Human beings have always wanted to be able to predict the future, and with good reason. Recognizing that the earth had seasons that were part of an annual cycle helped ancient peoples know which species to hunt and fish at any given time, when to collect crops and seeds and when to prepare for inhospitable weather before it arrived. Recognizing seasons was a great advancement for prehistoric societies.
That recognition was taken one giant step further more than 11,000 years ago, when early civilizations around the world began erecting large public structures to track the sun’s progress throughout the year. They wanted to know what was coming next, and how soon.
A 10,000-year-old arrangement of 12 pits and an arc, created in Scotland to track the moon, became known as the world’s first calendar in 2013. Most ancient calendars were lunisolar, in that they were based on the paths of both the moon and the sun.
What does all this have to do with predictive analytics? Hang on, we’re getting there.
In 45 BCE, Julius Caesar introduced a new calendar that did not rely on the position of the moon. To keep this new solar calendar accurate, the Julian calendar followed an algorithm of adding an extra day every fourth year. That was the beginning of Leap Year, and also one of the oldest arts and sciences in human history: predictive analytics. Thousands of years of testing different calendar models had produced one that generated the correct result by incorporating a simple algorithm.
The Julian calendar was such a good predictor of the daily relationship between earth and sun that it stayed in use until 1582, when it was revised by what became known as the Gregorian calendar. The Gregorian calendar remains the secular calendar used throughout the world today.
Throughout time, people have attempted to predict everything from the weather to the stock market to which horse will run the fastest to which email headline will generate a higher open rate. It would seem that predicting the future is in our DNA. But now we can do it better, because computers do it faster and more accurately than people do.
Predictive Analytics (PA) studies historic events and predicts future outcomes from those studies. It has become a staple of those modern businesses that make their decisions based on data instead of hunches. PA helps organizations maximize their resources by using Artificial Intelligence and Machine Learning to study past events and generate many models that predict future events in order to find the one that produces the most desired outcome.
While PA was initially available only to large enterprises, it is now taking hold among small and medium-market companies thanks to these shifts in the marketplace:
Data-centric organizations have raised the bar. Forward-thinking companies are making decisions based on data. These businesses are fast to take advantage of new opportunities because they see business intelligence in real time. They use Predictive Analytics to reduce risk, improve outcomes and go to market with confidence. Other companies in their vertical follow suit in order to remain competitive.
Faster, more powerful and less expensive computing. Predictive Analytics requires a lot of computing horsepower. Today, computing power has become a price-driven commodity, making machines with the power to handle Predictive Analytics available at prices less than the most basic computer cost 20 years ago.
Enhanced data collection and storage systems. Businesses collect first-party data from their customers via point-of-sale systems, online orders, social media interactions, phone centers, emails and more. They also collect second-party data about those customers from business partners and third-party data from companies that have no direct relationship with those customers. To save all that data and make it available there are on-premise servers, data warehouses, cloud servers, data lakes and more solutions on the way to solve even the most complex data storage challenges.
Even more data to analyze is on the way. The amount of data businesses collect is increasing at a massive rate. StorageCraft quotes a statistic from IBM. that 90 percent of all the world’s data is less than two years old. In other words, we have more data at our disposal than we will ever use. Even though some of that data will never be useful, the more relevant data you have about your customers, the more accurate your predictions about them will be. So, bring on the data.
Affordable software created specifically for Predictive Analytics. Businesses now have software at their disposal to handle all of their Predictive Analytics tasks. Some of this software is available free as open source software. Small and mid-market businesses that don’t have data scientists, data wranglers and data analysts on staff can engage data consultancies to lead their Predictive Analytics projects.
Depending on your point of view, all of this can read like something too good to be true, or, it can seem way too complicated. The answer is somewhere in-between. To help you understand why companies turn to PA, take a look at how they are using it today.
Thanks to widespread acceptance, Predictive Analytics (PA) is now integrated into many Business Intelligence (BI) platforms. Businesses recognize that the descriptive analytics provided by old BI platforms is important, but it’s not enough.
PA is used to improve results throughout organizations in a variety of ways, including:
Risk reduction. Banks and lenders of all types rely on credit scoring based on the applicant’s past behavior and current circumstances. PA also lets lenders know the likelihood that the buyer will pay off the loan early, which is another form of risk, as it deprives the lender of anticipated interest.
Cybersecurity. Increased sophistication of hackers and malware developers makes it impossible to recognize every threat by a tell-tale signature. Instead, today’s antimalware software relies on PA to provide real-time analysis of behavior in the network and quarantine the abnormalities.
Hospitality. Airlines use PA to automate ticket pricing. Hotels forecast occupancy rates and needed resources.
Manufacturing. Manufacturers predict parts inventory needs.
Automotive. Car and truck fleets schedule maintenance within projected windows of time while making sure they keep enough vehicles in service.
Entertainment. Netflix analyzes millions of options to make suggestions about what you want to see next.
Etail. Amazon does the same with products.
