From telecoms to finance, e-commerce to government, predictive models are being utilized across various sectors to tackle all kinds of business problems. Companies that have yet to benefit from this practice need to examine the ways in which they can do so…
You can download PDF version of this whitepaper here.
Using Predictive Modeling to Address Different Business Problems
For thousands of years, people have had the desire to (or claimed they could) predict the future. This desire to foresee what lies down the road is a common one among individuals, each of us wanting to know what our lives will be like one day (be it in regards to happiness, wealth, health, etc.). Naturally, companies also possess this desire, wanting to know whether certain products or services they plan on releasing will be successful, whether their customer base will expand or shrink based on a strategic decision, or whether their investments will pan out as desired.
Thankfully, the rise of the digital era has partially enabled this, (with the help of databases and the power of analytics), taking shape in the form of predictive modeling.
Predictive modeling, by definition, is the analysis of current and historical facts to make predictions about future events. Several techniques – according to the nature of the business problem and current conditions – can be used when conducting predictive modeling. These include regression techniques, time series models, decision trees, and machine learning methods, among others.
The phases of predictive modeling are rather straightforward, and involve activities aimed at ensuring a look into the past through the analysis of various data points will in fact help predict the future:
Companies from different sectors are using predictive modeling on a regular basis to add value to their business in several different ways. Some examples include:
Telecom companies, such as AT&T, use predictive modeling to predict customer demand for voice or data services by creating choice models. These models also enable companies to conduct “what-if” analysis by simulating the market, helping to predict results based on various scenarios. These ultimately help the company impact its bottom line through ensuring the right product features and pricing is used to ensure maximization of market share and revenues.
One of the most common usages of predictive models in the telecommunications sector is around predicting churn, trying to understand which customers are more likely to be leaving the company within a certain time frame. Cellular One, a telecommunications company in Puerto Rico, is an example of a company that does this particularly well. Every one of its customer service agents knows which customers are at-risk, and are able to offer tailor-made benefits to these customers during their interactions. Cellular One claims this practice has helped it reduce churn by 33%.
Estimating Customer Lifetime Value
ING Belgium uses predictive modeling techniques to estimate the potential value of a given customer over their entire lifetime (how much revenues and profits a given customer can generate for the company over xx number of years). Based on the findings for a given customer, resources are then reallocated accordingly, such that those with high potential lifetime value (though not necessarily valuable today) receive an added level of service.
Estimating Credit Risk
Predictive credit risk scoring is a stronghold of customer analytics in the finance sector. Financial institutions, for example, can estimate the likelihood of a loan being defaulted on by looking at several variables (such as income, credit history, outstanding load balances, etc.) that directly correlate to default behavior. Wachovia Bank uses this type of predictive modeling on a regular basis – by looking at financial indicators, demographics, and life events (such as divorce, loss of job, new business, children going to college, etc.), the bank changes the way it strategically addresses each and every one of its customers.
Capital One, known as a “test and learn factory” in the financial services sector, used predictive modeling to identify the most appropriate individuals to target for each specific campaign that will be launched. This reflects in the over 300 business experiments that are conducted at the company each day in helping to fine tune their predictive modeling capabilities.
Predicting Startup Success
Financial services companies are also exploring the power of predictive modeling as it applies to new businesses, essentially whether giving a loan or startup venture capital is wise or not. One website (Younoodle) is trying to do exactly this – to that end, they have built up a large database of detailed startup information of 60,000 companies and 350,000 people. Three different predictive models have been built based on this database, models which generate scores that signify the potential future value of the startup, its growth potential, etc.
Determining the Next Best Offer
If you are familiar with e-commerce sites such as Amazon or Netflix, you are most likely then familiar with the “our recommendations” section. The offers in this section are tailored based on each and every customer’s past behavior and preferences. Thanks to an intense usage of analytics, companies are quite successful at predicting which customer will buy what next (next best offer). Netflix declared that from 1999 to 2006, revenues generated directly from the practice of analyzing customer behavior and creating customized offerings increased from $5 million to $1 billion dollars.
Junk e-mails, AKA spam, represent some 80% of all emails sent in any given day. Predictive modeling techniques are used extensively in helping to determine which e-mails are more likely to be junk. Companies like HP, IBM, and AT&T rely on predictive modeling techniques to determine which emails to label as spam when it hits their employees’ email accounts.
Improving Care Services
New York City Health and Hospital Corporation uses predictive modeling to predict disease related risks for each of its members. According to the risk scores assigned to each member, NYCHHC then prioritize the high risk groups to take preventive / proactive actions to lower the possible clinical consequences. With limited resources, predictive modeling allows maximized effectiveness of disease management programs.
