Chapter 8
Convincing Your Management to Adopt Predictive Analytics
IN THIS CHAPTER
Measuring the benefits of adopting predictive analytics
Optimizing operational decisions
Developing unambiguous success metrics
Creating a proposal to management
Predictive analytics should be on every company's radar. If it isn't in your company's toolkit, take a good look at the surrounding competitive landscape: Margins are thinner; customers are more demanding and selective; customers have more and more choices. Companies that thrive in this environment are those that adapt to these changes and innovate to get ahead of the competition. They're lean; they're agile; they have embraced predictive analytics.
For examples of the effective use of predictive analytics, consider the success of Amazon, Netflix, LinkedIn, and Facebook. Amazon and Netflix have been around for about two decades, publicly traded for roughly half that time; LinkedIn and Facebook have been around for about a decade, publicly traded for only a few years. Although these companies are relatively new in comparison to blue-chip giants like Walmart or IBM, they are some of the biggest companies in the world. In fact, Amazon and Facebook have both surpassed Walmart and IBM in market capitalization. One major reason: These companies have pioneered the use of data and predictive analytics to make business decisions.
The market has rewarded this new generation of companies with very high stock valuations — and equally high expectations. To justify those high valuations, these companies are expected to grow their earnings at a very rapid pace for the next several years. The velocity of growth in these young companies reflects the explosive growth in data generation — and the use of that data to grow.
Amazon and Netflix are not just any ordinary e-commerce companies; they have transformed how we shop and watch movies by their innovative use of predictive analytics to perfect the retail business. Facebook and LinkedIn started as companies that provided an online service. They smartly used the predictive analytics on the data they were generating to create an even better service that is used by millions of people every day. So in essence, their data is the core of their product.
All of these companies collect, analyze, and monetize their data by using it to understand their customers' needs and offer them compelling and customized products and services. Without using data in innovative and ingenious ways, these companies probably wouldn't have grown as fast and as large as they have.
So, if your company wants to follow suit, the questions to ask are
- How did the analytics team or middle management (for companies that lacked data-science teams at the time) convince upper management to adopt predictive analytics?
- How did the executive management team have (or acquire) the insight to create data science teams?
Nothing like that had been done before in the open, which makes it all the more remarkable. At least we don't have to worry about that now. You can use the success of the pioneering companies as a real-world example to convince your management about the importance of applying predictive analytics to your organization.
Making the Business Case
Many CMOs and CIOs may already be including predictive analytics in the company roadmap, but may not yet know how to incorporate it into their business. Others may not have even heard of predictive analytics, or may confuse it with traditional descriptive analytics. After reading this chapter, you'll have an understanding of that distinction. If you're a company executive, you'll have a handle on the benefits of predictive analytics as part of your overall business strategy.
If you're in a middle management role, you're probably looking for ways to make effective recommendations to upper management. You want to take advantage of the buzz surrounding big data and predictive analytics — and find out how you can use it to benefit your company. After reading this chapter, you'll be armed with what you need to propose a predictive analytics solution for decision management to the executive team.
Benefits to the business
The benefits of adopting predictive analytics can be achieved in any company. However, larger companies see greater benefits than smaller ones due to the multitude of inefficiencies that can be found in large corporations.
In general, more established companies see greater benefits and have greater opportunity to
- Optimize their operations
- Look for new opportunities
- Acquire new customers
- Retain current customers
- Find new revenue streams from their current customer base
Startup companies may see a smaller benefit and defer adopting predictive analytics because their product roadmaps should already be prioritized for the next several months and should stay sharply focused on their current strategy. Smaller companies may also consider a predictive solution too costly and complex to implement. They may think differently later on: After all the core business strategies have been executed and the company has begun to search for ways to grow customer lifetime value and improve customer loyalty, the time is right to add predictive analytics to standard operations.
An exception to the preceding paragraph would be a startup company that has already centered its business on data and analytics. Whether such a company is already doing predictive analytics or already has the staff and tools to create a predictive analytics solution, cost and complexity may not be a barrier.
