It’s time for all business leaders to understand how predictive artificial intelligence (AI) and machine learning (ML) can help grow your business, and how it can be implemented and deployed with minimal risk and maximum value returned. In my experience in large companies as well as startups, most current efforts have failed to deploy or never realize the value expected.
I recommend that you treat this new technology as you do any other, with careful planning and a detailed deployment process, with controls in place to monitor results and change requirements. Specialized business practices are also springing up to help, like bizML, outlined in a new book, “The AI Playbook,” by Eric Siegel, a leading consultant and former Columbia University professor.
Here is my summary of Siegel’s six key steps to a successful deployment of your first foray into the use of this technology, whether it be called artificial intelligence, machine learning, or predictive analysis:
1. Quantify a positive business value proposition. First of all, document the business improvements targeted, such as revenue growth through increased ad response rates. Avoid any technology-first or solution-first thinking, where the focus is on technology rather than business results. Use this value to get approval to proceed to deployment.
2. Establish a machine learning prediction goal. In great detail, you must establish what will be predicted by your deployment, and what will be done about each prediction. This is the intersection between biz and tech, requiring a collaboration between business leaders and technologists to turn business intentions into a well-defined technical model.
3. Define specific model evaluation metrics. What you are looking for here are accuracy measurements for how well the model predicts correctly, or at least predicts better than guessing or no learning. Additional elements are the cost of a correct prediction, cost of a false positive or a false negative, and the improvement in learning potential over time.
4. Prepare the data sources for learning. Remember that the right data as fuel always trumps the best machine learning algorithms. Data has to be collected and reconfigured into relevant elements for model training, as well as deployment. For learning, it must contain both positive and negative cases, as well as noise and supportive elements.
5. Generate and train the predictive model. Here is where you develop the most powerful predictive technology, including the training element, where the computer is essentially reprogramming itself. Evaluate predictive analysis algorithms available, including decision trees, regression analysis, with learning from custom-built or purchased AI models.
6. Deploy and evaluate the machine learning model. Deployment means introducing the change to your operation, which requires buy-in and cooperation from the team at every level.to translate predictions into actions. Setting up a control group is recommended to mitigate risk, process the metrics, and make necessary adjustments to data and model.
In reality, these steps are only the beginning. Once the model has proven to be of value, its viability moving forward requires maintenance, monitoring, and ethical vigilance. Any new technology tends to lose its edge and stagnate over time as the world changes around us. The economy shifts and customer behavior patterns evolve.
You must also be sensitive to changes that can cause the learning model to adversely affect any protected class, show bias and lack of representation to any specific groups, or reveal personal attributes that need not ever be disclosed.
Thus your role as a business leader and professional becomes even more key in making sure that the result helps not only your business, but also your customers and society as a whole. It’s time for all of us to learn more about how to deploy new technologies and move forward effectively rather than striking out blindly or ignoring new business growth opportunities.
