Tell Me How You Shop Online and I’ll Tell You Who You Are

Business leaders are increasingly under pressure to grow sales and find avenues for innovation. Implementing rigorous price cuts or artificially increasing the salesforce quota used to work in the past, but not anymore, in a more digital world with well-informed customers who have greater expectations.

Disruptive technologies, in particular programming, have been perceived as distant and complex –a necessary evil for digitalization, a black box that is misunderstood and feared. Machine learning (ML), the principal machinery of artificial intelligence, is showing leaders worldwide exactly the opposite. Using algorithms (programs and instructions), ML paves the way for the creation of ideas, products and markets, capitalizing an asset companies already have: millions of data on their customers.

If customers log on to a virtual store (app or website) today with a username and password, ML can use their prior purchasing data to suggest to the system how they will behave: Did they go straight to the deals? Did they pay by credit card or PayPal? Which products did they look at before making a decision? Did they accept suggestions on complementary products? The system will be able to simplify their path and show the deals straight away, suggest paying in interest-free monthly installments, or recommend a product that not even the customers knew they needed.

What happens when customers decide to browse without signing in? Since there is no browsing history, ML has to ‘sniff out’ their behavior another way. The apps installed on their phones could indicate if they use a competitor’s services, if they are concerned about their mental health, or the type of news they prefer to view, thereby creating an initial customer profile that reveals their digital level (innovator or follower), or their predisposition (from 0 to 10) to acquire our product.

The phone model, IP address, and even the percentage of battery left are other clues that will help ML to ‘polish’ this profile. Knowing where the customer lives gives an idea of which products are consumed most in that region, while battery use can provide information on personality traits, for example: Is the customer cautious or always living on the edge and borrowing chargers? For a bank, this information could be highly valuable to determine whether the person will make their car loan payments on time.

Customers will also make a series of decisions while they browse: looking for a category or product (male or female), reading reviews or checking other sites before completing their final purchase. With these data, ML can place the customer in groups of users with similar characteristics (clusters) in order to offer a differentiated experience. Customer demographics (age or occupation) no longer matter here, but rather how they behave at the time of purchase. A thirty-year-old man might behave in the same way as a sixty-year-old woman and form part of the same consumer cluster, so the system would show them both the same offering.  People can have different purchase behaviors on their phone or in store and, therefore, will be in different clusters in each case. A person could buy certain products online and others in store; the system should recognize this difference.

ML is not limited to the company’s internal data or to the virtual environment: a cash register could suggest a different action depending on the weather or traffic. ML can integrate millions of real-time data, which a business’s Excel would take weeks (and several servers) to process, and has the advantage of using models that have already been created and tried and tested. A cafeteria manager may receive suggestions on the supplies he or she should order every day, based on thousands of data and the historical correlations of that point of sale: climate, upcoming holidays, new competitors, active campaigns, local events, etc.

Many executives are under the impression that ML is a crystal ball that will magically discover customer trends. The reality is that each of the system’s decisions has to be defined. If the objective is to earn the customers’ loyalty, the system should lead them towards actions that will increase the perceived value of their purchase, such as offering a special discount on their next purchase. Suggested actions will be very different if the objective is to acquire a new customer, such as offering free shipping for the first purchase. In order to make the most of ML, the company requires specialized talent: data scientists who combine expertise in mathematics and programming, and individuals who understand the business and the disruptive potential of technology.  

ML is not infallible (nor are we humans), but it is limitless in its capacity, scalability and enhancement, the more data and models it processes. While leading companies use ML to innovate and stay ahead of their customers (to achieve competitive differentiation), others will continue to look on Excel for someone to blame for last year’s low turnover. The big question is: what kind of company do we want to be?

Originally published in Expansión.