Companies are having a much more difficult time gaining an advantage over their competitors. Many competitive advantages that leading companies enjoyed were eventually eliminated as competitors adjusted or copied the leaders' practices. This is especially true for companies offering the same types of products.
Adopting predictive analytics is one of the last process enhancements that many companies pursue to gain a competitive edge. Most companies start with traditional business intelligence to answer:
- What has happened in their business?
- Why did it happen?
They use dashboards and ad hoc queries against their databases to create reports for the management team.
- A dashboard is a graphical display where users, at a glance, can view summaries of predefined key information about the business.
- Ad hoc queries are non-predefined and non-routine queries on the database, used when specific information is needed.
Using these tools just gives those businesses an understanding of what has happened in the past. It doesn't give them any insight as to what will happen in the future. You can, of course, use standard math and statistics to try to forecast the future by extrapolating from historical data. But sometimes that approach leads to incorrect or non-optimized results.
There is very little chance that a data analyst, using standard math or statistics, can find relationships among the hundreds or thousands of variables that can affect the outcome that he or she is trying to forecast. For example, using just a few predictor variables may lead the analyst to forecast an uptick in sales for the next few months for a product. But predictive analytics may present an entirely different picture.
By taking more variables into account, and faster, predictive analytics can provide a more reliable answer to the question that every business wants answered: What will happen in the future? In this case, suppose the predictive model has predicted a sales decline instead of an uptick for the next few months. The machine-learning algorithm discovered dozens of highly useful predictor variables that affect the outcome, sometimes too subtly to be easily understood as a contributing reason or cause. Not until a particular predictor has presented itself can the team think analytically about why the outcome would depend on it; here the subtler predictors were brought to light. The power you gain from predictive analytics is the insight to make better-informed decisions.
Even so, predictive analytics is still currently under-adopted (and incompletely understood) by many companies. Some companies that have invested the time and money to understand and implement a predictive solution for their operational processes have enjoyed a return on investment (ROI) double of that provided by non-predictive solutions.
In this tough competitive landscape, predictive analytics offer a way to differentiate from the competition.
Improving operational processes
The primary benefit of adopting predictive analytics is its ability to discover actionable insights that aren't easily found by the naked eye. With these insights, the company can optimize business processes.
Predictive analytics are typically applied to operational business decisions such as
- Predicting trends in sales:
- Forecasting sales of a particular product
- Forecasting the revenue contribution of a company division
- Lead scoring (predicting the value of a sales lead)
- Predicting trends in marketing:
- Target marketing: Predicting which customers will respond to advertisement
- Cross-selling: Predicting which products a customer will also be interested in.
- Predicting which customers will churn (leave the business relationship):
- Predicting which customers will leave to a competitor
- Predicting which customers will cancel a subscription
- Prediction of fraud:
- Predicting which retail transactions are likely to be fraudulent
- Predicting unauthorized access to accounts
- Predicting trends in employment:
- Predicting which employees will quit their jobs
- Predicting whether an applicant is the right fit for the job
Historically, the specific applications of predictive analytics are usually associated with the particular industry a company does business in. However, predictive analytics can be applied to many more domains and problems. The applications are only limited to the imagination of your staff and available resources. Start small with a prototype and then scale up from there.
Using predictive analytics to make data-driven decisions provides a couple of major advantages:
It alleviates the stresses and inefficiencies of traditional processes.
This relief not only optimizes decision-making throughout the organization, but also establishes a consistent decision, free of potentially harmful influences from other outside factors.
It frees up human resources to handle other operational issues that can't be (or haven't been) automated.
The result is an improvement in the overall quality of operational processes, allowing the company to scale up its operations without having to add an equal amount of human resources.
Knowing the score
The predictive “score” is the best example of using predictive analytics to make data-driven decisions for operational processes. The best way to describe the score is in terms of a familiar example — the credit score created by the FICO company to help creditors make decisions on consumer loans. Several other credit companies create and sell their own credit scores, but the FICO Score is by far the most popular. Here are some examples of scoring decisions and their benefits:
Only contacting those sales leads in the CRM (customer relations management) database who have high lead scores.
