Recent data from McKinsey revealed that artificial intelligence (AI) has the potential to create $1.4 to $2.6 trillion of value in marketing and sales across the world’s businesses. What's more, according to a survey of 341 marketers at for-profit U.S. companies by The CMO Survey, the top uses of AI in marketing include content personalization, predictive analytics for customer insights and targeting decisions. And social media advertising is estimated to surpass $100 billion in 2020.
With that much budget going toward social media marketing, brands are hungrier than ever for effective content. The creative process, however, is filled with guesswork, and the movement toward hyper-personalization amplifies the guesswork problem for marketers. Ad platforms like Facebook allow marketers to test their content but don’t explain why their content performs the way it does. This absence of explainability makes it hard for brands to allocate resources efficiently.
Is there a better way? Below are three cutting-edge computer vision techniques, based on my experience investing in computer vision companies, that can provide some answers and potentially transform the marketing landscape.
1. Generative Adversarial Networks
Recent breakthroughs in GANs have empowered startups and brands to start experimenting with the creation of art, music, fashion and marketing content, and just about anything digital, via computers. The quality of output today is still low, but it is only a matter of time before marketers can ask their machines to generate 100 variations of "a golden retriever on a surfboard at a beach with a sunset and a yacht in the background." This could be game-changing in today's increasingly hyper-personalized marketing world.
To be fair, GANs have their share of problems, as the technology is often associated with fake digital content. An example of this is when a real person’s facial expression and speech are replaced with a fake but convincing version, fooling the audience.
However, this is an area of AI where, once it clicks, the resulting output can significantly augment the work of marketers in terms of massive-scale content generation. The whole theory behind GANs is still in its early stages, but its future in marketing seems promising at the very least.
2. Explainable AI
We often hear about AI success stories where new research gave way to state-of-the-art object detection/tracking, realistic human face generation and other solutions in the marketing space. Although these breakthroughs could transform the marketing landscape, a lot must be done to enable effective use of these technologies in the real world. Limitations of research should be explored to understand when models behave unfairly. Models should be made transparent and show their flaws to prevent people from taking their assessments as absolute truth.
Explainability in AI is the ability to explain why a machine learning system produces a result. AI has historically been a black box, which makes it difficult to detect bias and model faults. This can have dire consequences. For example, last year, plus-sized influencers found that Instagram regularly flagged their content as “inappropriate” even though no nudity was present in the images. This was because Instagram’s algorithm simply identified a skin-to-clothing ratio, which tended to be higher on plus-sized bodies.
Today, AI models used in the industry provide little insight into why a certain decision was made. Explainability techniques both in the traditional machine learning (ML) setting, as well as the deep learning setting, could help marketers understand the performance of their creative. Detailed and explainable recommendations could boost key performance indicators (KPIs) by going beyond object detection and correlating performance.
The result of this is that marketers will not only know what content performed well, but also why it performed the way it did, allowing them to make data-driven decisions around their visual content.
3. Scene Graphs
Humans are able to form split-second impressions upon looking at an image and tell how a collection of objects relate to one another with a high level of accuracy. For example, we’d be able to quickly tell if a person was riding a horse or simply standing beside it just by looking at the image for a second. But this isn't true for machines -- yet.
Visual Genome is an ongoing effort pioneered by Stanford researchers to enable computers to understand relationships between the collection of objects in an image or the attributes present. Scene graphs are used to encode the object instances and explain relationships between objects in an image. The cutting-edge research has many potential applications in marketing, such as smarter auto-captioning and contextual image generation. In addition to identifying objects, encoding their attributes and the relationships between objects will allow marketers to gain a more global understanding and receive more insightful recommendations.
This is a relatively new area of applied computer vision, and most of the work surrounding it is still being done by researchers. However, there are already a small number of innovative startups exploring use cases for it as we speak.
Startups Disrupting The Space
Some computer vision technologies marketers are using to boost their campaign effectiveness come from startups like Moveable Ink, which generates creative with millions of unique variations, and GumGum, which serves contextual advertising alongside content that customers are engaging with. Additionally, through startups like Fiddler Labs, explainable AI is making its way into the enterprise.
Viralspace, a Stanford startup I invested in, uses both scene graphs and explainable techniques to help marketers make data-driven decisions around visual content. The insights are based on machine learning models custom to each brand that are supplemented by learnings across industries. Because Viralspace uses comprehensive scene graphs custom to each brand, the models can go far beyond object detection, as they not only recognize what an image contains but also their attributes and the relationships between them.
In the near future, I believe digital marketing decisions will largely be driven by AI. We may start seeing ads created by AI without even realizing it, and they'll be so on point and well targeted that they might just become part of our lives with minimal interruptions.
