Author: Meghan Villard
If your organization hasn’t yet invested in data science, they will soon. From manufacturing, to healthcare, and even to the non-profit world, mining data for new and actionable insights has become one of the most effective ways to grow a business or solve a problem.
Data science differs from more traditional analysis by applying the rigor of other scientific disciplines to process and review large amounts of data, often making connections that would otherwise go unrecognized. However, just like any other business function, data science needs support to strike insight gold.
How can an organization set themselves up for success? RevGen recommends using this strategic framework to make the most of your data science projects.
Step 1: Set the Right Goals and Ask the Right Questions
Much like any other initiative, a data science project must be aligned to the organization’s broader goals. Whether it’s a project to uncover why employee turnover is so high, identify where time can be saved on an assembly line, or better understand how customers are responding to your marketing, there must be a strategic impact in mind.
Some of the pre-work involved includes:
- Defining the problem: Often the most difficult step, but critical to the project’s success.
- Determining the end-user: Who will be utilizing the outcome, and how?
- Defining the baseline: Understanding the current situation is imperative to understand the context of the goal and ultimate success of the solution.
- Determine the type of solution: Does this goal require exploratory analysis, or does it need a larger data product, such as a predictive model?
- Measure success: You can’t fix something you can’t measure. Decide on a clear evaluation metric and track that from project launch.
Let’s say your organization wants to understand the customer base and why churn has increased month-over-month. If you are the Chief Marketing Officer tasked with understanding and preventing churn, the pre-project checklist might look like this:
- Define the problem: Do we want to know the key drivers of what causes churn? Do we want to predict churn proactively to tailor our marketing campaigns? Although both problems are related to customer churn, the approach would be quite different!
- Determine the end-user: Will this be a one-time insight provided to leadership to determine Q2 initiatives? Or a daily refresh of churn predictions to be used by the marketing team to automate campaigns?
- Define the baseline: How do you define that a customer is gone? And how do you calculate this? When your company doesn’t have a subscription service, it’s actually quite difficult to determine if a customer has churned.
- Determine the type of solution: Maybe you need a predictive score by customer. How often does this need to be refreshed? What does the output look like? Where will this be stored?
- Measure success: Calculate the baseline churn and set the goal you’re after. This can be broken out by segment or persona, as sometimes there are customers that you are okay losing.
In many cases, this step requires the most attention, as there is usually a plethora of problems, opinions, stakeholders, and KPIs to consider. It’s especially daunting when considering how to measure success, as there’s often not a single “silver bullet” metric. Working through these initial questions will help your organization strike the right balance and start your project off on the right foot.
Step 2: Data Science in Action
Most people assume this step is the entirety of a data science project. It’s true that the data discovery, cleansing, exploration, feature engineering, and building a model takes significant brain power and coding – but there are many pre-built packages and frameworks to streamline this process.
Typically, RevGen’s Data Science team follows this pattern when approaching a new project:
- Data ingestion: Inventory the available data and quality of each dataset. Determine the key requirements (type of data, time frame, sample versus entire set, etc.).
- Data exploration: Dive into the characteristics of the data set, with an understanding of the business context.
- Data cleansing: Analyze the data for completeness, consistency, and accuracy.
- Feature engineering: Enrich the data set with transformed or additional logic stemming from the raw data. This step requires business knowledge and creativity, and typically provides the most information gain.
- Model build: Choose which algorithm addresses the problem and desired output as defined in Step 1. Model selection is also based on the performance, complexity, and maintainability of the solution. Define your input and output data and begin building.
- Test and train: Determine what is ‘good enough’ for your model output. Evaluate the model’s performance and adjust the parameters until you reach your goal.
There are many different methodologies with which to approach your organization’s data. For instance, machine learning is a very popular approach but isn’t necessarily right for every project. This is another reason that understanding the goal is key – it’s far too easy to realize in hindsight that your team might have been better off taking a different analytical path.
Step 3: Operationalizing Your Learnings
Of course, insights are only as good as what you do with them. As we mention in our Business Strategy to Execution framework, understanding your business enablers will help your project take that final step from learning to action. This means looking into the people, processes, and technologies that can turn an idea into a reality.
Let’s use our marketing example:
- People: Connecting data science and marketing teams to define the business problem and goal.
- Process: Update business process for how the marketing drip campaigns get sent to customers.
- Technology: Setting up a cloud platform to streamline your data science model.
The power of data science is truly awesome – in the oldest sense of the word. It can be a massive boon to every organization, but without strategy and action behind it, projects quickly become overwhelming and difficult to capitalize upon.
No matter what your current data science capabilities are, RevGen can help guide you through the process of optimizing your business with the power of data. Contact us today to schedule a chat with one of our in-house data scientists.
Meghan Villard is a Manager of Data Science at RevGen Partners. She is passionate about empowering clients to make data-driven decisions that deliver value to their business.
Susan O’Connell is a Director of Client Services who understands the ins and outs of data management, business intelligence, and the importance of bringing people together to create the best possible solution.