Sales managers save valuable time and money by not calling contacts who won't buy.
Only targeting a segment of the customer database that has high response scores for a particular marketing campaign.
Saves marketing costs, decreases opt-out from email campaigns, and improves overall customer satisfaction rates by not contacting customers who aren't interested.
Contacting customers who have high churn scores.
Send a retention offer, such as a discount, to prevent customers from defecting.
Fraud system will deny transaction and/or send inquiry to customer regarding the suspicious transaction with a high fraud score.
Saves fraudulent payments made to criminals and enhances customer loyalty.
Table 8-1 shows some typical uses of predictive scoring.
TABLE 8-1 Typical Uses of Predictive Analytics
Use Cases |
Predictive Model |
Output |
Take Action |
Mobile phone contracts, cable subscriptions |
Churn model |
Customer churn score |
Send retention offer when score is high. |
Insurance claims, e-commerce transactions |
Fraud detection model |
Transaction fraud score |
Send claims to investigator or risk management when score is high. |
Stock trading account |
Risk model |
Portfolio risk score |
Send notification or route transaction to risk management when score is high. |
e-commerce, marketing |
Response model |
Customer response score |
Send advertisement to customers with high scores. |
Operational decisions are generally scored along a range of numbers. If the range is from 0 to 100, you can treat each score value as a percentage and allow the user to make a decision based on that information. Or you can create a rule that tells the system always to take action when a score is higher than a particular number (provided your business rules allow it).
Sales and marketing teams are among the largest users of predictive analytics. A 2016 survey by Gartner reported that analytics and business intelligence form the top priority for CIOs. That result is almost certainly in response to the rapidly growing trend among sales and marketing teams to use big data to deliver business insights. One popular and widely respected CEO even goes so far as to cite a Gartner prediction that a company's chief marketing officer will spend more on technology than the chief information officer by 2017.
Building relationships with your customers
Company representatives spend many hours trying to get to know their customers and build a relationship. As the relationship grows, so does the customer lifetime value (CLV): Loyal customers will continue to add to future profits. Loyal customers also refer new customers to your business. Your cost of customer acquisition eases off and your revenues improve — the result is closer to an ideal business environment.
Obviously, it takes time to know each customer — and it's practically impossible to get to know each and every one of your customers in equal detail. Most businesses find the acquisition cost for each customer way too high using the one-at-a-time method.
Predictive analytics offers a way to get a virtual understanding of the customer by combining and analyzing data from
- Registration forms filled out by customers
- Customer surveys
- Social networks
- Customer purchase histories
The machine-learning algorithm learns and predicts the customers' behavior. The data can be thought of as the aggregate of the customers' experiences; thus the algorithm is learning from experiences of all the customers. By scoring each customer for churn or responses, you can optimize operational processes across all customer touchpoints (instances of company contact with the customer, whether in person or through a message conveyed by salespeople, store cashiers, call centers, websites, direct mail, television, radio, billboards, and the like). This is the competitive edge you gain by applying predictive analytics on the customer.
Don't freak out your customer by obtaining or using data analytics in socially unacceptable ways. The result of doing so may be the complete opposite of what you're trying to accomplish. Customers are very sensitive to data privacy issues. For example, there was a big backlash when consumers found out that Target Corporation was using predictive analytics to predict pregnancy in customers.
IMPROVE CUSTOMER RETENTION
As the market becomes saturated, the cost of getting new customers (customer acquisition cost) generally increases. The supply of easy customers declines, so companies must work hard to keep the customers they spent so much to acquire.
Churn modeling is the application of predictive analytics for predicting a customer's propensity to leave the relationship with the company. If you know which customers are ready to leave, you can implement some type of retention program before they leave. It can be as simple as a discount offer for a product or service. The key is that you only offer these incentives to high-risk customers — keeping retention costs low and margins high.
Typical adopters of churn modeling are providers of service subscriptions such as those for mobile phone service. One implementation of such modeling is to integrate the churn score into the customer service application. Then the customer service representative can get a real-time notification to offer a discount for a new phone or service when the churn score is high, even during contact with the customer. The system can also generate a discount offer via text message, email or standard mail, if a customer's churn score is high when approaching renewal time.
