Audio version of the article
For organizations looking to capitalize on artificial intelligence and data analytics, the road forward can be daunting. From identifying use cases and data sources to developing models and sorting out the software and hardware stack, it’s easy to get lost in the complexities of a project.
At Dell Technologies, we understand these challenges firsthand — there’s an almost infinite number of ways to apply AI. We have come to recognize that like all organizations, we need to take a strategic approach to AI, one that allows the best practices identified on individual projects to be leveraged across the enterprise.
With that goal in mind, we have identified five fundamental tasks, or use cases, for implementing AI in the enterprise:
- Anomaly detection
- Natural language processing
Using this strategic framework, we look for similarities between different AI tasks and opportunities to use common approaches to address challenges in projects across the business. We’ve found this to be an effective way to evaluate the work effort required for AI projects.
Let’s take a look at each of these use-case tasks, along with some examples of AI projects that we’ve successfully implemented.
Anomaly detection is the process of using AI to identify outliers in a dataset. For example, organizations use anomaly detection to identify fraudulent transactions, spot malicious network behavior and recognize manufactured parts that are out of tolerance.
At Dell Technologies, we use AI-driven anomaly detection to fight warranty service fraud, which is a serious and costly threat to manufacturers. Our research found that as much as 10 percent of warranty reserves can be taken away by fraudulent claims. So, we looked at how we could get in front of fraud by predicting it with machine learning techniques.
We quickly determined that our machine learning models were much better at this task than the manual methods we used previously. The models identified key patterns of behavior that allowed our staff resources to focus more time on actual fraud cases instead of false positives.
The results were impressive. We went from a 4x ROI with people manually investigating questionable warranty claims to a 13x ROI with machine learning models. In a single month, we recovered $1 million that otherwise would have been lost to fraudulent claims.
Natural language processing
Natural language processing, or NLP, is a sub-field of AI focused on enabling computers to interpret and respond to language-based data.
We put NLP to work in an intelligent support tool that we developed to streamline the customer support experience. Our tool, which incorporates multiple machine learning models, gives our agents predictions for the best troubleshooting steps to suggest to customers who call in to report issues with products. The tool helps agents diagnose and solve problems quickly and accurately without having to navigate through a maze of web links, troubleshooting guides and decision trees.
Today, we have more than 3,000 agents using the tool, servicing more than 10,000 customers per day. And we’re seeing some great results. The tool has helped us achieve a 10 percent reduction in call times, along with improved customer satisfaction and a reduction in the numbers of customers who have to call back because they are still experiencing problems.
Recognition is the AI process of identifying what a given artifact is — for example, which person is in the photo, what type of car is in the video, and so on. Augmented reality, or AR, is one type of recognition application. With AR, your view in the physical world is enriched with overlaid graphical elements.
At Dell Technologies, we developed an AR Assistant to empower our customers to perform lifecycle management for Dell EMC hardware. This smart phone app displays simple animations over the top of the view of physical hardware to help users see how to perform certain hardware repair and upgrade functions, as if they had an assistant explaining each step. This makes it easy for users to see which components they should interact with and what tools they should use ― from any distance and angle — as they perform hardware servicing procedures.
For a view of how the AR Assistant works, check out this video demo. You’ll see firsthand why we are so excited about this AI-driven app.
Recommendation, which includes concepts such as “next best action,” encompasses a class of techniques and algorithms that are able to suggest what action to take in the future to provide an optimal result.
One way to use recommendation is to optimize IT operations with predictive analytics. We’re doing just that at Dell Technologies. To improve IT operations efficiency, our data scientists developed a propensity model to score the potential of various servers going down within our IT operations. This model proactively identifies specific servers that might be at risk and provides our IT operations team with advanced warning and the ability to respond proactively to potential outages.
The results have been great. The highly accurate model predicts and alerts on abnormal server behavior 40 minutes before system failure, and it has led to estimated operational cost savings of $25 million.
We would like to see our customer realize the same sorts of gains. With that goal in mind, we worked this modeling effort in combination with our ProSupport team so we could provide better estimates about abnormal server behavior to our customers.
Segmentation is the AI process of grouping together items with similar characteristics, typically for additional analysis or descriptive analytics.
We’re using segmentation for all kinds of processes at Dell Technologies. One of those is inventory management for parts repair.
We have 760 distribution centers globally, and because we offer two- and four-hour Service Level Agreements (SLAs) to many customers on Same Day Business Services, each center has to have the right part in the right location. This means that we need to inventory on hand in cities all around the world. Using segmentation and pooling modeling — by cities, regions, and inventory part — we were able to avoid costly overstocking of parts and greatly reduce our overall inventory requirements.
Here again, the benefits were huge. This initiative saved $18 million in additional inventory costs, all without impacting customer service levels. Our data science team is now working on continuing improvements to the model, and we plan to roll this model out to other regions and countries.
A team effort
The examples highlighted here are just a small sample of the many internal successes that we have had with AI implementations across the Dell Technologies enterprise.
So how did we get here? To build our AI solutions, we draw on the experiences and expertise from three sources: our internal data science team, our consulting services practice that advises our customers on their AI and data science projects, and our Dell Technologies HPC & AI Innovation Lab, which is our research center in Austin.
With the combined expertise from these groups, we have been able to think more strategically about how we can best capitalize on AI at Dell Technologies. And one way to do that is to start all AI initiatives with clearly defined use-case tasks, and then build your solution from that solid foundation.
This article has been published from a wire agency feed without modifications to the text. Only the headline has been changed.