AI chatbot usage and concept
It is time for all business leaders to understand how predictive artificial intelligence (AI) and machine learning (ML) can help grow their business and how they can be implemented and deployed to get maximum value with minimal risk. Our experience in both large enterprises and startups shows that most of the current efforts either fail to deploy or fail to deliver the expected value.
We recommend treating this new technology like any other, with careful planning, a detailed adoption process, and controls in place to monitor results and requirements for change.To help, specialized business practices such as bizML, outlined in the new book “The AI Playbook” by leading consultant and former Columbia University professor Eric Siegel, have also emerged.
Here are Siegel's six key steps to success as you first begin using this technology, whether it's artificial intelligence, machine learning or predictive analytics.
1. Quantify a positive business value proposition. Start by documenting the business improvement you're targeting, such as increased revenue from improved ad response rates. Avoid a technology-first or solutions-first mindset that focuses on technology instead of business outcomes. Use this value to gain approval to move forward with implementation.
2. Establish a machine learning prediction goal. You need to establish in detail what is expected from the deployment and what will be done for each prediction. business and TechnologyTranslating business intent into a well-defined technical model requires collaboration between business leaders and technologists.
3. Defines a specific model evaluation metric. What you're looking for here is an accuracy measure of how well the model can predict, or at least how well it can predict better than guessing or without learning. Additional factors are the cost of a correct prediction, the cost of a false positive or false negative, and the likelihood of learning over time.
Four. Prepare the data source for training. Remember, the right data always trumps the best machine learning algorithms. Data must be collected and restructured into relevant elements for training and deploying models. Learning should include both positive and negative cases, noise and supporting elements.
Five. Generate and train a predictive model. This is where we develop the most powerful predictive technology, including a training element where the computer essentially reprograms itself. We evaluate available predictive analytics algorithms such as decision trees, regression analysis, etc., by learning from custom-built or purchased AI models.
6. Deploy and evaluate machine learning models. Deployment means introducing change into operations. Turning predictions into actions requires buy-in and cooperation from all levels of the team. We recommend setting up control groups to mitigate risks, handle metrics, and make necessary adjustments to data and models.
In reality, these steps are just the beginning. Once a model has proven its value, it requires maintenance, oversight, and ethical vigilance to continue into the future. New technologies tend to lose their advantage and become stagnant over time as the world around us changes. Economies shift, and customer behavior patterns evolve.
They must also be sensitive to changes that could cause learning models to adversely affect protected classes, show bias or lack of representativeness toward certain groups, or reveal personal attributes that do not need to be disclosed.
Your role as a business leader and expert therefore becomes even more crucial to ensure that the results benefit not only your company, but your customers and society at large. Rather than acting blindly or ignoring new business growth opportunities, it is time for all of us to learn more about how to adopt new technologies to move forward effectively.
