Complete this sentence: “Data is ________.” Complex? Important? Boring? Whichever word you chose, we’re guessing groundbreaking wasn’t the first thing that came to mind. When we think about data, the first thing that comes to mind is often spreadsheets full of numbers or ones and zeroes flying across a computer screen. While this certainly constitutes one type of data, it only represents the tip of the iceberg of the world of data around us. We are continually absorbing and processing data as part of our everyday lives, and with accelerated advancements in AI and machine learning, we are now able to bring together human and machine capabilities like never before. These technologies allow us to play with data in new and profound ways, which can lead brands and companies to groundbreaking innovations and designs.
When you do something fun, like watch a movie or listen to music, your brain is processing the data of sight and sound. When you bite into an apple, it is processing the data of taste and texture. Until recently, it was all but impossible for computers to handle this type of data in the same way that our brains do. After all, how do you quantify a sight? How do you assign numbers to a sound? Big data now makes this type of information processing possible. Machine learning is helping us understand the data we collect in increasingly sophisticated ways – and many organizations are learning how to best leverage it.
The team at CircleUp, an investment platform that provides capital and resources to emerging consumer brands, is using machine learning and big data to disrupt the world of consumer goods. Their team of data scientists and investors leverage machine-learning algorithms to monitor an average of 92,000 data points per company*, allowing CircleUp to find, evaluate, and invest in early-stage consumer brands. Not only do the data scientists at CircleUp think about the way consumer products look, taste, and feel, but they also use modeling and big data analytics to capture this information at scale. This union between machine and mind is able to fundamentally change the way startups seek capital investment. CircleUp is not interested in the way one product looks or tastes; instead, they seek this information about all products. By drawing on this aggregate data and these machine-supplemented insights, the company can fuel capital and accelerate growth for startup brands.
The team recently built a model in Helio, their machine-learning platform, to analyze colors on packaging labels. As any marketer knows, color psychology plays an instrumental role in consumer purchasing behavior; indeed, a study by Satyendra Singh published in 2006 in Management Decision suggested that the colors used in packaging design can influence a consumer’s perception of a product by as much as 60-90%. Since the average grocery store stocks 50,000 products and can have up to 40 varieties for certain items, catching a customer’s eye quickly can mean the difference between the purchase of one product over another. The color analysis model in Helio was built to determine which packaging colors were more prominent in a particular product category, as well as to compare how colors differ across categories. This information could help marketers and designers re-envision the way they think about packaging design. While much has been written from a marketing standpoint about brand design and color psychology, there hasn’t been a large-scale color analysis of labels for products currently on the market. Though many designers are thinking about how to design a logo, a label, or a brand, marketers and designers have yet to look holistically at what’s already out there to uncover new insights and think about the use of color in new ways. The model is still in its early stages, but its preliminary findings are quite expansive and fun to look at.
How the Model Works
The team built the model to process information like the human brain does. It works by taking images of packaging labels and running them through an image-processing algorithm, which identifies prominent colors and assigns higher weights to colors of critical objects. For example, the algorithm would assign a higher weight to the color of a brand logo than it would to a background, even if the background color occupied a greater surface area of the packaging. The colors analyzed are represented by the color palette below, which features 27 colors:
The algorithm picks up these colors and then assigns a weight to them based on their prominence on the packaging. Take, for example, this Fage pineapple yogurt packaging:
When analyzing Fage’s packaging, the program picked up 13 primary colors. The higher the bar in the accompanying graph, the more noticeable the color is to the average viewer. Because the blue and red of Fage’s logo quickly draw the consumer’s attention, those colors score highly. Contrast this to the packaging for the plain Fage yogurt container below, which has far fewer colors:
Doing This at Scale
The same analysis was done for 5,400 yogurt packages to see what colors jumped out. It turns out that the Fage packaging was a bit of an anomaly in terms of colors. The average yogurt package looks more like this:
The average of colors is pretty telling and straightforward, but looking at all the colors at once is where this gets really interesting. For the visualization below, the CircleUp team ran the images through a k-means clustering algorithm with ten clusters to make it easier to see the most common color schemes.
At first blush, the chart may look like a really cool piece of modern art (it could be!), but there are also a lot of insights to unpack. Each horizontal line in the chart represents a different yogurt package. Without microscopic vision, no one could discern individual labels, but looking at all the labels at once does uncover noticeable patterns. The labels all have a lot of white, while the labels with orange usually have red or pink and those with a lot of grey and black feature fewer colors – not anything surprising to the experienced graphic designer.
Comparing and Contrasting Categories
The next step is to ask how yogurt packaging compares to other categories. Are the average colors found in yogurt packaging unique, or are these trends pretty uniform across other consumer products? The average color distributions for yogurt, coffee, ice cream, and nutrition bar packaging can be found below.
At first glance, there don’t seem to be dramatic differences across the different categories. In order to make these differences easier to analyze, the team subtracted the colors of one category from another to pinpoint net differences with yogurt packaging as the control. Here, more patterns emerged. We found yogurt labels to be quite a bit more colorful than the other categories examined, with nutrition bars coming closest to matching the variety of colors found.
Painting By Numbers
We have access to more kinds of data than ever before. Not only can we quantify what a human sees or tastes, but we can do so in such a way that a human brain alone can’t. This allows for valuable insights to be unlocked.
Does the packaging color data we explored have other implications? Will this color model help predict future brand strength, purchase drivers, or overall business growth? We think it might. It’s been very clear from package redesigns that simple changes to a product’s visual aesthetic can significantly impact sales. In 2017, for example, Lean Cuisine redesigned their packaging to look more modern and sleek, and their sales increased by $58M. Likewise, after its packaging was redesigned to reflect the product’s simplicity, sales for nutritional bar RXBar skyrocketed, leading to its acquisition by Kellogg for a staggering $600M.
Now, imagine if you could capture these data-driven insights by working with machines to augment human judgment. We’re just starting to understand the realm of possibility for combining the capabilities of the machine and mind, and we believe this new way of decision making will inform a great deal of groundbreaking work.
*CircleUp evaluated more than 92,000 data points per company between 2012 and 2015.