According to Netflix, orange is the new black. In the marketing world, machine learning is the new big data. Last year marketing teams across all industries, including retail, waded through their big data sets to manually pull out the useful conclusions. But what if there was a system that could do that for them? That’s where machine learning comes in. While machine learning could yield major benefits, it is first important to know how it works and answer the three key questions to determine if it’s right for your marketing strategy.
A little background
Machine learning is a branch of artificial intelligence that deals with the construction and study of systems that can learn from data. For example, a machine-learning system could be trained to score marketing leads as either “worth pursuing” or “not worth pursuing” based on the performance data of those leads in question. From a marketing perspective, the main takeaway is that machine-learning models are trained. Consequently, the effectiveness of machine learning is highly dependent upon the quantity and quality of training data available.
Retailers have utilized machine learning for many years. Amazon uses it to recommend products that you may want to purchase, while Netflix has machine-learning systems that determine which movies you may be interested in watching. Additionally, retailers like Target assign every customer a Guest ID number that collects purchase history to provide them with relevant offers. Many other retailers use machine learning to weed out good transactions from the manual review process or to analyze marketing lists.
There are lots of options for obtaining a machine-learning service. You may have heard that machine learning requires a data scientist on staff, but that’s not always true. Today, there are a variety of vendors that offer retailers of any size machine-learning services without having to hire a data scientist. That’s great news because those salaries can be anywhere north of $100,000. The costs of these solutions are widely variable and are often based on the volume of data that is being analyzed. So depending on the amount of data you have to analyze, implementing a machine-learning system could be more affordable than you think.
While considering the types of technology available and the possible overhead concerns is important, it’s best to first start by examining how well a machine-learning system fits within your marketing strategy. Here are three key questions to get you started:
Do you have a clear definition of “good” versus “bad”?
The more clearly you define your end goal the better the machine will be at building a model that predicts the outcome. While the types of inputs can be extremely varied, the outcomes need to be clear. For example, if you are a retailer, then “good” is whether the customer bought something over a period of time, and “bad” is if they didn’t. The machine-learning model will take those outcomes and build models based on the types of customers that are more likely to buy versus not buy.
Do you have complexity of inputs?
If you sell umbrellas, the inputs you would use don’t really require a machine-learning solution because it’s fairly easy to predict that customers will buy umbrellas when it is raining. On the other hand, say you are selling more complex goods like electronics. If you have a variety of inputs that determine sales and if your transaction values are variable, then machine learning could help you identify which inputs are more likely to predict whether a customer will be a big spender, small spender or spendthrift. More involved inputs could include time of day, domain name, IP address, geo location, and referral source as part of the machine-learning model. These complex inputs can help a machine-learning model tell you that the customers from the Western USA who are shopping from work in the morning tend to spend more. Or, on the risk prevention side, these inputs can help a machine-learning solution separate low-risk transactions from high-risk, streamlining your review and collections processes.
Do you have enough data?
In order to build a good predictive model, machine learning needs lots of data. Lots of data can be defined as either a high volume of data elements from which to pull statistical samples or it can be defined as lots of iterations on a smaller data set where outcomes can be tested more frequently. Ideally, you would have both.
Where did you land?
If you answered no to any of the questions above, then there are likely cheaper, simpler statistical solutions out there that will meet your needs. If you answered yes to all of the above questions, then machine learning may be for you. There are a number of companies out there building specialized machine-learning solutions worth investigating, and then there is always the option of building your own, or just hiring a data scientist to do it for you.
Whitepages PRO is a division of Whitepages that provides customer contact information for businesses.