There is tremendous interest in artificial intelligence this year, which is ironic since the technology is more than 60 years old. In fact, artificial intelligence actually encompasses a few different technologies, each of which can help an organization better understand and engage with customers.
Forrester Research Senior Analyst Brandon Purcell has authored two reports on the current strong adoption of artificial intelligence. In part one of Information Management’s interviews with Purcell, we discuss “The Top Emerging Technologies in Artificial Intelligence.” Part two will discuss the report “Artificial Intelligence Technologies and Solutions, Q1 2017.”
Information Management: Artificial intelligence is all the rage in 2017. Why is there so much interest in AI currently?
Brandon Purcell: Since the term “artificial intelligence” was coined at Dartmouth College in 1956, the field of study has gone through several periods of exuberance and disillusionment. The previous period of disillusionment, commonly called an “AI winter” lasted several decades and was characterized by a lack of investment and research in the fields.
This changed in the last few years due to several factors. First, data is the lifeblood of AI systems, and we are currently producing and capturing more data than at any other time in history. Second, while machine learning and even artificial neural networks have been around for decades, it wasn’t until a few years ago that we discovered deep learning’s uncanny ability to classify unstructured data such as images, video, speech, and text. Still, it took ages to train deep learning models with traditional CPUs but the advent of GPUs accelerated this process from days to hours.
Most importantly, there is a clear need for AI. Increasingly empowered customers are demanding personalized experiences, and the only way to meet their expectations at scale is through artificial intelligence. AI provides the ability to understand customers and anticipate their needs, then deliver optimized experiences across channels and touch points. Companies that don’t embrace AI are likely to fall behind.
IM: You describe four specific technology areas that will help customer insight professionals “sense, think and act”. How do these technologies related to competitive advantage, and which organizations or industries are they most appropriate for?
Purcell: Organizations interested in adopting AI must first identify the use case and the overarching business objective, and then determine whether they have the requisite data on hand. For example, speech analytics is quite useful for companies with large volume of customer phone calls who would like to understand what drives customers to that channel and improve the customer experience. Retail banks, travel and hospitality companies, and insurers all benefit greatly from the insights speech analytics uncovers.
Deep learning is especially useful for categorizing unstructured data, so companies with a large collection on image, video, speech, or text data may be interested in investing here. The most important thing to remember when considering a deep learning platform is that most of them required you to “bring your own training data.” This means that they provide access to the deep neural network, but you’ll need well-labeled training data to feed into the algorithm. This is a non-starter for many companies.
Finally, natural language generation will be most useful for organizations interested in developing their own customer- or employee-facing apps that converse with users.
IM: Deep learning is an area that more companies are starting to explore. What exactly is deep learning, and how can an organization best use the insights gained from deep learning to make better business decisions?
Purcell: Deep learning is a branch of machine learning that leverages artificial neural networks. Today, deep learning is mostly used to analyze or classify unstructured data sources such as text, speech, or image / video. Organizations who don’t want to start from scratch can look to vendors who have already trained neural networks for a specific use case. For example, Caffe and Clarifai offer access to pre-trained image recognition algorithms. Through a process called transfer learning, organizations can adapt these pre-built models and tailor them to their own needs using their own proprietary data.
IM: You discuss AI-enhanced analytics solutions in the report. How does this differ from traditional analytics, and what does an organization have to do/have in place to achieve best results from AI-enhanced analytics?
Purcell: Today, AI-enhanced analytics solutions come in two flavors. Some offer a conversational interface that combines NLP and NLG so users can query data using natural language. Many also explain analyses in natural language. Others learn from a user’s behavior the types of KPIs that user is interested in and will automatically track those KPIs and surface insights. Some AI-enhanced analytics solutions offer all of the above.
Existing BI vendors such as Tableau and TIBCO Spotfire are entering this arena by partnering with Automated Insights to add AI enhancement to their existing platforms. Other vendors such as IBM Watson Analytics and Salesforce BeyondCore have these capabilities built in. The good news for enterprises interested in these solutions is that they don’t require any special preparation - just expect the typical integration challenges you’d encounter with any BI integration.