The Electronic Retailing Association

MCMS 2018 Speaker Insight: Dimitris Karamitsos discusses AI & Machine Learning for a new generation of marketing & content generation

MCMS 2018: Dimitris Karamitsos (Spott), Tom Bowers (Observatory) and Charles Dawes (TiVo), discuss emerging technologies with host, Ben Keen.

Buzzwords come and go, and some of the most popular ones at the moment are AI and Machine Learning (ML). They’re going to take our jobs, they’re going to disrupt the economy, they’re going to make our lives better...but what are they?

What are AI and Machine Learning?

AI, or Artificial Intelligence, is all ‘intelligence’ by machines. There’s general AI, where computers mimic human intelligence, and narrow AI, which is focused on specific tasks. A narrow AI might be really good at driving, but nothing else.

ML on the other hand is a way of creating AI. You can manually create an intelligent system, but it’s a massive amount of work. With ML, the system teaches itself.

Dimitris Karamitsos – Head of International Sales & Business Development, Spott

Dimitris' growth into innovation started in robotic laboratories, progressed to power plants technologies, followed by the co-founding of a digital startup and in 2017 he joined the world of artificial intelligence & machine learning technologies with SpottTM.

Dimitris has a Masters in Mechanical Engineering from Imperial College London, and a Masters in Management, Economics and Entrepreneurship from the University of ETH Zurich. Possessing a passion for bringing innovative solutions to the market, after setting-up his first company, Dimitris decided to dedicate his career to becoming an entrepreneur and impacting society positively. In 2017, he met the two co-founders of SpottTM (Jonas De Cooman and Michel De Wachter) with whom he felt he could share his passion. SpottTM is today the reference in making content interactive and shoppable using AI and ML technologies. Within SpottTM, Dimitris is responsible for international business development and sales.

email: Dimitris.karamitsos@spott.tv

mobile: +41 762665462

website: https://business.spott.tv/

You could program an AI to recognize t-shirts, but this would require an incredibly detailed description of every single possible t-shirt. With ML, you show the system thousands of pictures of t-shirts. Based on this, the AI generates its own definition of the concept ‘t-shirt’. Over time, as the system is shown more examples and is corrected when it makes mistakes, it gets better and better at recognizing t-shirts. Or cats, Johnny Depp, fraudulent payments, ...

How can this be done today?

The two requirements for ML to work are powerful computing resources and massive amounts of data to process. Because both of these are available today, AI and ML have made massive advances in recent years.

AI is still far away from real intelligence, and won’t be able to mimic human thought for a long time. But in many ways, it is already superhuman. It can process insane amounts of data in a short amount of time, finding information and generating real value. Below are some of the areas where AI is already having an impact.

AI & ML impact today

The amount of text, images, video and audio that is created every single minute is mind-boggling - 300 hours of video are uploaded on YouTube alone! And even if you only generate a fraction of this content, it can still be a challenge to keep control of it all.

This is where automatic generation of metadata can be a big help for your content management. Who is in the video or picture, what are they wearing, what are they doing, what are they saying, ... manually tracking all of this in an accurate way is not an efficient or scalable solution.

When we process a scene from New Girl, algorithms assist us with a number of steps that would traditionally have been time-consuming:

  • Selection of high-quality frames: within the thousands of frames that make up even a short video many will be blurry or will not feature relevant content. By selecting only the highest-quality frames, the workload to process them is reduced

  • Recognition of faces: while it’s easy enough to teach someone what Brad Pitt looks like, recognizing tens of thousands of faces is a lot harder. AI can quickly compare the faces in the scene with a vast database of potential matches, identifying the actor and actress and the role they play

  • Identifying products: if recognizing celebrities is a hard task for humans, efficiently identifying which shirt and dress they are wearing is almost impossible. There are simply so many possible options out there that manually searching for the right one takes a massive amount of time

The end result of this process is an interactive piece of content:

Era_ai__machine_learning_for_a_new_generation_of_marketing_content_generation
Automatic generation of metadata can be a big help for your content management. Who is in the video or picture, what are they wearing, what are they doing, what are they saying.

