Digital Marketing and AI: How is AI being used now?

In this article:

  • The potential for AI
  • What has happened with AI
  • Interesting observations with AI
  • Who are the Big Players?
  • What tools are using AI?

Artificial Intelligence(AI) has arrived!

No longer is AI a “future technology” like nuclear fusion, but a technology that is now being commercially developed and available to you as a digital marketer.

These technologies and tools will save you time, and money.

The potential for AI

Digital has traditionally solved discrete black and white questions – logical problems. If “this”, then do “this”. Classify “this” item in “this” category.

The codified problem solving model has been the foundation for our entire digital age. And now codified problem solving has a growing sibling in AI that is showing just as much potential.

AI offers a way of solving problems that are not easily codified with a set of specific conditions, problems that would have been impossible in the past with digital systems. Image recognition, voice recognition, pattern matching, video analysis.

It is both scary and amazing that a computer could do a complicated task that would normally require a person to perform. Many rudimentary and advanced tasks make up a good part of many people’s day to day job.

Example – Create tags from an image

Say for example you manage an eCommerce store and you have a series of product images that you need to upload, and you want to extract out attributes of the garment when you upload like the sleeve length, type of neck or similarity to another items. You can’t do this using traditional coding practices, you do this yourself and it’s likely going to be a drudge job.

But with AI, you can now do this automatically with a product like Okkular. An AI based tagging and visual search system. Okkular has been trained to recognise a traditional garments display within an image, you then work on training the system to best understand your garment attributes.

The digital of old allowed the automation of simple logic problems. AI will enable the automation of complex human-oriented classification and decision making problems, and in some instances potentially better than a person. For better or worse.

Example – Guess my age

This got quite a lot of PR when it came out. The technology demo Guess My Age from Microsoft that allowed you to upload a photo of yourself to guess your age. It technically didn’t have a marketing campaign behind it other than being a PR piece for Microsoft technology. But it was a good representation of the rich processing ability of AI.

What has happened with AI?

So why is AI now attracting investment and becoming mainstream? The key breakthrough that happened in the early 2000’s when researchers shifted their thinking from an algorithmic approach of intelligence to one that is based on learning through data.

A data-oriented approach to teaching a neural system how to perform a task better reflects the complex nature of human thinking.

The second enabler (or limiter) has been computer processing power. AI based algorithms take a lot of processing power. But the explosion of computing power available on-demand means that the average coder can now tap into supercomputer like processing power as they need.

Interesting observations with AI

We’ve had some time to play around with AI development frameworks(like TensorFlow) to get a sense of how mature they are and the kind of results you can get if you were given a blank canvas to develop an AI solution.

There are some very interesting takeaways from our experimentation.

Ones and zero’s becomes probabilities

When you strip away the interface and interact directly with you code. You find that the answers you get are tethered to a degree of confidence rather than a fixed outcome.  So the algorithm may be 80% confident that a bird is in the image, but a 50% confidence that it is a cat.

This reflects the nature of the data we are working with. In discrete systems the data is clear, but with human and AI problems the data may not be complete. If you are looking at a picture, what if the bird is blurred, or cut off for some reason? What if the bird is looking away in the camera.

AI can return a degree of confidence which literally can be translated simply to “I don’t know”. And very interestingly, as you are training an AI, you can almost see how some of the mistakes are almost “human”.

In complex systems, it becomes impossible to know how results are generated

Digital processing systems are built for audit-ability. But with AI, the system becomes a complex web of nodes that reflect the behaviour of neurons. When you are dealing with a small system of a couple dozen nodes you can try and figure out how things happen.

But very complex systems simulate the behaviour of the brain, which you can’t simply plug into and back track why someone made the decision they did.

We can’t yet ask AI ‘why did you make that decision’. This creates an interesting situation where we create solutions that work, but we don’t know why it works.

Having enough training data is key

The sophistication of your AI is dependant on the quality of data that you feed in. And this usually is a lot of data.

This means that preparing the data for training can be a very time intensive activity. You need to not only provide the source data, but you need to tell the AI that the expected outcomes of that data should be.

If you can get that data together time effectively, the potential for training is almost unlimited. You can train an AI with more data for a problem than a human could consume in a life time. You could run a training program on a computer for a few days that would reflect hundreds of years of human experience.

Solution providers will tend to manage a lot of this themselves, but the more custom the circumstance to your company, the more likely you will be training an AI.

Who are the big players?

The bigger players are investing billions into AI research. Acquisitions for $500m+ are not unheard of.

Amazon & Apple

Everyone has seen and interacted with these two systems: Alexa and Siri both have voice recognition tool that use AI. As the sophistication of the AI improves the utility of “voice search” is going to become more important to marketers.

Google

Google seriously is investing in AI.

  • Google Translate is now using AI.
  • Google Adwords has a ‘Smart Bidding’ system that uses machine learning.
  • Google also relies on it’s Rank Brain system to help refine it’s search results system.

What we believe will happen with AI and the large providers is that there will be AI brains handling data, but the impact on you will be seamless. You may just tick a box to choose a ‘smart’ or ‘auto’ system.

What digital marketing tools are using AI?

Current AI tools are biased heavily towards enterprise levels problems and eCommerce. eCommerce tends to see investment as the improvement it offers is easily justified return on investment.

Okkular – Automatic image processing and tagging for eCommerce images.

  • Time saving
  • Improvement the quality of searchable product attributes
Stye – Take a photo of a piece of inspiration clothing and the app will find a similar match in your store.

  • Increase the conversion of clicks from searches (improved relevancy)
Nosto – Personalisation engine for content

  • Uses AI to make recommendations based on complex set of potential inputs like shopper’s location and their individual tastes and preferences, including price sensitivity, brand affinity, and purchase history.
NS8 – Fraud protection

  • Uses AI to give users a credibility rating from 0 to 1000.

Breathe easy, AI can’t do “people” (or creation)

In spite of all the potential for AI. You will probably notice that AI is fantastic for processing and classifying data.

An AI is not at the point where it can create a marketing campaign, or quite simply any kind of unique creation – which sits at the very core of the marketing professional.

The most potential we see for AI is going to be related to:

  • Personalisation – search, product recommendations, incentives
  • Advertising cost management (through personalisation)
  • Search (voice, image, video content..?)
  • Advanced data crunching and pattern recognition

As a marketer, this is going to help you manage the volume that is the foundation to marketing. Getting your message in front of a lot of people.

Additional resources for digital marketing and artificial intelligence

 

Tell us what you think!
If you would like to talk to us about this article, drop us a line on [email protected].