Insights | Analytics & Insights

Is Machine Learning Right for Your Project?

Machine Learning has limitless potential and can create incredible value for businesses across many different industries if a successful data science strategy is used —but there are significant pitfalls to avoid to harness its full value

Is Machine Learning Right for Your Project?

 

 

While Data Science and Machine Learning demonstrate undeniable value in multiple industries, the long development times and low success rates can be daunting. However, with the right approach, Machine Learning models can be developed and deployed quickly to create substantial value. Creating a sound business case, intelligently weighing the tradeoff of model performance vs. development time, and picking the right tools and technology for the job lead to quicker, more efficient, and more valuable Data Science solutions.

 

Building a Business Case

Defining a business goal is the first, and most critical, step to a successful Data Science initiative. Addressing these foundational questions ensure we choose the right Machine Learning projects, with faster development times and more value added:

  • What is the business problem to solve? Are we trying to reduce churn? Increase profit margins? Having a clearly-defined use case for Data Science significantly decreases time to value.
  • Is ML necessary for this business problem? If this problem can be solved with simpler logic, expending resources for an ML model is unnecessary.
  • What is being measured? A Machine Learning model needs a quantifiable metric that can be improved throughout the process. For example, if we are trying to reduce churn with binary classification, the key metric might be accuracy, precision, or recall. The goal would be to improve these metrics as the project moves along.
  • Is the data ready for a Machine Learning project? Do we have the data at the right granularity? Is our business case feasible (e.g., making predictions for one month in the future is more feasible than two years)? Are we capturing enough data?
  • What is the expected ROI on this project and how is it measured? Compare the projected ROI of this Machine Learning project to other past projects to determine if the project is a good use of resources.
  • Do we have the skillset and infrastructure necessary to perform ML modeling? If the answer is no, it is necessary to hire resources internally or hire an outside consultancy to get the job done.

 

Model Build and the Time to Value Ratio

If we are confident that Machine Learning is the best way to solve our business problem in a timely manner, there are tradeoffs that need to be considered.

There is a wide array of Machine Learning algorithms at our disposal. Some models require more development and training time than others but might lead to better results in the end. For example, a deep neural network might yield better results than a linear regression model, but the time and cost to get value from the linear regression model is typically less than that of the neural network.

Ultimately, the question to answer is “what is the most time efficient, simplest, and cheapest model that can yield me the results that I desire?”. If a logistic regression returns 85% accuracy with less effort required, it may not be worth it to throw time and money at a neural network if it only increases your accuracy by 2%.

Machine Learning can improve the speed to business insights, but is the time it takes to build the ML model adding enough value?

As seen in the chart above [1], taking longer to complete Data Science projects does not always lead to a proportional value add.

 

Tools and Technologies

An ecosystem of tools and technologies has sprung up alongside the wave of Data Science that makes model development and deployment fast and efficient. Some of these include Docker, Python, SQL, cloud computing, and many others. Python has a vast number of data cleansing and Machine Learning algorithms. It is also a user-friendly and flexible language, which makes setting up a Machine Learning problem framework comparatively quick.

To speed up the time used on data exploration, the Python library pandas_profiling is an excellent tool to consider. It provides high level statistics on your features and visualizations to get a better feel for your data, quickly! Docker can create a script with a pre-configured environment to deploy on any machine or server.

Cloud computing allows for quick scaling of applications with a matter of a few button clicks. Plus, it keeps costs under control as we are only billed for resources utilized. The cloud ecosystem provides tools such as Azure ML, AWS SageMaker, and GCP’s AI & ML services to help expedite the data science lifecycle.

Used harmoniously, these tools and technologies significantly decrease time to insights, deployment, and ultimately boost business value for your organization.

 

Ensure Speed to Value with Data Science Initiatives

Data Science and Machine Learning have helped companies unlock the power of their own data to make actionable insights and create immense business value. But to harness Data Science’s full potential, your team should have a clearly defined business goal, pick the models that will yield the most value in the minimum amount of time, and use the optimal combination of tools and technologies that save you money and time.

If you are considering Data Science to tackle your most challenging and high-value business objectives, RevGen Partners has the expertise, experience, and passion to help you succeed. Contact us today to schedule a quick chat.

 

 

Sources

[1] VB Staff. “Why Do 87% of Data Science Projects Never Make It into Production?” VentureBeat, VentureBeat, 22 July 2019, venturebeat.com/2019/07/19/why-do-87-of-data-science-projects-never-make-it-into-production/.

 

 

 

 

 

Subscribe to our Newsletter

Get the latest updates and Insights from RevGen delivered straight to your inbox.