If you’re like us, a quick Google search for ‘artificial intelligence’ or ‘machine learning 101’ does little to help your understanding of what exactly Artificial Intelligence and machine learning are, and can do for your business. In fact, you’re more often met with complex mathematical concepts often qualified by the author as ‘simple’ but are anything but.
First, let’s try to draw a distinction between the terms AI and ML:
- Artificial Intelligence is an all-encompassing umbrella term that represents a field of research, like, for example, Biology.
- Machine learning is a subset of Artificial Intelligence, specifically as it relates to software and the ability of the software (specifically, the algorithms) to adjust itself without human intervention.
Second, it’s important to understand what has changed in recent years that allows machine learning to flourish. The most relevant difference is that, thanks to Moore’s Law and cloud computing, computing horsepower has increased exponentially. This is enormously important as machine learning involves a massive amount of calculations take place – and that crunching takes time for computers to process and execute.
The effectiveness of machine learning depends largely on how quickly and intelligently algorithms can respond to or align with changing conditions. Today, that response can be almost instantaneous due to new technologies that allow computers to perform these calculations practically in the blink of an eye. This means that machine learning improves our ability to quickly solve problems, even in situations where it’s not clear what the right answer might be. For example, what products a retailer might recommend for different first-time shoppers.
Third, let’s break down the main components of a transaction using machine learning:
- Model: this is the predictor (i.e. Shopper ‘A’ has 90% probably of purchasing Product ‘X’ at time ‘Nexus’)
- Parameters: these are the signals used by the model (purchases, product views, email opens/clicks, etc.)
- Learner: this carefully adjusts the model based on the results of the model.
Consider this scenario:
Shopper X browses men’s pants, women’s jewelry and toddler pajamas. Eventually, she ends up purchasing jewelry and subscribing to email promotions.
Shopper Z lands on the same site and browses jewelry. She is recommended men’s pants and toddler pajamas, doesn’t purchase anything immediately but later returns and continues to browse through jewelry. After clicking through some product recommendations, shopper Z ends up browsing and purchasing a pair of men’s pants, and does not sign up for email promotions.
While both shoppers are browsing and purchasing on site, the machines observes all of these minute interaction points and derives meaning from them. Using that meaning, the machine powers recommendations for that individual and other shoppers. It continues to learn, in real-time, from the result of each and every prediction it makes. Shopper X’s behavior not only helps to build the model that powers her recommendations but it can also help power other shoppers’, like shopper Z’s, recommendations and predictions.
Check out the video below, from a session at Dreamforce 2016 featuring True Religion Brand Jeans, in which its Senior Vice President talked about the promise and challenges of personalization.
Machine learning makes it easier to know your customers whether they shop online or off, uncovering correlations that would be impossible for any human to uncover on their own. Retailers are in the early stages of implementing machine learning technologies but are already far down the path of understanding its possibilities. The ability to know and serve your customers on a 1:1 level will be huge differentiator for first movers.
Will you be one of them?