Big data is only getting bigger—and thus becoming more valuable than ever. That’s why large organizations, suddenly awash in operational and customer data, are leveraging data science to help them sort through their voluminous datasets and uncover strategic insights.
According to Accelerate Your Data-Driven Transformation, a commissioned study conducted by Forrester Consulting on behalf of RapidMiner, companies surveyed said that they currently see 4.4 times average ROI on their data science initiatives, but in the next two to three years, they expect that number to grow to 6.7.
It’s clear that data science is critical to business operations now and into the near future. But what is data science? How do you define it? How does it relate to computer science? And most important, how can you apply it to your organization so that you aren’t left behind? Let’s address each of these questions in turn to help you better understand data science and why it’s so critical to every kind of business.
What is Data Science?
In the most basic sense, data science is a multidisciplinary approach to unlocking value from data by applying statistical analysis, machine learning, artificial intelligence, and other advanced analytics techniques.
Data science is not just a set of tools, though. After all, anyone can gain access to basic machine learning algorithms, plug in some data, and see results. Data science provides necessary insight about what these numbers mean, how data should be collected and managed, and how analysis is best applied to answer core business questions.
It’s important to note that, because the goal of data science is to use data to drive insight and create business value, the work of data science isn’t (or at least shouldn’t be) conducted only by data scientists—anyone with domain expertise and business acumen should be able to harness the tricks and tools of data science to drive insight for their organization.
How Does Data Science Work?
Data science isn’t magic—anybody can look under the hood to see the statistics and algorithms driving the analysis. In fact, much of the data science process is intuitive, beginning with figuring out how data science questions should be framed, which is why domain expertise can be so valuable.
Five steps to data science insights
Let’s briefly walk through the five key steps to any data science project to give you a sense of what happens at each stage, and how everything fits together to create business value.
1. Asking the right questions
This sounds obvious, but asking the right question is absolutely essential to getting actionable and insightful data science results. For instance, if you’re posing a question that’s too broad (“how can our business save money”) or too ambiguous (“what type of customer is best for our business”), you’re never going to get an adequate data solution. Data scientists can help you better frame your questions (e.g., “when are the most energy efficient times to run our heavy equipment” or “what demographic segments provide the highest lifetime value for our service offerings”) so that you get clear answers and actionable directions.
Churn is another great example. Many businesses would like to be able to predict when customers are going to churn, but simply knowing that fact isn’t terribly helpful. What’s critical is that you frame the question in a way that is both answerable and valuable; you really want to be asking is “how much of a discount should I offer my customers who are going to churn so that they decide to stay, while also maximizing my profits.”
2. Getting the right data
With the right question in hand, a data scientist can then help determine how (and where) to collect the data necessary for analysis. In some cases, this will be data that you already have or are in the process of collecting. This scenario is becoming more and more common as digitization takes place across industries, and in some verticals that have embraced digital transformations—such as manufacturing and the Industry 4.0 revolution—it’s very likely you already have data to answer a lot of critical business questions.
In other cases, you may need to seek out new data sources, or think about how you can implement collection processes into your current workflows to start building a database of critical business information that can be used for projects like this. But it’s important to remember that your data will never be perfect; don’t put off getting started just because you think you need to wait until a hypothetical future when the data is exactly what you want. In our experience, people who decide to wait usually end up not starting at all.
3. Cleaning and wrangling the data
Because data is never perfect, data cleaning and wrangling is one of the most important parts of the data science process. It’s the step when you prepare datasets for analysis by fixing or removing data points or data categories that are inaccurate, incomplete, duplicated, corrupted, incorrectly formatted, or otherwise misleading. Data cleaning is a critical step in data science, as neglecting to remove things like strong outliers and irrelevant categories can lead to completely unreliable or uninterpretable results. The cleaning process can often take far longer than the actual analysis, with an oft-cited metric saying that data scientists spend up to 80% of their time on data wrangling and cleaning. That’s why if you want good results, you need to have a good data prep process in place.
4. Analyzing the data
This stage is obviously the core of data science, and it’s an opportunity to leverage all of the field’s techniques and knowledge to uncover valuable information. Data analysis and model development are an iterative process, and various parameters might need to be adjusted as you go along.
