No Business Analyst Left Behind

Products in Gartner’s Latest Magic Quadrant for Data Science and Machine Learning Aren’t a Good Fit for Business Analysts

 

In February, Gartner published its annual Magic Quadrant for Data Science and Machine-Learning Platforms for comparing 16 vendors according to their “completeness of vision” and “ability to execute” on that vision. As I read the report and looked at the relevant placement of the vendors, it became clear that most vendors prioritize the needs of the data scientist over those of the business analyst.

Data scientists have created the vast majority of the predictive models currently in production, but with Gartner’s recent prediction, “by 2019, the analytics output of business users with self-service capabilities will surpass that of professional data scientists.” Vendors will need to address the unique needs of the business analyst in order to help them become more proficient in data science, and turn them into what Gartner defines as “citizen data scientists.” More on that in a bit.

 

Anything that currently has to be done in an iterative manner is ideal for applying automation.

Today’s users interact with the products in the Magic Quadrant in one of two primary ways:

  1. Manual coding. Data scientists write code in Python, R, Scala, or Spark to build models using commercial platforms that license open-source data science libraries and content. DataRobot’s partner Domino (a Magic Quadrant visionary) is a perfect example of a platform that has taken this approach to market. This works well for the portion of data scientists who have invested the time to learn these programming languages, and have the experience to optimize the many parameters and options associated with each algorithm.
  2. Manual GUI. There is also a significant portion of users who do not have the time nor desire to learn how to write the code for building models. Several platforms have attempted to address the needs of these users by creating a visual workflow for analytics that generates the underlying code automatically. DataRobot’s partner Alteryx moved into the Leaders quadrant this year with this approach as an extension of the process that business analysts use to prep and blend data for modeling.
    Approaches that use a manual GUI require users to have two very important skill sets in order to develop the most accurate models: knowing in advance what type of predictive model is needed, and optimizing the data preparation and cleansing steps for the chosen model.

Users must understand the applications and limitations of all of the models at their disposal so they can    select the type of predictive model for their use case. For example, they need to decide if a logistic regression will yield the best result vs. a decision tree, and then insert the corresponding logistic regression model icon into their analytics workflow. If they’re unsure of which modeling approach to use, they’ll need to insert multiple model icons in parallel and review the results manually to determine the best model for their application.

Once the appropriate model is determined, they need to look earlier in their analytics workflow to review the data preparation and cleansing steps they used to make sure they are optimized for the chosen model. Data scientists call this process “feature engineering” because they often define new “features” (aka “variables”) that might be useful to the model to generate better results. As you may surmise, this can be a highly iterative process because the introduction of new features could actually change the type of model that generates the best results.

 

Our understanding of the limitations involved with manual approaches, and our belief on how the data science market will evolve, is why we chose to lead the development of a third route to model building — via automated machine learning.

Anything that currently has to be done in an iterative manner is ideal for applying automation. In fact, Gartner believes the automation of data science tasks will be key to helping data scientists increase their productivity, and enable business users to cross the skills gap to become citizen data scientists. They estimate that “more than 40 percent of data science tasks will be automated by 2020.”

Our understanding of the limitations involved with manual approaches, and our belief on how the data science market will evolve, is why we chose to lead the development of a third route to model building — via automated machine learning.

 

Automation for Business Analysts

Everyone benefits from automated approaches. The greatest benefits work out for business analysts who have advanced domain knowledge in Sales, Marketing, Finance, HR, etc., but lack the coding skills and detailed understanding to differentiate between each model type.

 

Instead of presenting what happened in the PAST, they can recommend exactly what their business should do in the FUTURE. Instant value to the organization!

Consider a business analyst who has significant experience creating detailed dashboards in Tableau. Imagine how much more sophisticated their analysis could be if they were able to use the same data to optimize the feature engineering process to appropriately train and compare dozens of machine learning models automatically. Instead of presenting what happened in the PAST, they can recommend exactly what their business should do in the FUTURE. Instant value to the organization!

Analytics authority Jen Underwood’s recent blog on how business analysts can advance their career with automated machine learning details the supply and demand statistics for business analysts with these more advanced skills. Organizations are facing a shortage of data scientists who can build models using the manual coding approach described above, and are paying higher salaries for data scientists and business analysts who can build predictive models using any of the three approaches I’ve described. If you want to learn more about automated approaches, join Jen for a live webinar - Advancing Your Analytics Career with Automated Machine Learning - on April 26, 2018 at 1 pm Eastern.

 

But, before you make a purchasing decision you need to examine the areas where they are applying automation, and determine the degree to which automation is used.

It’s important to keep in mind that automation in the wrong hands can have disastrous results. That’s why DataRobot has over 120 highly-accomplished data scientists on our staff, including four top-ranked data scientists, who collectively have achieved over 50 top-three finishes in Kaggle competitions. They’re focused on incorporating data science best practices for feature engineering, model training, evaluation, and more, into the automated blueprints inside DataRobot, while also putting the appropriate guardrails in place so users of all skill and experience levels can achieve the greatest results.

So, in 2018 you can expect to see a lot of the vendors in this year’s Magic Quadrant talking about doing some degree of automation for data science and machine learning. But, before you make a purchasing decision you need to examine the areas where they are applying automation, and determine the degree to which automation is used. That will help you decide if these enhancements will benefit a small subset of potential users, or enable an entirely new class of users – the business analysts – who can help you realize your vision of being an AI-driven enterprise.