Blog: Brand Marketers Keep Up. AI Is The New TV!
Not too long ago, in order to build a successful brand and sell a huge amount of goods and services there was a formula:
- build a product people needed.
- Put your dollars into TV advertising and see sales go up.
- Reinvest your profit in TV ads and repeat.
Products on shelves had a label “As seen on TV”!
TV was hyped and cool. More importantly, it was a huge megaphone. Brands had a reach they never experienced before. A direct result of that was that they achieved global scale. Almost all major brands (Coca-Cola, GE, McDonald’s to name a few) which had a breakthrough in the last century have built their success on TV advertising.
But things have changed.
First, The market is saturated. Consumers have most if not all their basic needs met. We don’t buy anymore because we “need” the products, we buy them because of how they make us “feel”.
Second, We are busier than ever. Consumer attention is continuously shifting. We don’t have time for ads (especially TV ads). People don’t have time to think and compare which one of the 89 different shampoos will better wash their hair, they go with the most attractive packaging or the one that smells better.
The conclusion is only marketers who develop the ability to show their advertisements when the consumer is “emotionally receptive” win.
Fortunately, with the shift towards mobile (Mobile web usage overtakes desktop for the first time). We produce an extraordinary amount of information that can be used. New computational systems and AI enable the creation of systems that can make sense of this huge amount of data, produce laser-focused personalizations and deliver in real time to a very large population.
An intelligent system can help improve ads delivery in different ways:
• Making sense of all the noise in this world
The decision of choosing between showing a “grooming product ad” and a “pregnancy test product ad” is easier for a system if the gender of the user is available information. But what if you have a full inventory of ads about cars, movies, health products, financial services…?
Suddenly the problem becomes much harder. You need more information to make a good decision. At the same time with more data, your options grow exponentially. Knowing the city of your consumer, for example, will increase the possible options to choose from: Just adding the 69 cities in the UK as source of data, increases the number of possibilities from few (male or female) to all possible combination between the 69 cities and being male or female (that is 138 options). What if we throw in other sources of information: what type of phone is the customer using? what time of the day is it? What is their age? What is their history of interactions? Millions of options are to be considered for each user.
Machine Learning (one of the biggest branches of AI) called “non-linear” methods can help make sense of interactions. The combination of these methods with the latest cloud computing technologies helps distill the relevant interactions between all the information from the overall noise.
• Complete your data
Usually, it is hard to get data about the consumer. AI can help “infer” this information to all the consumers in our databases. For example, we can learn, when we have the data, what characterizes a male from a female user (they use different applications, they interact differently with the same content…) and predict the gender for new users based on models previously built.
This new “predicted gender” can be used then as an input for the main predictive engine that is making the decision.
• Augment your data
Other techniques of AI can be used to learn “meta-data” that can be helpful for targeting. Natural Language Processing (NLP) enables machines to understand human language. Not only the words but the context too. Like NLP, Image Recognition is another set of techniques that enable machines to understand images. These powerful abilities can be used to extract new features to enhance an AI’s targeting ability.
Input to these algorithms can be written descriptions of your campaigns, photos, and videos of your advertisement, social media posts about your brand or even news reports and articles scrapped from the internet. The output will be a deep understanding of patterns and correlation that can be used to predict user behavior when it comes to interaction with advertisements.
• Uncover the best delivery medium
AI can be used as a top layer that optimizes spending on campaigns running on different mediums. How much should you allocate to Facebook Ads, Google Ads, Youtube Ads, Programmatic…? How much to allocate, within each medium, to the different formats you are running? The answer will probably change depending on the results.
If we take into consideration that results also can change over time (especially for seasoned businesses) we can quickly see the advantage of using machine intelligence and adaptability to keep our campaigns always optimized.
• Filter Fraudulent traffic
Bots, huge networks of computers operated remotely, are mimicking human behavior online and are used to generate ad revenue sometimes mostly by ill-intentioned people.
It is no secret that this practice is a big bottleneck for the industry. Digital ad fraud takes $1 for every $3 spent on digital ads, and online advertisers are estimated to lose $7.2 Billion globally to bots in 2016!
By teaching AI what normal traffic should look like, fraudulent traffic can be flagged. Algorithms can catch unexpected events and filter them on the fly (a user clicking on the same ad more than 80 times at a specific time each day??? fishy…) and can learn to counter fraudulent techniques as they evolve automatically.
• Generate Insights
Beyond the insights about which medium works and to what extent, the use of AI enables the discovery of meaningful insights about customers. It can explain why certain audiences behave in a certain way and reveal new unexpected special customers.
All this information can be used by brands to create better stories, refine their message and personalize customer offers with relevant information.
But some problems have to be solved to completely unleash the power of AI:
• User Privacy
Using AI involves the storage of a huge amount of data about the users. Anonymizing it (targeting on audiences level, not on an individual level), protecting it (cyber attacks against data holders are common) is also primordial business practices.
• The ultimate currency “Attention”
Users don’t want to be interrupted. We all hate ads popping up! Even if the content is highly targeted, creativity is key. Marketers should turn to new technologies to enhance the user experience.
Chatbots, voice interactions are all new technologies that can be embedded in the advertisements themselves.
• Interpretability of results
The majority of the powerful Machine Learning techniques are “black box” models. The number of parameters and features that are used to get high accuracy can be overwhelming for our 3-dimensional brains to understand.
In other words, we might not know how the exact model works inside and out, but we understand the learning algorithm that created it. Even if there is an incredible effort in the academic world to enhance our understanding of such techniques, there is still a long way to go.
• Data quality and availability
The AI is as intelligent as the data you feed it. Quality of data is essential. In the mobile world, publishers that are directly linked to AI providers will give the most reliable data. Quantity of data also matters.
The human brain is a fascinating thing, a child can learn to recognize a tiger just after showing him a single image 3 times. But to train an AI, you need millions of data points to have good results. That is why tech giants are dominating the space. Who is in the best position to build an AI that can perform Face Recognition? Facebook of course. They have more than one billion faces in their databases.
• Which metrics to use?
Nowadays, the metrics that marketers use are CTR (Click Through Rate), VCR (Video Complete Rate) among many others. Using AI to optimize on these metrics is suboptimal at best and a complete waste at worst. How does a 70% video completion rate translate to how aware you made the user about your offer? How does a 3% CTR inform you about how far did you move a user toward a purchase?
AI should be used to optimize towards meaningful metrics: brand awareness, consideration, purchase intent, footfall… These can be measured by asking the user. In “Data Jujitsu”, DJ Patil argues that if done correctly, one can engage the user to give more useful, high-quality data. In the context of the mobile world, surveying a sample of users, before and after showing them an ad, can help quantify these metrics. Using some kind of Uplift Modelling, optimization can be performed. That is exactly what a product like PurchaseLoop by LoopMe tries to achieve.
Finally, AI is undoubtedly a fantastic tool. It is altering the way we live at the very core. The change can be latent but it is of a higher order of magnitude than users perceive. Nevertheless, it is just a tool. It only provides results when the marketing message is to the point and can be a platform for virality when the product is a “Purple Cow” (using Seth Godin words).
Only when the creative communicate the “why” of the brand that the full potential of AI is completely used. It is the ultimate solution to the distribution problem in the digital age. It is the only approach that can get us closer the holy grail: the message will get to the user, at the right time, in the right place when he is in the right state of mind.