What is a good data science use case?

Almost all industries can benefit from data science, which is reflected in the fact that currently most businesses significantly increase their data analytic capabilities.

It is easy to find many high-level descriptions of possible data science applications like fraud detection, predictive maintenance or recommendation systems. These descriptions mostly use buzz words such as AI, data science, machine learning and deep learning along with superlatives about their business potential. However, when it comes to real, tangible data science applications, business units often struggle with the evaluation and prioritization of use case ideas.

Step 1: Understand the problem

Use case evaluation should always start with understanding the problem from an end-user perspective. While it is very tempting to think about solutions from the beginning, focusing on the problem first helps to determine the major requirements of the solution like system performance and prediction accuracy. Furthermore, it helps to estimate financial and non-financial benefits of the solution. If possible, idea owners should include future users into this step.

Step 2: Assess data availability

One major prerequisite for a data science use case is (unsurprisingly) data. Having the problem and solution statement defined, the next step is to find the required data. Early assessment of data quantity and quality provides important insights into the feasibility of the use case idea. For this step, it is recommended to consult a data scientist, who is able to judge if the goal of the application can ever be achieved with the available data. Furthermore, it is important to check if the data usage is subject to restrictions, which means that data can only be used in a limited way, or — worst case — not at all. Many use cases turn out to be not feasible at this point due to lacking data availability.

Transparency about data as well as clearly documented data usage terms are massive enables for this step. Without a well-structured data catalogue, idea owners can get frustrated at this point and are forced to make assumptions about the data availability.

Step 3: Check business readiness

Business readiness is the major success factor for any newly introduced solution. Being fully aware of the implications before starting use case implementation is very important. For example, introducing a decision support system for a particular user group within the company will affect their way of working and change business processes. In order to leverage the potential of the solution, business units have to be willing to take the necessary steps to gain and retain user adoption. If the solution is a stand-alone digital product for external customers, the successful market introduction will require capabilities, which might be new for the company. Additionally, a proper analysis of the competitive landscape is needed.

This step should not discourage the ideation on new, disruptive ideas! The purpose is to think through the implications and enable stakeholders to make a conscious decision for the use case before much time and money is invested for the implementation.

Step 4: Evaluate technical complexity

If the use case idea has successfully passed the data availability and business readiness check, the next step is to evaluate the technical complexity of the implementation.

Additionally to the already collected information, this step requires more knowledge about how the solution should be build: Is there a scalable of-the-shelf solution that matches the needs or does it make sense to develop a custom-made application? Is it necessary to integrate data from different systems? What would be the complexity, estimated cost and time-lines for the implementation?

This assessment certainly has to happen jointly with the IT department, which can consult idea owners on different options.

Step 5: Determine the priority category

Finally, all available information about the use case idea is merged into a priority category, which is determined by using a value-complexity matrix. The identified benefits of the solution will enable idea owners to rate the use case on a scale from low to high value, even if financial benefits can only be estimated. Similarly, the available information about required business unit measures and complexity of the technical implementation is used to rate the overall complexity on a scale from low to high.

The high-value, low-complexity use cases are the “low-hanging fruits”, which should be implemented first. High-value, high-complexity use cases should be, despite their higher cost, the second priority. Low-value, low-complexity use cases should get third priority and only be implemented if there are opportunities to generate greater value with the solution in the future. Needless to mention that low-value, high-complexity use cases should not be considered any further.

It is crucial to be open-minded when doing this evaluation in order to prevent that ratings are (unconsciously) tweaked until the priority is as desired. Idea owners might like their use case very much and have a hard time to accept that it should not be implemented.

Some uncertainty about the use case has to be accepted in the beginning. However, following the steps above helps to identify the most valuable data science use cases in your company.