IMPROVE RESPONSE RATES
Wouldn't it be nice to launch marketing campaigns that have higher response rates than you're used to? By applying predictive analytics, you can target customers with advertisements that make better sense in terms of expected return. The model can segment out a portion of your entire contact list with customers that have a high response score (those most likely to respond) for a particular product or marketing campaign. To increase response rates, companies must shift from mass marketing to such segmented marketing.
Segmented marketing also reduces the expensive side effects incurred by mass marketing. When enough ads miss the target, customers start to see your marketing messages as spam. Customers eventually become irritated and opt out of future contact. This in turn will reduce your contact list, customer satisfaction, and company image.
Many customers are only willing to spend a limited amount of money. By marketing products to them that are customized and have ranked high in the recommendation list, not only do you increase your response rate, you also increase customer satisfaction and loyalty.
Check your data regularly. Customers often change their emails or create multiple user profiles with different emails (that get forwarded to their main email). Find the email that is valid and try not to pollute their inboxes with repetitions of the same advertisement. You don't want them to get irritated and opt out.
Gathering Support from Stakeholders
Most people don't like change; convincing management or other stakeholders to adopt change is hard. To make matters worse, the adoption of predictive analytics is low compared to other forms of analytics. Company executives either don't know what it is or they've heard of it but don't believe in its value, so convincing them to see the value in it will require some work.
The good news is that, as hard as it sounds to make the pitch, it can be accomplished. Many new CMOs, VPs, marketing directors and marketing managers are technically savvy and value the benefits of technology. Some companies have product management executives that are also engineering executives. Some software engineering managers are changing roles into engineering product managers. These executives are looking hard for technological ways to find an advantage over their competitors and using predictive analytics is an innovative and unique way to distinguish them from the competition.
In any case, starting a predictive analytics initiative begins with the leadership team, as shown in Figure 8-1.
FIGURE 8-1: A predictive analytics initiative starts with the leadership team.
Working with your sponsors
In order to develop a successful and optimal predictive analytics initiative, the sponsors must think collectively about the problem(s) that they're trying to solve and how they will solve it. Developing a predictive analytics solution includes some indispensable activities:
- Thinking strategically
- Asking questions
- Defining success as a group at the beginning of the process
- Investing in human resources and tools
The upcoming subsections examine these developmental aspects.
STRATEGIC THINKING
As with most company initiatives, getting the executive team together in a room and having them agree on the strategy is critical to the success of this initiative. Doing so requires a lot of thinking about current objectives, business practices, investments, resources, and competition. Having a clearly defined plan from the executive team allows middle managers to prioritize and deploy the necessary resources to execute the plan.
ASKING THE RIGHT QUESTIONS
You need to know what insights you're looking for — and what operational decisions they help solve. If those aren't known right off the bat, then asking some concrete questions will help flesh out the answers. The executive team must answer these apparently simple questions in great detail:
- What do you want to know?
- Which actions will you take when you get the answers?
- What is the expected outcome?
- What are the success metrics?
These questions should lead to a series of follow-up questions. When the team is satisfied with the degree of detail in the answers, you should have a picture of the business problem they want most to solve. The time that the team invests in the beginning of this process — to define and clearly understand all aspects of the task — will contribute to the success of the project.
Try to think from the customers' point of view. Creating customer personas will help you understand who your customers are, their needs and pain points (problems that the business wants to solve). Ask questions that you think your customers would ask. Doing so helps identify the most important operational goals.
Defining success
To measure the expected ROI of the predictive analytics solution, you must be very specific when defining the prediction goals. Having an ambiguous definition of success can lead to misleading results and conflicting interpretations among stakeholders. Start with small projects that have clear outcomes. A clear win gives the executive team more confidence in predictive analytics generally, and can lead to additional projects of larger scope and value. A track record of success also boosts the confidence of the teams that work on the project, and can improve cross-functional and cross-organizational collaboration.