Utilities. Utilities anticipate demand as well as equipment failures.
Insurance. Health insurance companies identify fraudulent claims.
The following use cases are a quick look at some of the ways Predictive Analytics is used by marketers today.
Lead scoring. Instead of using best guesses to score leads, marketers use PA to study the behavior of previous leads and predict how close a lead is to being ready to hear from a salesperson. Leads that score as sales-qualified can be automatically sent to sales.
Lead segmentation. Marketers have different nurturing campaigns for different personas at different stages of their buyer’s journey, and PA automatically moves leads that are not yet sales-qualified into the right nurturing segments.
Content distribution. PA determines which content is most relevant to different customers and automatically serves the correct content on web pages and in emails, newsletters and direct mail.
Lifetime value. By finding former customers whose behavior and characteristics are similar to that of current customers, PA predicts the lifetime value of current customers. This figure is essential to determining the level of resources that should be expended in marketing to each customer.
Churn candidates. PA analyzes past customer behaviors — and how your company responded to them — to find patterns that indicate when a customer is going to stop doing business with you. As you become more adept at stopping churn before it happens, that process can be automated.
Upsell and cross-sell. By reviewing customers who have purchased additional products after their original product purchase, PA can predict which customers will be more likely to respond to upsell and cross-sell offers and which products to offer them.
We’ll go deeper into how PA is used in one specific field, direct mail marketing, in a bit, but first let’s take a look at how Predictive Analytics works.
Generally speaking, Predictive Analytics (PA) helps businesses solve problems in discrete stages.
1. Identify the problem you want to solve. Narrowing the focus of your project limits the scope and makes it realistic. A given project could be to predict any one of a number of outcomes, including:
1.1. Which customers are most likely to respond to a particular offer?
1.2. What pricing strategy will produce the highest return for the offer?
1.3. How many items are likely to be sold?
2. Make sure you have someone to lead the project who understands the problem you want to solve as well as PA. This person is responsible for identifying any existing data specialist needs and the source of third-party data to be used to enrich your data.
3. Cleanse and integrate your first-party data so it’s ready to be used. Poor data quality is a big issue with most companies. There are duplicate records and records with missing data. Integrating the cleansed data pulls it together from multiple sources, such as online sales, email, CRM, credit card purchases and more. Data integration also transforms the data so that each column uses the same format, no matter the format of the data source.
4. Enrich your first-party data with third-party data to create a fuller picture of your customers. Your customer list is matched to records in a dataset with data that complements your information. Data can be enriched a near-limitless number of ways, so choose those that are most relevant to your campaign.
5. Develop an accurate predictive model through an iterative process in which historic data is used as training data. This is called training the model. Machine Learning is used — sometimes involving different techniques such as neural networks, recommendation algorithms and deep learning — to generate many models and identify which are most effective. Models that pass muster are tested and validated to determine their predictive accuracy.
Once you have an accurate model, put it to work for your campaign. If your campaign is happening online, you’ll need your web developer and IT team to get involved.
You won’t need their assistance if you’re launching a direct mail campaign.
With its use of Machine Learning (ML), Predictive Analytics (PA) is a natural for direct mail marketing.
Like all forms of direct marketing, direct mail succeeds or fails on the numbers. For years it’s been said that the success of a direct mail marketing campaign is a matter of matching the right list with the right offer. Predictive Analytics helps perfect both. It does that by using ML to generate thousands of predictive models — in the time that human beings could produce a small handful — and identifies the models that produce the best outcomes.
PA tells marketers which customers are most likely to respond to a given offer and which are most likely to not respond. By focusing on those most likely to respond and eliminating those least likely, marketers can raise response rates while lowering overall costs.
Thanks to ML, Predictive Analytics tells marketers which products are most likely to be sold and predicts the average order.
Marketers also use PA to acquire new customers. How? By matching the traits of the most-likely buyers in the original list to find similar customers in a third-party list of customers. These customers don’t already have a relationship with the brand, so they are unlikely to convert at the same rate as existing customers. However, because they share the same traits as likely buyers on the existing customer list, these new customers will be acquired at a cost significantly lower than that of standard new customer acquisition techniques.
To sum it up, Predictive Analytics helps in direct mail marketing by:
Raising response rates 20 percent to 30 percent or more
Eliminating the wasted cost of marketing to customers who are unlikely to buy
Acquiring new customers at a fraction of the costs of standard new customer marketing and advertising
As mentioned in the beginning of this article, the Gregorian calendar has remained in use for more than 400 years. Given the speed of business today, many things that are built to be “evergreen’ are outdated in less than 400 hours.
The only constant with PA is that it is constantly fine-tuned as business circumstances change.
Predictive Analytics helps organizations stay current and competitive by showing them challenges and opportunities across all lines of business, and most certainly in marketing.
In marketing, PA is capable of providing all of the following benefits:
Enabling better campaign management
Lifting response rates
Boosting cross sells and upsells
Improving customer experience
In short, Predictive Analytics improves decision making in every area where it is applied.