This is another reason why effectively framing the question you’re trying to answer is so important. If we return to the example of churn, after you’ve identified the factors that make a customer likely to churn, you might think that you’re done with your work. But if you have a better question in mind—“what action should I take to maximize my profits?”—you know that identifying customers who are going to churn is only the first part of your analysis; you also need to figure out what actions you can take to help keep as many of them as possible, while providing the lowest discount possible.
5. Communicating the results
The best data in the world isn’t worth much if it isn’t comprehensible. That’s why data scientists need to be able to not only crunch numbers but also explain what their analysis means and how it can be applied. Data visualization is obviously a great way to facilitate this. The secret to displaying data is understanding what key points you’re trying to convey, ensuring that those insights can be intuitively understood by viewers, and then providing enough context that key relationships and trends are also accounted for.
Once data is effectively visualized (or broken down within a report or presentation), it will be easier for both other data scientists and key decision makers to begin interpreting it, drawing conclusions, and taking suggested actions.
Data Science Versus Computer Science
Data science and computer science are closely linked, but the two fields do differ in several important ways. Data science is ultimately about studying, storing, managing, and analyzing large volumes of data, while computer science is about the operating methods through which digital data can be isolated and manipulated. Computer science is what computers run on; data science is what businesses (should) run on.
How Can Data Science be Applied?
Now that you have a firmer grasp on how data science is conducted, let’s look at a few specific use cases that help highlight the true breadth of data science processes in different industries and domains.
Online ads
Data science helps ad vendors like Facebook present ads to consumers likely to have an interest in the content of the ad. That’s because data science helps structure the type of data generated and continuously evaluate metrics related to demographic reach, performance versus cost, and conversion rate versus media type. You can use these techniques to optimize your own campaigns and ensure that the right audiences are seeing your messaging at the right times.
Recommendation algorithms
Data science is why Netflix knows you’d enjoy binge watching Ozark right after you finish Mindhunter, and why Amazon keeps suggesting baby toys after you view a few onesies. Algorithms help companies predict what customers will enjoy, based on existing information about them. Any business can benefit from this kind of insight, as it allows you to change the way your website communicates with users and personalize customer engagement-based preferences and purchases history.
Image recognition
Image recognition has advanced tremendously in recent years with AI and deep learning, and can now reliably be employed to identify people, places, logos, patterns, colors, and shapes. The uses for this are nearly endless, beginning with greater automation and extending to quality control, targeted advertising, and security applications.
Speech recognition
Data science is behind speech recognition technologies like Alexa, Siri, and Cortana that allow people to interact with their devices, homes, and cars. Even less robust versions of these programs—now available through simple APIs—allow anyone to interact with computers or digital devices and get responses to vocalized questions and prompts.
Price determination
Finding the best price for a product isn’t as simple as it once was. Today, factors like your IP address, how many times you’ve viewed a price, and larger forces like demand and weather all factor in. Using complex algorithms to determine when your business should buy critical products—and what you should be charging for your products—can translate into substantial savings and profits.
Fraud detection
Machine learning is at the core of modern fraud detection. There’s never enough time for a human to review individual cases to determine whether there’s a possibility of fraud (let alone comprehend the sheer number of potential variables). That’s why algorithms that are used to identify purchasing patterns and irregularities, halt potential fraud immediately, and take appropriate actions—including escalating cases to humans where appropriate—can lead to major savings for business.
Delivery logistics
Determining how variables like weather, traffic, driver availability, regular vehicle maintenance, and legal requirements come together is a daunting task for any human. Data science techniques allow modern businesses to take these factors into account simultaneously and derive accurate estimates that your business can rely on.
What’s the Future of Data Science?
The fact that investments in data science and machine learning are simultaneously coming from governments, educational institutions, private companies, and even curious individuals suggests the field has finally moved past the boom-and-bust cycles that defined its early years. Current investment is driven by recognition that data science is consistently delivering ROI through proven and replicable use cases across industries.
The key for most businesses starting out with these technologies is empowering them to see beyond the hype and complexity to the real value that data science can provide.
If you’d like to learn more about how other companies are using data science, machine learning, and advanced analytics, check out Accelerate Your Data-Driven Transformation, a commissioned study conducted by Forrester Consulting on behalf of RapidMiner.
