Artificial Intelligence (AI) and machine learning have generated a lot of discussion and a lot of ambiguity. With these new techniques, we can transform the ways we do many things in our everyday lives: how we drive our cars, manage our money and make smart decisions. If we asked you to define them and how they work together, however, could you?
Simply put, artificial intelligence covers a broad range of evolving technologies that combine statistical methods, data and modern computing power to transform human potential. Machine Learning (ML) uses math (algorithms) to identify patterns in large data sets and recommend actions. The differentiator with ML models is that over time, the algorithms ‘learn’ from their performance and refine how they execute over time to improve. For example, an ML model can flag potential fraud in financial data based on patterns that are expected and unexpected. The model refines how patterns are detected using new data which brings new scenarios. Machines are very good at reviewing massive amounts of data and finding these patterns and exceptions and can do so at a far larger scale and speed than humans.
All Machine Learning is AI, but not all AI is Machine Learning
When you hear someone discussing machine learning, they are speaking about artificial intelligence. As one of the core branches of AI, machine learning has brought us some of the most noteworthy advances – Siri on Apple’s iPhone, Amazon Alexa and even the engine behind your Facebook newsfeed. AI also contains many other fields such as computer vision and robotics, but we won’t be focusing on those.
Neural Networks are a class of models within machine learning that have revolutionized the field. Neural networks are a specific set of algorithms that are inspired by biological neural networks and were originally developed to mimic the computational structure of the brain. Recently, however, they have moved away from their biological inspirations towards new, mathematically inspired frontiers.
Within the study of artificial neural networks, there is a specialty area that you may have heard called deep learning. Deep learning is an advanced subset of machine learning and uses a hierarchical level of artificial neural networks to carry out the process of machine learning. These networks have several layers of computation and are often constructed for tasks such as satellite image recognition, medical diagnosis and speech recognition.
Machine Learning in Action
Now that you know what machine learning is, the next question is how to use it. Here are two examples of solutions we’ve delivered using machine learning.
Intelligent Assistant Reduces Burden on Overloaded Call Center.
Our client wanted an intelligent assistant that could predict and use the right customer service templates to respond to specific, common questions in order to reduce call volume for these predictable requests. This goes beyond the basic experience of common online chatbots by tailoring the response to the person’s specific situation. We first conducted research with the Product Owner to understand how customer service agents work. While there were officially 400 defined templates, we discovered customer service agents adjust the templates by adding or removing information. Instead of 400 templates, there were thousands. For an effective experience, we mapped all the different answers to the 400 official templates so that we could classify the data. Then, we developed a reusable machine learning framework that starts with simple methods, sets benchmarks and increases in complexity as necessary. Through this process, the team compared the relative efficacy of several models (neural networks, ngram SVM, catboost, and xgboost models,) allowing us to improve accuracy by 30% from our initial model.
Advanced Models and Open-Source Data Combine to Increase Nationwide Donations.
When the ALS Association needed to turn data into action, they called on Excella’s experts to help them build tools to improve donor events and drive more dollars back to the cause. The ALSA realized they needed to explore donor overlap and repeat giving across the Association. This data-driven approach would enable a more donor-centric, efficient approach to engage their constituents. The team’s first attempts at data analysis projects were time-consuming and produced limited results. Excella’s data scientists jumped right in, working with disparate data sources like event location and timing to look for patterns in giving across chapters. The Excella team delivered two solutions to help the ALSA team better leverage their data. The first solution layered publicly available data with historical donation patterns to develop an Event Recommendation System in order to help managers decide when and where to hold donor events. The system is powered by advanced machine learning models with an interactive front-end, making it easy for managers to plan for events. The second solution provided insights into the organization’s own data by displaying historical patterns and trends using Tableau. These solutions enable a more data-driven approach for the organization to increase efficiencies and help the team do what they do best – make the world a little better for those with ALS.