This is the first article of a two part series on the impact of Artificial Intelligence on the marketing profession.
Marketing. The American Marketing Association defines this profession as:
the activity, set of institutions, and processes for creating, communicating, delivering and exchanging offerings that have value for customers, clients, partners, and society at large.
Alongside this broad definition, a marketer takes on diverse roles. From developing brand strategies to hands-on tasks such as promotional campaigns, marketing departments seem to be a catch-all for most companies. However, the eventual goal for every marketer is the same: to match the company’s product and/or service to people who need and want it.
Enter Artificial Intelligence (“AI”) and the new era of AI marketing. AI is beginning to transform the marketing industry and, as with any digital disruption, there is a fear that computers will begin to replace marketers overtime.
However I believe that this fear is exaggerated. AI is neither magic nor rocket science. Behind this cloud of hype exists several inherent limitations. Lets cut through this buzzword and look at what AI is doing to the marketing industry today. And perhaps like these computers, we can “learn” their implications for the marketing industry.
All aboard the AI hype train.
“AI is the next big thing for marketing”
I’m sure many of us have heard this phrase in one form or another. Many articles talk about AI but few dare to explore its nuances, let alone define it. This has led to many misunderstanding about AI where some articles even describe new technologies as AI despite no application of AI at all.
Hopefully, understanding these three key terms would make the picture infinitely clearer.
Artificial Intelligence. Almost half a century since the inception of Alan Turing’s test, computers are now smarter than ever. In a field that is rapidly changing, there is no consensus to the exact definition of AI, but many academics roughly agree that AI is,
“[the] activity devoted to making machines intelligent, and intelligence is that quality that enables an entity to function appropriately and with foresight in its environment” (Stanford AI 100)
Simply put, AI is a broad category of technologies which make machines think like humans. Examples of machine intelligence range from crossword solving to optical image recognition to autonomous cars.
Machine Learning is a subset of AI. Breaking it down, Machine Learning (ML) is the field of study that gives computers the ability to learn without being explicitly programmed. Most ML technologies today exhibit “Supervised Learning” where input data (A) is used to generate a response (B). Professor Andrew Ng from Stanford University gives an excellent overview in his article here.
Our computers can do much more when we add in artificial neural networks (more on them below) which draws inspiration from biological neural networks. With this ability to learn through logic and repetition, ML brings the machine one step closer to its human creators.
Data Analytics is not AI. Data analytics does not to make machines intelligent. It is the field which investigates the properties and behaviour of data, governing the flow of data and the means of processing data for optimum accessibility and usability. In other words, data analytics is the processing and understanding of information/data around us.
However, data analytics is a very important piece in the picture as data is what powers machine learning. Through input data, the computer “learns” and comes up with the desired output.
To sum it all up,
But why is this segmentation important? Today where every type of technology is being marketed as AI, understanding what actually makes computers intelligent (i.e. AI) allows us to distinguish real technologies from fads. As mentioned, data analytics is merely business as usual but done by computers. A technology that merely analyses data and provide information is hardly AI. A business that sell the service of data analytics is not AI. And if we determine that the technology uses machine learning, we can begin to recognise the current inherent limitations behind them and their implications for the marketing industry.
What are the main areas where AI is making a headway in the marketing industry? We break it down into three areas:
The examples of AI technologies below are not exhaustive. Feel free to let us know if we have missed out on any existing or upcoming technologies.
AI which provide consumer insights
How great would it be to have an AI handle all marketing executions, testing and decisions for you?
Meet Albert. In a nutshell, Albert “performs many of the time-consuming, manual tasks which humans are unable to perform at the speed and scale required for efficient and effective consumer interactions”. He figures out the best mix of advertising and marketing communications across your channels across an assigned time period and budget. Through machine learning, Albert learns and gathers data on what works and what doesn't. This allows him to adapt the campaign process “automatically on every channel and device, every step of the way”.