Time to market
Your first predictive analytics project will take some time, and that time will vary greatly depending on many factors such as the size of the company and what prediction goal is being attempted. Ask your marketing, analytics, IT, and engineering team to determine a timeframe for a small project. An alternative is to get an expert consultant who can discuss the project plan and review your infrastructure. Such a consultant can provide a cost estimate and a realistic timeframe.
Typically, most of the time spent is in data preparation. Most, if not all, commercial vendors provide capabilities to help you prepare your data for analysis. For example: After you load the data, you can explore the data and decide whether (and how) to handle columns that have missing data.
You may have the option to use your data in its current state — rather than having to spend precious time cleaning it — as you build a predictive model with. If you do so, your results may be less accurate than they'd be with thoroughly prepared data; it really depends on your specific situation. Your company may feel the time pressure to get the solution deployed quickly; dirty data may be less important than speed in such a case. Communicating to the leadership team that the results of the proof of concept may be dramatically different than a production-ready model is a must in this case.
Budgeting
Realistically, implementing a predictive analytics project will cost more than doing a non-predictive project. Such costs can include the addition of human resources in marketing, analytics, IT, and engineering. You need both people and tools to store the data, prepare the data, model the data, and deploy the model — which can make the whole undertaking appear too costly to many smaller companies. Larger companies can absorb the initial cost more easily, and may already have staff who can be reassigned to handle the task.
To determine whether to do a predictive analytics project, you have to evaluate whether the projected benefits exceed the costs. Predictive projects have returned up to two times the ROI realized by their non-predictive counterparts. If building a custom solution isn't feasible, products available from predictive analytics vendors can be of help.
Getting business and operations buy-in
After the strategy has been defined and the solution deployed, the key players in the success of the project will be the business and operations personnel.
You need people managers (such as program managers and project managers) to shepherd the team and coordinate with the other stakeholders. This group of people will keep the project on track, resolve road blocks, schedule cross-functional training, and help resolve any difficult cross-organizational issues that may arise.
Your product managers and analysts will have to create requirements, find meaning in the model's output (throughout the various stages of the project), and report the performance metrics to upper levels of management. Your customer operations team will make sure that the data is flowing smoothly between the company and the customer.
To get the full support of these people, you must be able to explain to them why these initiatives are being implemented — and why their routine operational processes are being altered. The project is about optimizing operational decisions, which targets the core of what many employees have been doing over time. The changes in their work requirements will take getting used to. They will also be the first people to feel the effects if the project is working — or if it's not.
Cross-organizational training
Cross-organizational training will be required for members who have customer contact, whether in person, by phone, or online. Salespeople will need to understand what the lead scores mean. And they may want to know how those scores were derived so they can use that information to strengthen their knowledge of the customer.
You must also train your customer support team not only to understand data scoring but also to apply it when the customer contact happens. Your customer service representatives must know how to respond to a high fraud score (ask for further verification or decline high-risk transactions) or a high churn score (offer a specific discount or freebie).
Standard operating procedures must be created to guide the interaction when such situations occur. To make things simpler, you can display the score as a color that shows the level of attention. You can also embed the correct decisions directly into the application.
A simple scenario between Customer Service and a mobile phone customer would look like this:
- The customer calls Customer Service to inquire about her abnormally high bill this month.
In Customer Service, a red indicator on the screen shows that the churn score is high; the customer is predicted to cancel her service.
The model scored this high because:
- Her service contract has about one month left.
- She has called numerous times in the past complaining about the high cost of service.
- She has only been a customer for a year.
- Action taken: The Customer Service representative offers a discount for a new phone if she renews for another year.
A more advanced version of this same scenario would change the churn score dynamically as the customer service representative inputs notes from the conversation. The model uses text and sentiment analysis to predict the likelihood that the customer will terminate her service contract.
Fostering a data-sharing environment
Being a data-driven company doesn't stop with using predictive analytics to create data in silos. The data should be shared across all business functions. As each team produces more data from the existing data they have, they should store it and make it readily available for other teams to use.
