In honor of Geek Pride Day on May 25, today we’re gonna get our geek on as we dig a little deeper into Nuance Pathfinder, an upcoming product that was introduced in our previous post entitled “Designing effortless customer journeys.”
Building conversational AI systems—like virtual assistants, IVRs, chatbots—generally includes three main technologies:
- Natural Language Understanding (NLU)
- Question Answering
- Dialog Systems
AI and, more specifically, machine learning and deep learning have had the biggest impact on NLU, which is implemented using classification and entity extraction techniques. In the research world, question answering is also being impacted by machine learning (for example, see SQuAD), but this technology is still making its way into conversational AI applications. But dialog, wherein a system will engage in a back-and-forth conversation with a user over multiple turns, is still very much driven by conversation and voice UI designers.
Today, the most labor-intensive steps to creating these conversational systems are still:
- Training a model: Training a natural language understanding (NLU) model to recognize a user’s intent, which requires gathering and manually labeling reams of data to train the model
- Writing a dialogue: Many virtual assistants rely on a script; their AI simply functions by choosing which branch of the script to follow. Writing this script requires specially trained conversation designers to work extensively with subject matter experts in understanding how a business functions and what questions an agent must ask to solve its customers’ problems.
Nuance Pathfinder makes both steps easier. This diagram explains Pathfinder at a high-level:
First, we start with a company’s conversational data. This conversational data consists of conversations between customers and (human) customer service agents, and it can come from both live chat and phone calls transcribed manually or using speech recognition like Nuance Transcription Engine.
Next, we feed this data through Pathfinder’s proprietary intent discovery algorithms, which analyze the data and automatically identify users’ intents and common topics in the conversations. The process then groups conversations together according to these intents and topics, saving much of the manual labeling effort.
Then we move beyond the level of topic and intent modeling, to analyze the structure of the conversations themselves. By analyzing the turns in the conversations, or the individual utterances between a customer and an agent, we can identify the states in a conversation and the connections between them. By collecting these states and connections, Pathfinder builds a visual representation of all the different paths your customer service conversations take, from first question to each follow-up, to reveal the best paths to resolution as well as unknown problem areas. We call this a dialog graph or model.
With insights and dialog graphs in hand, dialogue designers can rapidly produce more sophisticated, natural and accurate scripts. With most customer service interactions being “monitored or recorded for quality assurance and training purposes” or compliance reasons, and ever greater adoption of live chat options, companies are amassing this kind of data at a blistering pace. However, few companies are truly able to leverage this data using the modern, data-driven techniques now possible with AI and machine learning.
Now, customer insights long hidden within terabytes of this conversational data can be surfaced to drive new opportunities for building stronger relationships with customers and reducing cost of service—thanks to Nuance’s army of geeks who get their rush from delving into the science of conversational design.
So here’s to the Geek! Without them, you’d be lost under a mountain of data and your customers would be following paths to a dead end.
Interested in a demo? Click here to engage us – select “Analytics” in the drop down menu and in the comments section type “I’d like a Project Pathfinder demo” and a member of our team will get in touch with you shortly.