This data enables new methods of performance tracking. By combining information about content, marketing and communications efforts and sales results, new connections can be found.

With millions or billions of data points, finding meaningful connections can be like finding a needle in a haystack. But what might take a human decades to process can be scanned in seconds by a computer. By combining different sources of data, the AI can find over- and underperformance, so you’ll know whether the red or blue shirts are performing best and should be featured more. Or maybe videos with Jack Smith perform better on Tuesday, while John Doe has better results during the weekend?

This new data and the new ways in which we can process it also enables unprecedented personalization & prediction. While segmentation used to be done in broad strokes (you’re an 18 year old woman from Belgium, so you probably like X and Y), we can now predict future interests based on past behavior.

This is the Netflix and Spotify model of segmentation to an audience of one. Based on what I’ve watched and listened to in the past, it can surface new content that I am likely to be interested in, not just because of who I am and where I live, but because of what I do.

The same segmentation methods can also be applied to different communication channels, so instead of bombarding everyone with the same email or notification, people receive the message that is most likely to have an impact.

Combining content & data also allows us to enrich this content. Because we know not only which picture or video our audience is watching but have a full view of what is featured in the video, we can offer our audience a more engaging experience. Really like the dress your favorite actress is wearing? We’ll show you how much it costs and where you can buy it. This improves the user’s experience, turning passive engagement into active interaction.

At the same time the extra data that is gathered generates more value. Because now we not only know which videos you like to watch, but which characters you like most, or which brands you are most interested in. Building on this data we can create content that better fits our audience’s interests, while also forming partnerships with the brands they care about most. A real win-win-win!

Caution is needed however when working with these new technologies. As humans we are programmed from birth to spot results that are ‘off’, slightly different from what we were expecting. AI recommendations and results can quickly fall into this uncanny valley. We have all been stalked by online advertising that assumes we have certain interests because of a one-off search online, while we don’t actually really care all that much.

And while telling cats and dogs apart is a trivial challenge for humans, even the smartest AI algorithm can be fooled with some tricky camerawork.

This makes the quantity and quality of data that is collected even more important. There is a need for datasets that are both massive and accurate to ‘train’ the AI using ML, with more and higher quality data leading to improved future performance by the AI algorithms.

Some of the many ways in which these new technologies can benefit you today include:

  • Improved content management: reduce the workload of gathering meta-data about the content you produce by employing algorithms that automatically scan and categorize it. By enriching your content with data you can create value in new ways, like the two below

  • Reuse content across channels: you’ve created a great video for TV - now ‘recycle’ it by using the same content across your other channels, like your website or social media. By using the content’s meta-data you can convert it in an efficient and scalable way. You can even add a shoppable layer to your content, allowing your audience to quickly move from inspiration to activation

  • Cross-channel performance analytics: you probably know which days or which products are your best performers, but by combining content meta-data with performance data and unleashing AI algorithms, you can get insights on a granular level. Determine exactly where you are under- or overperforming so you can make targeted changes that will have the biggest ROI

The advent of AI and ML is a modern-day gold rush. For too long we have been building up massive piles of data, with buried treasure of information that went untapped because we were not able to efficiently mine it. New tools are finally unlocking this information, and the race is on to extract the gold.

About Spott

Spott is the reference for interactive and shoppable content; from well-known TV-shows to vloggers or other branded content. Spott was selected as one of the top 100 most disruptive technologies in the world by Disrupt 100. Spott enriches the viewing experience on every screen (smart TV, iOS Android, website, video player or set-top boxes) with a real-time view on the ROI of every product & brand shown in the content. Increase the value of your content thanks to Spott's interactive plug-in. Spott has offices in Brussels, Lisbon, Rio De Janeiro, Sao Paulo and Los Angeles.