Predicting Equipment Failure
The US Army has created several predictive models for the purpose of estimating how and when the various equipment it has on hand will fail. With such models in place, the US Army operation planners can more effectively manage required resources by answering how long they can rely on any given piece of equipment, or, when they should start seeking a replacement. With the value of equipment on hand in the hundreds of billions USD, the predictive models are of significant importance.
Optimizing Customer Service Levels
The Canadian Automobile Association (CAA) uses predictive models to optimize its customer service levels. Using results of a member-based survey around overall satisfaction levels in regards to emergency roadside services, CAA built predictive models to determine which customers need to receive an added level of service to prevent membership cancellation, thus allocating capacity-constrained resources to the customers that most need the assistance.
Organizing Air Traffic Systems
Airlines predict airspace performance with the help of analytics. Continental Airlines used predictive models so they can better organize and manage air traffic in United States, especially in territories where weather challenges are frequently observed.
How Companies Can Get the Most Out of Predictive Modeling
Designing predictive models that work is not as simple as 1-2-3. It requires that the company has a strong analytical team is in place, that it is truly dedicated to collecting data, and that it can take time to reap the benefits.
Some principles companies should adhere to in their trek to benefit from predictive modeling:
1. Increase awareness across the organization about what can be achieved with analytics
In many companies, analytics is perceived to be in the focus of BI or IT departments. On the contrary, analytical models (including the predictive ones) need to be owned and triggered by the business, in such units as marketing or sales. Executives need to make sure that employees in such departments have a vision about what can be achieved with analytics, and how it can help them in their day-to-day business activities.
Business analytics workshops can be held to help achieve this, with business intelligence experts sharing best practices with business units, showing how companies have used predictive modeling to impact the bottom line. This can then be followed by a roadmap building session, defining when and how predictive modeling will be used in the business units.
2. Define a data strategy to clarify which data should be collected, calculated, or stored
The single indispensible element of predictive models is data. A rich set of data fields with a high level of granularity can enable several different analyses/models to be generated – without the data little can be done. Thus, having a data strategy in place (specifically focusing on which data needs to be collected at what regularity) is of critical importance for ensuring predictive models can be generated.
Long lists of forms with too much detail frustrate customers where as a very limited number of data fields limit the potential benefits of predictive analytics. Rather, based on the desired predictive model, a company can work backwards to determine what data fields are needed at what detail. This is a step that can be built into the predictive modeling roadmap, as an early but vital step.
3. Manage data quality to keep it high
Predictive models, like with other analytical models, solely rely on data. Successful predictive models can only be designed with data that is of high quality. Companies should measure the quality of their data regularly and understand the root causes of possible data quality defects. The level of quality of the data should also be assigned to one individual or team, with data quality targets / KPIs in place to ensure acceptable levels are achieved such that predictive models and their continuity can be ensured.
4. Ensure a data-driven decision making culture is in place
To ensure that predictive models can be created, a data-driven organization needs to be in place, one that relies on facts rather than intuition in their decision-making process. As such, from top-down, a culture needs to be built into the organization that emphasizes the importance of utilizing data in all business activities to the greatest extent possible. Such a culture should even be reflected in job descriptions and performance reviews, whereby employees are made responsible for ensuring data is tapped into regularly, valued by the entire organization at every level.
5. Establish a test and learn environment
While establishing such customer analytical models such as predictive ones, establishing a test and learn environment is of critical importance. Models need to be tested over and over to ensure they become as accurate as possible in predicting whatever it is that wants to be predicted. “Rushing to market” with a predictive model just because it’s ready is not recommended. Rather, the model should be deployed behind the scenes and fine-tuned if necessary based on its accuracy. An example of this would be a model that is meant to predict customer churn – a company should check to see that the model accurately does actually predict churn before building strategies based on its predictions. Otherwise, the wrong strategy could be applied to the wrong customer based on flaws in the model.
6. Conduct pilots and spread the results to get buy-in
The most effective way to achieve and accelerate momentum around predictive models is through demonstrating bottom-line business results achieved because of the models. Companies should be communicating the learnings and achieved levels of success with pilots across the company. Any type of business case showing the potential success of large-scale rollout of the model can help with this. Predictive models stand to succeed once employees embrace them, which can ultimately be achieved through demonstrating bottom-line impact.
Be it small-scale models that help predict call volumes to those that predict a customer’s potential lifetime value, predictive models have proven their worth time and time again in practically every sector on practically any issue. Companies need to strive to ensure they make these models a part of their day-to-day business to the greatest extent possible.
To learn more about building or enhancing your predictive modeling capabilities and practices, please contact email@example.com.