Another team may find just the piece of data they're looking for — which would otherwise be a waste of resources to redo the effort. Just as data should be shared, so should the techniques used to produce the data be shared. To make efficient use of the knowledge across the entire business ensures that the business reaps the maximum return on investment.
Getting IT buy-in
Predictive analytics is a science. To ensure that you get the best possible results, you need the talents and skill-sets of your engineering team. IT plays a critical role in the success of your predictive analytics projects.
In addition to data scientists (especially if your company lacks data scientists), you need computer scientists and software engineers who are experienced in machine learning, natural language processing, and data mining. They can help you choose the appropriate algorithm (say, a clustering or classification algorithm or decision tree) to build your predictive model. IT may also own the databases to which you would need access. They're the ones who set up access rights so the data doesn't get accidentally modified or deleted during the creation of the model. After building your predictive model, coordinating with IT again becomes necessary to deploy the model you've built.
From its inception, to data access, data integration with third-party vendors, data preparation, data mining and applying statistical analysis, and finally to model deployment and maintenance, getting IT buy-in will be critical to the success of your project.
Building a data-science team
Building a data-science team is an essential part of your predictive project. This is the team that helps you create the predictive model to solve the business problem.
You can hire new talent if your organization doesn't have qualified employees who can perform the task. Or you can recruit from within the organization. Recruiting personnel from within to be part of your data-science team may actually increase your chance of success; these people already have the much-needed business domain knowledge. Nevertheless, the team must have a strong lead member who has experience with the whole predictive analytics lifecycle.
The team's background should be diverse; the team should encompass business domain experts, data scientists, and IT personnel. All business departments should have representation in the makeup of your data-science team. Your data scientists should have deep mathematical knowledge and preferably proficiency in your line of business, and broader experience building different kinds of models. IT staff should include computer scientists who know machine-learning algorithms, software architects who can help in the design and implementation of the whole project, database and data warehouse experts, and personnel experienced in running IT operations.
If your organization lacks the skill-sets you need to produce a predictive analytics solution, you can look into the trade-off between hiring a consulting company versus building a team in-house. Many enterprise-level software solutions are available that managers and business analysts can use to build predictive models without having to know much about machine learning or algorithms. You can hire consultants to build the initial models with the software and also train your team to use the models. Your team can maintain and enhance the models.
Depending on your specific needs, it may be easier and more cost effective for you to buy a simple-but-powerful tool designed to enable business users to build predictive models. Such software products have built-in workflow templates for predictive modeling. The tools make assumptions based on the data you're loading into the program, which can get you started right away. This option may be faster than hiring specialized and diverse professionals to build your data-science team.
Choosing your tools
If you buy a software tool for your predictive analytics project, which tool should you use? As with all important business decisions, you have several factors to evaluate. Choosing the right tool(s) depends on
- The budget
- The scope of your project
- The data you'll be analyzing
- The main users of the tool
You may need to hire consultants to do a feasibility study for you to see what tools are available and which ones would work best for you, given the business questions you'd like to answer. Experienced consultants who have working knowledge of at least a few predictive analytics tools — who know the pros and cons of those tools, as well as the business problems they're built to solve — will be best able to help you choose a tool. Or you can have them build a pilot model if that's within your company's budget.
As with all information technologies, predictive analytics tools are getting more powerful and commodity hardware to store data are getting less expensive. This is an impressive trend, given that data is only getting bigger and more complex. Look for predictive analytics to become more widely adopted as data availability increases and the technology to analyze it becomes more accessible.
The tools are being built with user-friendly interfaces. They're easy to use and come already powered with all necessary features for managers with technical backgrounds and data analysts to start using them to derive value right away. Analysts can select the data, process it, run multiple algorithms on it, and view results as visualizations or reports; all in a few clicks.
There are also open-source tools available online: They're mostly free of charge and have active communities where you can discuss and post questions. Keep in mind, however, that open-source tools
- Target more advanced users (some, for example, require programming knowledge).
- Are often less user-friendly (some lack advanced graphical user interfaces).
- Have limited capacity for handling big-data problems.
- May run out of memory when handling big datasets.
- May not be able to handle real-time analytics or streaming data.