Then we have the creative director for McCann Japan, AI-CD β, which sparked the online debate about whether AI can perform a creative director’s job. AI-CD β runs data from deconstructed ads from commercial winners in the past 10 years through its algorithm, pulls out themes which “creates emotionally moving or stimulating visuals” and proposes a creative direction for the current campaign. Using a competition against a well known human creative director as one of its selling point, Al-CD β’s advertisement narrowly lost on a public opinion poll by just 8 percentage points.
Many competitors are beginning to emerge in this space as well. There’s Einstein by Salesforce and Sensei by creative powerhouse Adobe. While they each have their own nuances, all these AI platforms incorporate a degree of machine learning. They address a common pain point by automating menial, tedious tasks and deciding areas where marketers should focus on to get the best ROI.
With these technologies, it seems that we’re one step closer to replacing human creative directors...
Limitations of Machine Learning
Nothing in this article seeks to criticise the technology identified above. In fact, we are all for AI marketing solutions and especially, those that show amazing potential.
But behind every machine learning project lies the following inherent limitations.
Machine learning is highly dependant on its algorithm. Where neural networks are involved, think of the many layers of neurons in a human brain. At the risk of oversimplifying this, we understand that each decision is made on many factors and each factor is assigned its relative importance or “weight”. The machine learns and regularly updates the “weights” assigned to each path to reach the correct decision. When a machine starts learning however, they are usually likened to “black boxes” where you will not know the internal workings but only the input layer and output layer.
An important question is how does one determine the relevant factors which affect a decision? A neural network can be accompanied by many hidden layers. Having too many layers will lead to overfitting where the computer takes into account "irrelevant" factors or "noise" which skews the results. Having too little hidden layers on the other hand will lead to inaccurate results. There is no actual science behind this and is usually context-dependent and trial and error. As my lecturer puts it, “the hidden layers of deep learning is black science” and rightly so.
Garbage in, garbage out. All arguments are unsound if their premises are flawed. Data must first be pre-processed and normalised, else the results produced will be nonsensical and impractical. This means that the data must take the same form and parameters for the computer to be able to make sense out of it. As a result, training the neural network also requires a tremendous amount of human time and guidance which have to be taken into account for any company wishing to rely on it.
Open and dynamic environments. Neural networks today are extremely efficient working in closed environments but not open systems which are constantly changing. This is where deep learning/neural networks come in. Dynamic environments are extremely difficult for neural networks to model especially marketing where user behavior, attitudes and trends are fluid and change on the go.
Absence of human element. Machine learning and AI in general will help to deliver relevant content to the targetted user. However, marketing has always been a subtle blend of art and science. If one is expected to create messages that will motivate human beings, form or change opinions or introduce an element of virality, it will require some imagination, empathy and artistry. And this is something which cannot be provided by a machine. Yet.
Implications for Marketing
Albert, Einstein and AI-CD β target the tedious executions components of marketing. So decisions like where to publish, what content to push and who to push content to will be done by AI. Many of these AI that provide consumer insights do not create content but only provide guidance and direction.
Knowing who to target and what message to use is not enough to convert. From a marketer’s point of view, he has to create campaigns that will reach his target audience and build relationships with potential clients to ensure conversion. Marketers who embrace AI will find themselves freer to focus on the creative direction of campaigns. Leaving menial and time-consuming tasks to machines will give marketers more time to focus on the creative aspects of their job.
Technologies which provide consumer insights open up more areas for marketers to work on. For example, as Albert identifies more targeted groups to reach out to, it is the marketer's job to understand the behaviour of each group and create different visuals or copy write which will appeal to them. This involves a deeper understanding and research of more diverse target audience for your product and/or service.
AI today still have a narrow application and are subject to inherent limitations. For example, AI-CD β is only able to propose a creative direction for a marketing campaign but the everyday content still has to be created by marketers like you and I. For whats its worth, my 2 cents is that AI still has a long way to go to be able to replace marketers in the near future.
In Part II, we explore two other areas where AI has made an impact on the marketing industry. Till then keep learning and adapting.
Photo credits: Raydar