How to Choose a Data Science Use Case

Use Cases & Projects Marie Merveilleux du Vignaux

When choosing a data science use case, the ultimate goal is to leverage data to support a business decision or action. In simpler terms, just ask yourself: “What question am I trying to answer?” To respond, you will need to communicate and collaborate with your teams to determine any unsolved issues or ideas the company can benefit from — inclusivity is key to identifying all potential use cases or questions you will want to answer using data. Once you have put together a thorough list of potential business questions, you can use the following steps to choose the data science use case that is right for your organization.

With dozens of potential use cases but limited resources, it is important to prioritize projects that have both high business value and a high likelihood of success.”

→ Read more:  Defining a Successful AI Project

A 3-Step Process

The process of choosing which data science use case is the best fit for your company can be split into three steps. Each step determines one important aspect that you must consider when deciding which use case is most relevant to you and which one to prioritize in your planning. You can then mimic these steps for future use cases.

1. The Business Value

This first aspect involves two types of value: the value in terms of people and the value in terms of capital.
  • Who will this project benefit?
  • What are their needs?
  • What are their current processes and habits?
  • Will this use case help internal or external agents?
  • Will it advise one role in particular?

These questions are crucial, as data science teams will design the project with the audience’s application of the solution in mind. Therefore, you must begin by defining your audience. Once again, remember that collaboration is key — those developing AI projects need to work closely with business stakeholders or other subject matter experts to not just choose, but truly understand the audience.

As for capital value, ask yourself: “How will this use case specifically improve experience or outcomes and how can this improvement be measured?” Data science use cases should focus on opportunities with real and measurable business results. In order to keep track of and demonstrate the return on investment (ROI), you must think about how you can quantitatively measure the progress of your data science use case once it has been implemented. Before you do this however, you need to know what you are looking to analyze exactly. Begin by asking yourself and your teams how this data science use case helps you make money, save money, or do something else that you can’t currently do.

Another point you should ask yourself at this stage is: “Why is using AI for this purpose better than existing processes?” Don’t create a new data science process just for the sake of using AI. Your data science use case has to bring concrete additional value at scale that can justify the effort devoted to the project. Take a look at your existing processes and their metrics. If your data science use case does not provide any larger value or ROI than existing processes, go back to your list of potential use cases and choose another use case.

2. The Level of Necessary Effort

Data science use cases can take different shapes and sizes. Some will require hours of work and others months of work. Either way, the time needed to complete the use case needs to be justified, optimized, and well organized.

Where will the data come from, and does it already exist? You might want to consider minimizing external dependencies for your data science use case as this will not only reduce complexity, but also steer teams toward early projects with a high speed to value, which is important for gaining organizational traction. As repeated during this process, collaboration between the business, IT, and data science groups will be critical for this consideration as well, so involve relevant stakeholders early and often in project design.

When should an initial working prototype and, subsequently, a final solution in production be delivered? Like in every project, the timeline of the deliverable should always be omnipresent. In order to build credibility with internal stakeholders, it’s best practice to have a limited slice of the solution working from end-to-end in a short period of time rather than aiming for a fully baked solution in the first pass. Note that your project timeline should include time for development and a deadline for a working prototype, but also deadlines for deployment into production and for second (or third) iterations, as per the MVP methodology.

the simpsons effort gif3. The Likelihood of Success (the Risk)

What is the upside if it succeeds and what are the consequences if it fails? As with every project, your data science use case will come with a risk. While you may have already grasped the benefits of your project, it is important to also be aware of the risks that might occur if worse comes to worse. It’s important for team morale and motivation to remain optimistic, but the reality is that 87%of data science projects never make it into production.

Thus, it’s important to evaluate the consequences should the worst happen and assure that you are choosing a use case that won’t have devastating consequences in case of failure. Consider project scope, access to subject matter experts, availability of analytical resources (both human and infrastructure) to identify the impact your project can have on your company.
man standing on top of a mountain with open arms
Now that you have a clear understanding of the steps to take to choose the data science use case that best benefits your company, it’s time to get started! As you begin to tackle more and more use cases, you will notice that it is not enough to scale. Indeed, you will then have to take the next step with capitalization and reuse.

You May Also Like

US, China, EU: How the Rise of AI Highlights Cultural Differences

Read More

Data Pipeline Struggles and Solutions

Read More

Why Is My Data Drifting?

Read More

DEEP BEERS: Improving the Performance of Deep Recommendation Engines Using Keras

Read More