- Require that you invest time in learning them.
Usually don't have a professional services division for problems and needs. You will have to seek consulting help from third-party vendors.
The learning curve and lack of commercial support may make an open-source tool unsuitable for wide adoption in a large corporation. Using a commercial tool designed as an enterprise solution can lead to a much smoother deployment.
Having too little or too much data
Data is abundant and getting more so; big data has become a familiar buzzword. Everyone is collecting it and talking about the large amount of data that is available. Organizations are amassing more data than ever before.
In this environment, it's hard to imagine a scenario where a company has a shortage of data. Whether such a scenario exists, the quickest way to get more data is to ask your customers to fill out surveys, or ask your customers to rate the products they've already purchased. Here the goal is to generate valuable feedback that you can put to use for data analysis.
You should have already stored all past transactions that your business has conducted. Building a database that contains detailed profiles of your customers can be of great value to your predictive analytics model. You can even buy third-party data.
On the other hand, having too much data also poses challenges. How do you determine which items are important? Where do you begin? How do you handle the constant influx? Increasingly, software tools and industry literature are addressing such big-data issues. These challenges will continue to grow as data complexities, in terms of data velocity and data volume, deepen.
Leading software products in predictive analytics are making it much easier to use big data, rendering its analysis more manageable. By the same token, those same software leaders are providing tools that smaller companies can use and apply effectively, even on smaller datasets.
Employing data you're already storing
As a result of normal operations, your organization must have already collected a fair amount of data it can use for predictive analytics. That data may not be analysis-ready, but with some effort you can make that data work for you.
One of the important steps in data preparation is to remove your duplicate entries. Make sure you don't have (for example) multiple IDs for the same customer profile. When you merged your data, you may have brought over the same customer with a different email address. The same customer may have registered multiple times by accident or through different channels. For openers, make sure that the email address you have for each customer is the most recent.
You can also buy third-party data, such as that derived from (and by) social media for the purpose of selling it to interested parties such as (perhaps) your company. Adding such data can help you build a better picture and richer profile of your customers — and it's readily available. Increasingly, data and its storage are affordable commodities.
To cut down on data preparation, you can enforce input checking at the time your customer information is collected, or (better yet) provide customers with multiple-choice surveys, and limited ranges of responses to choose from, to minimize data-input errors.
Buying data from third-party vendors
By itself, the data your company has may not be enough to provide you with a good basis for predictive analytics. For example, to produce actionable insights at today's ever-increasing pace, simply using demographics and basic inferences from your data isn't enough. The demographics of loyal customers may look exactly the same as demographics of churning customers.
You may also have transactional data from your operations to combine with customer profiles. But, in some cases, your existing customer base may be limited or belong to a specific segment of clientele. You might want to make predictions or target a whole different segment of customers; your transactional data may not be able to achieve that. Also, there are cases where you may have transactional data and very limited customer profile data. For example, you only have name and location, but you also want age, income, and gender.
To generate actionable insights, you'd need more complete data. To get it, you'd probably need to pull that data from various sources; one way of doing that is to buy it from third-party suppliers. Some suppliers offer what is called a DMP (data management platform). They can integrate your first-party data with third-party data to create personalization and segmentation models. If you can get hold of third-party data that's relevant to your predictive model, the next step is to integrate it into your existing datasets to create better models.
With the rise of social media, an abundance of data is being generated, some of which is collected by companies whose business model involves collecting this data, packaging it, and selling it. Such companies view that data as their product. Your company may have a use for that product.
Rapid prototyping
Predictive analytics is a complex science in every aspect. It requires a lot of strategic thinking by executives and domain experts. It requires technology and specialized skills. It requires time and money. When companies finally decide to take the plunge, they have to know where to start.
After the executive team has come to a consensus on what operational problem they want to solve, the spotlight is on you. You want to prove to them that predictive analytics will work. You want them to see some results in a month or two, not in six months to a year. Without their confidence in the science of predictive modeling, any attempt to launch a company-wide predictive initiative won't get off the ground.
