Across many industries, innovators are looking at machine learning as a solution for bringing their organization to the next level of productivity. Retail is no different. Many organizations have already adopted machine learning and artificial intelligence solutions and are reaping the rewards. In my experience working with retail companies, those that have implemented AI solutions have an overall greater awareness of business practices and have been more successful at making important and difficult decisions effectively.
To operate successfully, retailers encounter and must make thousands of decisions every day, from creating a new product to its placement in a store and, ultimately, to protection of revenue through asset monitoring. There’s a simple way to make sense of retail and consumer goods companies’ data and enjoy the best business rewards, and it starts with understanding AI solutions.
• Pragmatic AI: This is a branch of artificial intelligence that uses cognitive search and machine learning capabilities. It can be used to help retailers make more effective business decisions. These decisions can include aspects of product design, shipping efficiencies, product placement and how to reduce unsellable item.
• Prescriptive Analytics: This tool identifies opportunities for improved operations and shares an easy-to-understand, prescribed plan of action and resolution in plain language. The prescribed task can be assigned to an employee and monitored to ensure it is finished. It could also link to videos, forms or other helpful resources.
Prescriptive analytics is the part of AI that makes it truly pragmatic: It’s a computer-driven operation that takes raw data and automatically identifies opportunities to improve processes and customer service, increase revenue and strengthen margins. For example, a computer might identify that a bottle of soda hasn’t been sold all day in a convenience store, despite having enough “on hand” (an anomaly in the data). It then will “prescribe” a corrective action for the store clerk, directing them to go check the shelf and make sure the sodas are restocked, either from their back-room inventory or expedited shipment.
A good analytics solution analyzes data in near-real time (as fast as the data can be passed along) to clearly define a course of action. In the case of the soda sale, prescriptive analytics can identify the lost sales quickly. Instead of the store going all day without soda sales, the shelves are stocked, and sales return to normal. Here is a real-world case that demonstrates how you can use pragmatic AI in retail.
Catch Vendor Issues
A grocery customer used prescriptive analytics to identify and solve a vendor quality issue. Not only did the data show that large eggs were being delivered broken more often than usual; it showed that the trend was happening in stores in a single region and with a single brand of eggs. Based on that data, the grocer contacted the egg vendor and determined that the vendor had suffered a fire at the plant that produced large egg cartons. To avoid a gap in supply, they had started shipping the large eggs in cartons designed for medium eggs — causing additional friction that led to the breaks. Without prescriptive analytics identifying the trend and tracing the root cause to the vendor, the grocer would never have course-corrected with its supplier and would have lost significantly more money. Instead, they were able to secure a credit of more than $25,000.
How To Implement Pragmatic AI In Your Company
1. Overcome Misconceptions
A common myth I encounter when talking to retailers or colleagues in other fields is that AI is a self-running monster that will quickly absorb the jobs of human employees. In reality, pragmatic AI is enabled by machine learning, and it requires human intervention to deal with the errors or business improvements it identifies. Humans are left to use their creativity and critical thinking skills to consider the machine’s findings and improve them with a feedback loop. For retail, this might play out as simply a machine learning solution that identifies that a store is out of soda. The human receives an alert from the system, goes to the shelf and either restocks it, or finds that the machine was wrong and overrides the analysis, training the machine to know what to look for in the future.
2. Start With The Data You Already Have
Data doesn’t have to be pretty. Many executives think that in order to have a successful pragmatic AI implementation, they need to sort and clean their data. In reality, raw data (structured and unstructured), as much as you can provide, is best to inject into a machine learning-enabled pragmatic AI solution. Even if what you insert is unintelligible to you, the system will learn over time, realizing what works and fixing what doesn’t.
3. Consider Additional Costs
Before you pick a solution, ask yourself a few questions. How quickly will the system be live and creating a measurable impact on my business? How much will I have to rely on my IT organization or other departments? How much will I be charged for data validation, training and additional users? You can start by loading the data you need for the first phase of implementation, then stacking the additional types of data and measures for later. While there was once a significant amount of mony, time and productivity lost during a typical software implementation, the right type of pragmatic AI solution should be flexible and easy to integration across an enterprise. IT involvement, unless they are leading the project, should be minimal. There should be no upcharges or additional fees for data validation, training or more users. Everything should be included as one fixed cost so that as many people as possible are able to jump on the system and see value from it.
The right type of pragmatic AI solution is not just a software-as-a-service, but a software and a service. Executives should feel supported by their AI solution and its provider throughout the engagement.