Data Science Economics

The field of data science is very exciting, especially for those involved with the IT side of the company. Intuitively, many see the importance of data science and tend to think of developing data science capability as a long-term investment that will improve the organization as a whole. But not everyone has this background and will view data science with a more critical eye. They want to know what the immediate results will be - operationally and financially. Approaching data science projects with this in mind will greatly improve the chances of success for getting them funded.

Priorities

The excitement of starting something new and people wanting to get involved can lead to losing sight of both organizational expectations and goals. Unless a directive from the top decrees that you should prove that you can learn data science, the projects you pick will be very important. Projects cannot serve to just highlight the fact that you can do data science, they must show a return on investment (ROI). The financial arm of the organization will grade your efforts using financial metrics, so finances must be considered when you’re prioritizing projects. Generally speaking, something with a high ROI expands within the organization by nature which will help promote the data science program.

When prioritizing data science projects, consider this:

• Does your team have the data science ability to successfully execute it?
• Does your team have the capability - access to data, software, access, and financially?
• How will the results of the data science be incorporated into operations and show improvement?
• What is the ROI? Will the returns be greater than the expenditures?

Answering these questions will require input from a diverse group of people throughout the organization and that they have some background knowledge in data science. It may be necessary have an education phase to ensure everyone has the basics of data science down so they can effectively weigh in on priorities.

ROI is a deep topic and specific methods for calculating should reference specific models LINK: ROI Investopedia

Where to Look

Prioritization of projects requires a list of potential projects and how they fit into the organization, so knowing where to look is crucial. Look for areas of the organization that have inefficiencies where small improvements drive sizable efficiency gains. These gains have to be proportionate to the cost of the team. This requires knowledge of the organization’s financial metrics, consider using data science and dashboards to get this information. This can be a tricky task, but there are some standard areas to look at depending on the operational area.

This is not an exhaustive list, but it highlights some areas to look at. For first projects, it may be easier to focus on tedious operations that can use data science to automate. The ROI is then straightforward on time saved. Whereas determining ROI on maintenance models or forecasting can be challenging. Another thing to consider is data science clustering. One model may not show great ROI, but if there are numerous similar models that can be used in the area this could show a stronger total ROI.

Framing a Model in an ROI Perspective

This is a basic walkthrough, but make sure to check if there are formal requirements at the target organization. Finance and accounting departments will have their preferred method of assessing the costs of a project, but be prepared to drive ROI considerations.

The Basic Costs

• Labor - Regardless of specialty someone has to administrate, plan, data engineer, data science, and conduct business integration.
• Support - Building a model will require support from other teams and services and their time costs money too.
• Software & Licenses - Tools related to data science have costs that must be considered.
• Meetings - This is its own category because meetings are a spearate beast to deal with.
• Maintenance - How much will it cost to keep the model running and operational?

Risks to Consider

Data science is a challenging endeavor and new to many organizations and people. It takes a while for people to get used to how to effectively use data science. Consider this when designing the program and make sure to build in education and training. Below is a list of challenges and risks to keep in mind.

• Influencing the system: Data science makes a prediction and suggests or supports a decision. Therefore, the data science owner now has influence over how the organization behaves. For example, a data science model that broadcasts its predictions on new sales opportunities in specific sales regions to all sales members will flood that region with sales efforts while leaving other regions neglected. The prediction of opportunity should be managed within sales to ensure resources are allocated properly.
• Placebo effect: Just studying and attempting to measure a system the organization can shift focus to that system and improve operations. The manner in which the models are released and used has to be considered so that the actual results can be determined. The risk is more than a degrading operation once interest shifts it is thinking the model worked and extending that approach to different areas of the organization.
• Lime Light: If an initiative is new and exciting, people will be far more willing to help, but once that newness wears off, the favors and support from other departments can dry up, grinding progress to a halt. It is important to learn how data science is fitting into the organization and work to get the team formal access and support for the services it needs.
• No Fruit: Most organizations have some low-hanging-fruit data science projects waiting for someone to tackle. These require little strategy and can be quick wins. Quick wins are great for getting a buy-in, but they can also delay the team’s understanding of the complexity of the organization and how a lasting data science program will address the fundamental challenges. For example, if the goal is to build a recommendation engine, the team will have to first understand how users use the site and if the current use patterns align with the recommendation engine. If the site has a high bounce rate (where users come to the site and leave immediately) then understanding why they leave is more important than a recommendation engine and data science resources might be better spent tackling that issue.

While this post covers many challenges associated with achieving and maintaining an ROI for data science models, data science is a great asset for a majority of organizations. An organization probably needs data science unless proven otherwise, so the risk is just when and what. A data science team that listens to the customers and is self-aware has a high chance of success. So, err on the side of data science!