Companies across all industry spectrums are still only at the brink of what’s possible in conquering the limitations of our cognitive capacity with machine learning (ML). While the technology, finance, manufacturing, and healthcare sectors have taken the greatest strides in applying the newest developments in this field, the opportunities for any organization are unprecedented. This new era of outsourcing decision making to machines will eventually reshape our lives. According to a survey conducted by Algorithmia, the business-case benefits for ML include reducing costs, generating customer insights, and improving customer experience.
ML is a complex subject, usually reserved for data scientists, yet management accountants can earn their seat at the table by contributing their business acumen and expertise in its implementation. After all, success of projects aided by ML isn’t typically measured by the level of sophistication in algorithms used, but rather by the substance of success realized. When it comes to understanding which levers to pull to drive business performance, no profession comes closer than management accounting.
MACHINE LEARNING FOR ALL
The rise of automated ML (autoML) democratizes access to benefits offered by ML with little to no code required. In other words, one doesn’t have to learn to program nor earn a Ph.D. in statistics to run these advanced procedures. Effectively, autoML helps realize efficiencies in ML processes through automation of the ML model life cycle. It could be viewed as a conveyor line of helping manufacturing ML models in a mass vs. manual process. A typical autoML process includes three core steps:
- Model Selection: Candidate algorithms are ranked based on a model’s predicted score, gradually adding data to increase accuracy and eliminate underperforming models.
- Feature Selection: The platform extracts data set features and a combination of features to examine subsets yielding the most accurate predictions. The result is optimized for the previously selected model.
- Hyperparameter Tuning: In this phase, an autoML solution optimizes performance by reducing the evaluation time of each iteration and finding the best choice.
Leading technology companies began introducing competing autoML tool kits in the last three years. Here’s a summary list of the offerings of the best-known names in the industry, in alphabetical order:
Amazon: The world’s biggest e-tailer also moonlights as the world’s largest provider of computing services. Its autoML offering, Amazon SageMaker Autopilot, was released in December 2019. Billed as a no-compromise solution, this no-code platform is true to the name “automated ML.” Users simply need to specify a column in their data set that they want to predict. While the model produces the results after a simple click, model quality can also be improved based on the iteration of recommended solutions.
In fact, the program shows the leaderboard of best-ranked solutions, and users can choose the best one or even modify parameters to favor speed of execution vs. accuracy of results. As promising as SageMaker Autopilot sounds, its biggest downfall is the fact that, at the time of writing, it supports only the algorithms of regression and classification.
Google: The owner of the most popular search engine introduced Google Cloud AutoML on May 3, 2018. Offering a mainly no-code user interface similar to other vendors on our list, it has a unique approach to solutions offered: While users can tap into Google’s own ML capabilities, they also have an opportunity to build custom models better suited for their business needs. When it comes to algorithms offered, Google uses niche cases to help users perform text analytics and video intelligence, among others.
Microsoft: The second-largest technology company in the world makes it easy for users to build ML models without code, at scale, and with speed. Unlike other offerings on our list, Microsoft AutoML is available from within a data visualization solution.
In fact, its Power BI product offers automated ML algorithms to perform text analytics and image recognition, as well as custom models utilizing other algorithms with or without integration of Azure Machine Learning. One practical example of the power of this solution is the key driver analysis feature built into Power BI, which automatically highlights metrics that make the most impact for the data being visualized.
FINDING YOUR TOOL KIT
There is certainly no shortage of business problems that finance professionals can tackle using autoML tools. The first step would be to find a tool kit that offers the best fit for the problem and business use cases considered. Most reputable companies offer free trial versions of their products, so users have an opportunity to get comfortable prior to committing to a specific vendor.
One option would be authoring a minimum-viable solution via a limited test-and-learn process where management accountants would formulate a hypothesis driving targeted business results and iteratively test it by running their data through an autoML algorithm. In cases when a model shows a different story than expected, keeping an open mind helps absorb seemingly implausible or radically different solutions that we as humans might not have had sufficient creativity to imagine.
While autoML opens up a host of opportunities for aspiring data scientists, it isn’t intended to replace actual data scientists who often need to implement solutions not currently supported by autoML. In the pursuit of lifelong learning, finance professionals can embark on their own journey to master the art and science of ML. Being the original data analysts, CMAs have already mastered the most challenging aspect of the data science skill set: business knowledge and acumen.
To learn the missing pieces in the fields of statistics and programming, you can turn to training on Kaggle, the world’s largest community for data scientists. Once you’re ready to progress to a more structured learning experience, various massive open online courses (MOOCs) can help you obtain ML minimal knowledge: Coursera, Udacity, or even Carnegie Mellon University. Lastly, when you’re serious about studying this subject, immersive boot camps (Springboard, Galvanize, etc.) could become a viable option on your path of becoming a data scientist.