will come with their individual size limits, and that's just Google. There's ads for Facebook, Instagram, Snapchat, email, Amazon ads, and regional platforms from other countries that are requested everyday. Managing a template library would mean designing almost a hundred different templates. Automating asset production based on those templates then becomes a hundred mini programs that have to be written. Also, my engineering partner and I had a month left. We could have fairly pushed back, but I had an idea I was curious to try and our engineering partners are always game for a challenge.
What preparing for Composer used to mean, this illustration captures 3 placements. Only 90+ left to go...
The idea for generative templates was switching from designing each template to designing relationships between template components. Clear relationships between headline and sub-headline, spacing, padding, buttons, logos, ad imagery, and brand colors were established so that a given solution might include or exclude any combination in any order.
A unit of measurement was based on the most limiting dimension of the requested template and all values for a given design, like type-size, line-height, border-radius of a curve, blur-radius of a drop-shadow, dimensions of an image etc. were determined in terms of that unit.
These two ideas together enable generating creative assets of any dimension, regardless of aspect ratio, viewing size, or content requests instantly. If we wanted to make a visual update, we'd be updating a set of relationships, not a hundred design templates and programs to automate their production.
This idea took a week to prototype with engineering partners, and within the month, we could generate any Amazon Fresh external ad for any channel, placement, or market internationally. And the machine is still running smoothly. We haven't had to design external ads for Amazon Fresh since, and we're using the same solution to generate templates for merchant partner stores on Amazon and for Amazon grocery in general.
Credits
Engineering partners: Marty Cortez, Saswat Bastia, & Michael Shortt Creative team: Zian Dinh, & me Copywriting: Julie Kline