Machine Learning for Business Managers

 As a data and analytics consultancy, we are regularly asked by business executives what machine learning (ML) is all about, and to that extent, have been running workshops with clients to provide further education around ML. 

As a data and analytics consultancy, we are regularly asked by business executives what machine learning (ML) is all about, and to that extent, have been running workshops with clients to provide further education around ML. 

We were recently asked to run a workshop for Executives in the Public Sector.  The Department has a relatively mature level of understanding of the importance of using data for decision making, has established data governance principles and practice, and there is a genuine desire to learn more about how data science techniques might help run the business more efficiently.  

Of particular importance to the client was learning more about data science, and specifically, trying to unravel what machine learning was all about.  They are increasingly hearing more about the impact of machine learning, advanced analytics and artificial intelligence, but struggling to separate the marketing hype from the reality. 

We came to the conclusion that to dispel a few myths and provide some hands on education we would focus our discussion around answering 5 key questions, which I’ve described below: 

What is data science?

It’s important to establish what Data Science is and isn’t, to ensure that everyone is on the same page, and to enable people to be clear about the types of skills and competencies that are required by organisations embarking upon data science programs. 

The key message is that it typically involves large volumes of data, and done correctly can develop business understanding and help create strategies that generate actionable results.  Further, machine learning falls within the broad field of data science and artificial intelligence. 

Impact of data science upon business Intelligence?

Most people have some understanding of business intelligence and how it can add value to a business operation. However, the inherent problem for BI users is that answering the question “what is likely to happen?” or “what should I do?” involves thinking about the future, and trying to predict what might occur. 

Whilst business intuition and understanding can help to answer these types of questions, it is the field of machine learning which is bringing new techniques, largely grounded in mathematics and statistics, to business people.


What is the best process for data science?

The discussion here centres around two globally accepted approaches (CRISP-DM and KDD), which interestingly enough were developed in the 90’s and 2000’s, and are as relevant today as they were back then. 

CRISP-DM is the Cross Industry Standard Process for Data Mining, which was developed in 1996 by a consortium of companies, and is considered the most widely used approach towards data analytics.    The process involves six stages, is iterative in nature, technology agnostic, and aligns the solving of business problems with analytics and data modelling techniques. 


The importance of walking through CRISP-DM is to demonstrate that machine learning is essentially part of a broader process that is iterative in nature, and is hugely dependent on collecting, manipulating, and transforming data before machine learning modelling can even by leveraged for problem analysis.

What problems does Machine Learning solve?

This is where the workshop gets interesting!  We discussed the types of problems that Machine Learning solves, which can be broadly categorised into Regression (predicting a value), Classification (predicting a ‘yes’ or ‘no’) and Clustering (predicting a grouping of data) problems.  The easiest way to bring this to life is to use a series of examples, which we walked through with the group.

The Modelling stage of CRISP-DM can be further broken down into a number of activities, specifically around building, training, and evaluating a machine learning model.  We ran through the important concept of splitting data into training and testing, and then training and evaluating the strength of the model. 


By evaluating the models, we can introduce the participants to the multitude of machine learning algorithms that enable models to be built quickly and efficiently, and to provide the confidence that most business problems have patterns that have been systemised and programmed into these algorithms.

By using real life examples, we were thus able to explain how different types of business problems can be approached by building a machine learning model.       

What technologies support Machine Learning?

The final part of the workshop was discussing how to build machine learning models using Microsoft’s Machine Learning Studio.  The tool is rapidly gaining market acceptance due to its ease of use, and its ability to easily connect to many data sources and incorporate traditional open source data science tools such as R and Python.

In a short space of time we were able to visually show the steps required to turn data into a machine learning model, and with simple API Integration, use Microsoft Excel to input and successfully run the model. 

The advantage with Microsoft ML Studio is that it forces end users to think about the workflow required to connect, ingest, and transform data for a machine learning experiment.  It also has a great gallery of publicly available experiments to help educate business users around the use of the tool. 


In Summary

The workshop generated a lot of great discussion amongst the various participants, who were representing business, finance, IT and operations, and provided interesting insights into why business people want to understand more about the topic.  It also dispelled a lot of myths and enabled our participants to walk out of the session with an organisational view of the value that machine learning might bring to the Department in the future.