We’ve all read the reports about artificial intelligence and its impact helping large organizations improve customer experiences, bolster security and tighten compliance. As business leaders consider AI remember putting cognitive computing to work is a process. Don’t jump ahead or take short cuts. Before you dive in, make sure you understand where AI produces the best outcomes through this five-stage journey.

A 2018 Harvard Business Review survey of 250 executives who are familiar with how their companies use cognitive technology shows that three-quarters believe artificial intelligence or AI will substantially transform their companies within three years.

Alas, there’s no silver bullet for AI at the enterprise level. Putting cognitive computing to work in a large organization is a process that requires a strategic and methodical approach. AI requires massive datasets and accurate modeling to produce useful predictions. Early missteps can cast a shadow over everything that follows. Ultimately, the goal is to develop a human-centered approach that removes friction and boosts efficiency, creating a brighter future for the organization and everybody who depends on it.

Before you dive in, make sure you understand where AI can produce the best outcomes. Be especially alert for opportunities to personalize and streamline the customer/user experience. And give your project room for trial-and-error processes that do the most good with the least disruption to your business. Here are five stages to take in your journey.


AI requires you to integrate credible information from multiple sources. The algorithms use statistical modeling to comb numerous datasets for business intelligence at speeds far beyond human capability. The data needs to be organized, cleaned, prepared and classified. It’s mandatory to create well-thought-out controls for challenges including security, personal information and compliance. IT departments have to create data warehouses or data lakes to store, share and manage all of this information. Technologies such as APIs and microservices will connect all of the data sources and create a data pipeline pumping information into your AI engine. Once you’ve integrated all your data sources, it’s time to develop reporting processes that demonstrate the impact of your project.


Data integration enables the creation of dashboards that provide single-point-of-truth insights on the well-being of an organization. Learning algorithms turbocharge the reporting process because business users can crunch real-time data to get a better idea of what’s on the horizon. Moving reporting power into the hands of business users marks a fundamental shift. In a traditional structure, the IT department handled much more of the data-analysis process. Today, the power of reporting goes directly to the people who need it most. Your IT teams still have plenty of work to do, because the network and storage requirements are pretty high to do AI because it requires access to core operation data without sacrificing security. Some cloud providers actually have an AI-based agent that can warn customers about personal information stored in the cloud without proper data protections.


In the data-exploration phase, data scientists have three fundamental responsibilities: Understanding the data, identifying the optimal algorithms and combining these insights to enable the development of predictive models. About three-quarters of a data scientist’s job involves exploring all the variables and inputs that go into an AI model, a process called feature engineering. Companies often hire AI expertise at the data exploration phase.


The potential for AI to create actionable insights becomes crystal clear in the predictive modeling phase. Here, data scientists run the information from the first three phases through learning algorithms that predict problems early enough to avert troublesome outcomes. For instance, a predictive model could identify when a car’s battery is about to die. An algorithm then informs the owner it’s time to change the battery and sets up an appointment at a local dealer, averting the aggravation and expense of a dead battery. Predictive algorithms also provide nuanced insights on consumer sentiment. Picture a mortgage provider getting calls about interest rates from current customers. While conventional consumer surveys might find people are perfectly happy with their current mortgages, interest rate inquiries could signal people planning to refinance. A predictive algorithm can interpret this signal and help the company reduce customer churn.


Prescriptive modeling represents the future of AI. It means using cognitive computing to chart the future of your organization. In the prescriptive phase, companies need to analyze the performance of their predictive model and make sure they still perform well. You also need to be sure you understand the rationale of your algorithms’ design. The ability to explain machine learning interpretability and modeling are key features that you will need if you are in a highly regulated industry or are required to provide rules for model behavior. It’s crucial to be able to explain why the algorithm produces a specific outcome to benefit the consumer.

Niraj Patel is senior vice president of artificial intelligence for DMI.

Story Continues Below