Uncovering insights with IBM
Max Kelsen chose an IBM Watson Discovery solution to build an insight engine for powering its new offering, the cognitiveCX platform. The accuracy of the new business knowledge revealed using the Watson toolset, which combines cloud-based cognitive search, machine learning, relevancy training, NLP and query capabilities, propelled its decision. “We tested a number of commercial and open source solutions, and Watson consistently outperformed them,” says Samuel Irvine Casey, Chief Operating Officer (COO) and Cofounder at Max Kelsen.
Analysts’ ability to quickly build and train highly accurate customized models using IBM Watson Knowledge Studio , which does not require coding, also stood out. “No solutions on the market allowed us to train a custom model for any domain as easily or effectively as WKS [Watson Knowledge Studio],” notes Irvine Casey.
Therkelsen-Terry agrees. “For us, that was game-changing. It brought a lot of what we were thinking alive and allowed us to execute the way that we wanted to. That was really the moment for us, a light bulb moment, when we first got onto WKS and said, Wow, this is insane.”
Irvine Casey and his team in the Insights and Analytics department collaborated with the government entity and Sherlok, the human analytics services division of the marketing research firm, to build and test the Watson insight engine on the IBM Cloud™ platform. The government organization, which like many businesses receives feedback from multiple channels and departments, presented an ideal use case for identifying the real pain points preventing delivery of excellent experiences.
Max Kelsen began its extensive analysis using Watson Knowledge Studio, a cloud-based solution that facilitates annotation of custom, industry-specific components in unstructured data. Government staff provided approximately two million source documents collected from hundreds of citizen touchpoints dating back three years and from more than 40 distinct service areas. The Max Kelsen team randomly selected 5,000 of the documents for annotation. In preparation for working with Watson Knowledge Studio, it mapped out relevant entity types, including service departments; web, social and call center channels; and citizens’ emails and letters to officials. Then, without coding, the team trained a machine learning model using approximately 4,000 of the annotated documents, which provided the equivalent of the 300,000 words of content IBM recommends for analysis.
Within six weeks of receiving the source documents, Max Kelsen sent the government officials a preliminary report of the analysis results. During that period, the team became acquainted with the application’s user-friendly features for training rule-based and machine-based models.
“When we first started,” says Therkelsen-Terry, “we didn’t grasp the power of some features—like preannotation tools, dictionaries and a visual rule editor—that we now use instinctively to really speed up the training process. Done manually, using 300,000 words, this would be an incredibly arduous task.” For instance, the team relied heavily on the application’s human annotator feature to help ensure the model accurately and consistently represented organizational and industry terminology.
Having completed the custom annotation model, the team generated qualitative insights using IBM Watson Natural Language Understanding (US). Now embedded with IBM Watson Discovery, the API interrogates unstructured text on an entity-by-entity basis for tone, intent, sentiment, service area and contextual details. The team analyzed the results using the Watson Discovery search engine, which provides core AI capabilities for quickly and affordably uploading, enriching, indexing and querying large collections of private and public data.
“Our analysts can quickly search through millions or billions of records in an instant to get individual outputs out of what customers are saying about certain issues and why they’re frustrated,” elaborates Therkelsen-Terry. “We can really get a handle on how customers are feeling, what they’re talking about and the actual words they use.”
Max Kelsen also relied on leading data visualization tools to present the insights in meaningful, engaging ways.