Several years ago when I was just starting my digital marketing career, I had read about an experiment conducted by Marissa Mayer of Google. During the course of this experiment, Google tested 41 shades of blue color for links to improve CTR.
Taking a leaf out of the book from Google if you decide to report your creatives for colors or any other objects represented in creatives it will be practically impossible to tag creatives manually for colors or objects, especially when you are running 100s of creatives. But with help of Microsoft Azure Computer Vision and MixedAnalytics. You can automate this whole process without writing a single line of code.
When I started exploring options to automate this type of creative report the questions I had in my mind were :
- What background color has better CTR for male model creatives and which is better for female model creatives?
- Which type of object has the highest CTR Shoes / Electronics / Apparel?
- Which type of object has the lowest CPMs with which background color?
- Which color has the highest CTR for your call-to-action buttons?
After testing out multiple tools and possible ways to achive these results I finally stumbled across the simplest way to make it work. Microsoft Azure Computer Vision API can give you information like colors, objects, gender, faces, etc about any image. But working with Computer Vision API required complex coding, a skill which I haven’t honed in several years, this sent me down the rabbit hole of finding no-code platforms which can help me achieve this result. This search lead me to MixedAnalytics , it’s a tool that lets you query any API you want and transform the Json results into a simple tabular format on Google Spreadsheets.
Let’s take an example
I started by checking out Computer Vision API in action at Computer Vision Demos here.
I would encourage you to try the same. These demos gave me glimpse into what information I can get for the images i process with Computer Vision API. The information included Objects, Colors, Faces and much more for each image but there were 2 challneges in making this scalable.
- I needed server-side coding to query Computer Vision API at scale with my own images
- The response that we get from this API is in Json format which is not exactly easily consumable for a non coder
Having this API at my disposal but not being able to use it scalably was frustrating but finally I found MixedAnalytics. This is a tool which allowed me to query this API without a single line of code and also tabulated the Json results into Google spreadsheet tables in easily consumable formats.
Now I will spare you tecno babble on how actually I did all this but lets take a look at one example of how the data looks like and how a marketer can use it.
Here are a few demo banners which we are going to run through Microsoft Azure API
These are dummy banners and I put some dummy metrics against them for Impressions, Clicks and CTRs which you can get for your creatives at each creative level from Google ads, Facebook ads or any other marketing platform you are using. But the magic starts when I run all these images through Computer Vision API and tabulate results in Google sheets with help of MixedAnalytics.
This sheet shows information returned by Vision API against each image. For example first image shows tags as fruits, food, snacks etc with colors as Red and Blue. While the 3rd image shows Man, Turban, Beard, Moustache. This should be able to give you glimpse into how accurate and precise the image information is. A marketer can do wonders with information this accurate about 100s banners and only task left to us as humans is to make sure you group this each creative level information into actionable insights for our creative teams.
This sheet has dummy metrics for impressions clicks and CTR while real dimensions which have been retrieved using Computer Vision API.
Against each of the assets above Computer Vision API returns multiple tags which are basically objects present in each image, background, foreground, dominant colors present in the image, Primary object type present in the image, etc. And trust me this is just tip of iceberg. If you dig deeper you can find much more about each image including text on the image, landmarks in the image, celebrities in the image, facial expressions of faces in image and much more. You can find more in Computer Vision Documentation here
At a larger level this API helps you with following information about your images
- Tags : These are all visual features of an image (e.g. Person, Food, Phone, etc)
- Objects: This is very similar to tags but more precises
- Brands: This API will detect brands if there are any logos shown in the image
- Categorize image : Categorizes images in defined category taxonomy
- Faces: Face detection allows you to get data for Gender, Age, Smile, Facial hair, Glasses, Emotion etc
- Colors : This detects foreground, background, dominant, etc.
In the demo sheet above I have used Tags, Objects, Faces and Colors to get data on 5 images shown in this blog. Feel free to play around with that data.
Once you have all these details against each creative in a spreadsheet its quite easy to creatie pivots based on whichever dimension or dimensions you find important for your success metric. For example I have given 2 examples in the demo sheet which is grouping the results based on color and gender. But you can chose any other dimension or group of dimensions you want to report. For example Gender and Color Creative report, Objects and Gender creative report, Color and Objects Creative report