Businesses are looking to predictive analytics that promise to increase sales, reduce costs, prevent fraud and make operations more efficient.
Yet most organizations are falling short of achieving the desired results: A study by MIT Sloan Management Review and Boston Consulting Group found that only 10% of companies saw significant financial benefits from their AI investments, and a Rexer Analytics survey found that just 22% of data scientists said new initiatives are typically deployed and operationalized across the enterprise.
Eric Siegel, a consultant and former professor at Columbia University and the University of Virginia, says many predictive machine learning projects fail because they focus too much on the technology rather than advancing it as a strategic business project.
In his new book, “AI Playbook: Mastering the Rare Art of Machine Learning Adoption,” Siegel argues that organizations are unable to realize the value of AI because they lack an effective business paradigm for executing machine learning projects. Most machine learning projects are highly technical and therefore often the domain of experienced data science professionals. This creates a disconnect between the data professionals who prepare the data and develop and operate the AI models, and the business stakeholders who are responsible for running the operations at scale that will benefit from predictive insights.
“Putting too much emphasis on the modeling science and not enough emphasis on the deployment is like being more excited about rocket science than the actual launch of a rocket,” Siegel said during a recent MIT Sloan Management Review webinar. “That's where we are today.”
To drive success, Siegel said companies need a standardized playbook for machine learning projects that business experts can access and participate in the lifecycle of predictive analytics projects.
Otherwise, “each side points to the other and says, 'Executing and managing this business-level process is not my job,'” he explains. “It remains a no-man's land, and that's the final piece before you can achieve larger-scale success and deployment.”
6 steps to kickstart your machine learning project
To bridge this gap, Siegel proposes what he calls “BizML,” a set of business practices for running predictive machine learning projects.
He outlined six steps to foster collaboration between business and technical stakeholders throughout all stages of machine learning adoption.
Establish deployment goals. To get real value from machine learning, you need a clear value proposition that details how the technology will impact your business. Data scientists can't do this alone. It's important that business stakeholders who know the problems and opportunities are familiar enough with the technology to be involved in setting realistic goals.
Establish forecast goals. Modeling and prediction involves complex mathematics, but it must be done with the business goal in mind. Business users need to have a semi-technical understanding of the technology and share their specific domain knowledge to define what the machine learning model is trying to predict in each use case.
Establish the right metrics. Determine what key benchmarks you will track during both model training and deployment. Additionally, identify what performance levels your machine learning project must achieve for it to be considered successful. Typically, most machine learning projects are based on technical metrics such as precision, recall, and accuracy. Organizations need to shift their focus to business metrics such as profits, ROI, savings, and customer acquisition, Siegel said.
Prepare your data. Defining what your training data looks like and making sure it's in the format you want it to be is a critical step, and one that Siegel says is non-negotiable, as it's key to getting valuable results.
Train the model. The prepared data is then used to train and generate predictive models, where data experts take the lead, but there is always room for additional input from the business.
Related article
Deploy the model. The model is used to render a predictive score, which is then used to improve business operations. It is also important to maintain the model through ongoing monitoring and regular updates.
The last three steps are more technical than the first three, but they all require close collaboration between technology and business stakeholders. Building a bridge between the two requires investment and commitment to good change management practices to ensure machine learning is well understood among stakeholders across the business.
“While change management challenges in general are not new, the need to manage change well is often overlooked when it comes to machine learning projects,” Siegel said. “Machine learning launches the rocket, but a responsible person still has to orchestrate the launch.”
Watch the Webinar: How to Succeed with Predictive AI