When an executive wants something, time is your enemy. We know that most of the time creating a predictive model is in the data preparation. Don't spend that much time doing it. Creating a rapid prototype of the predictive model is the way to go. Such a prototype doesn't necessarily have to be pinpoint-accurate to be valuable; the idea is to prove the concept. You just want to build something for the executive team to see in action.
Take a clean subset of the data and use that for starters; that should be plenty for most prototypes. After the execs see the model at work, they may just get that “aha” moment that demonstrates the value of predictive analytics and makes a full adoption a lot more likely. Even if they don't get it right away, you may be buying some time with the demonstration. If they see that you're capable of building a predictive model, they'll probably wait a little longer to see clearer results.
You can build on the prototype and show incremental progress to the executive team. That way they can evaluate your progress and make recommendations along the way — which also firms up their relationship with the project. The investment is also allocated in small chunks instead of one big lump sum, which can make the decision to do the project much easier. Upper management loves small investments and quick returns.
Presenting Your Proposal
When you're ready to sit down with your executive management to propose a predictive analytics initiative, you must be able to explain predictive analytics clearly, and in simple terms. Granted, many executives already know about it (or have at least heard of it), but be prepared for those who have not. Be prepared to frame that explanation in many different ways.
Here are several definitions of predictive analytics that you can use for your specific needs:
- A tool that uses data to assist a business to make smarter operational and strategic decisions.
- A tool that uses advanced machine-learning and data-mining algorithms to extract hidden patterns that produce actionable business insights.
- A tool that uses computer science to compute the best action from data.
- Software that models future events based on historical data.
- Software that allows you to simulate business decisions and see their predicted outcomes.
- A technology that predicts customer behavior by using data from purchase histories, demographics, social media, and web logs.
Okay, here's another business truism: Executives are (by definition) busy, so they're generally pressed for time and skeptical of being sold to. You need to be concise. They got to where they are because they have been making decisions that got these results. They know what initiatives not to do, what initiatives to do, and probably have strong opinions about both. You had better have a thoroughly well-thought-out proposal.
With the right message, you may get their attention. To seal the deal, you'll have to wow them with a vision and a demo. When you have them interested, ask them to sponsor a rapid prototype.
Here are some tips on presenting the proposal.
- This is a complex topic, so expect — and be prepared — to answer a lot of questions.
- No presentation is complete without providing a summary and answering questions at the end. You can find answers to many of these questions in this chapter. Other questions will be specific to your situation. Generally you'll need to answer
- How will predictive analytics add value to the business?
- How long will it take?
- How much will it cost?
- Who's going to do it?
- Tell them at least one interesting story about who is using it and how they're reaping the benefits, before trotting out the details of how to implement predictive analytics.
- Be armed with success stories of predictive analytics as implemented by other companies — especially your competitors. If you can't find something specific to your domain, talk about these more famous success stories:
- Amazon's product recommendation engine
- Netflix's movie recommendation engine
- Facebook's recommendations based on users' “likes”
- LinkedIn's “people you may know” feature
Show them a list of specific benefits that your company will get from applying predictive analytics.
For a refresher, see the “Benefits to the business” section of this chapter.- You can frame the benefits in terms of an overarching goal: to gain a competitive advantage over your competitors by
- Optimizing company operations
- Looking for new business opportunities
- Acquiring new customers
- Retaining current customers
- Finding new revenue streams from your current customer base
- Show that the costs of missing these benefits are greater than the small cost of approving the pilot program.
Have a list of employees (and the reasons why) you think are ideal prospective members of your data-science team.
The pilot program should be short-term and small in scope. The employees on your team can still work on their current projects if they can't be dedicated full-time to the predictive analytics pilot.
Employees may find it fun and rewarding to work on something new to the company. Finding meaning in data is very challenging and gratifying. Big data and predictive analytics form one of the hottest topics in technology right now — and employees are sure to dive right in to learn these new skills if given the opportunity.Ask them to sponsor you to build a pilot model in a short period of time.
Use open-source tools to create the predictive model to solve a simple problem. For example, build a simple recommendation engine that would address a small percentage of your overall web traffic or a small targeted mailing.