This article is based on a presentation at the Spring 2024 Analytics Conference on Decoding Consumer Analytics. by Sergei Ananyan, CEO of Megaputer Intelligence.
From the moment a customer walks into a store, there are dozens of invisible processes taking place behind the scenes to make the experience possible: things like inventory management, workforce management and support centers, equipment management for point-of-sale transactions, online shopping, loyalty cards and promotions, and survey feedback.
“Companies initially collected this data for operational purposes, so customers could return merchandise, for example,” explains Sergei Ananyan, CEO for Megaputer Intelligence. “Then, they discovered that if you look hard enough at the data, you can discover wonderful things that will help you to be more efficient and outpace your competition.”
As they say in Harry Potter, Ananyan jokes, “magic always leaves traces.”
Ananyan is describing retail analytics, which provide insights related to sales, inventory, customers, and other aspects of the merchant’s decision-making process. Retail analytics help businesses make better choices, run more efficiently, and deliver improved customer service through techniques including data mining, text analytics, AI, and business intelligence.
Companies like Megaputer Intelligence generate tools for retailers to mine and interpret this magical data and gain a competitive advantage. Megaputer Intelligence uses a platform called PolyAnalyst to cover all steps of a typical data analysis process, such as data loading, performing manipulations with data, analysis of structured data, machine learning, and AI, then generates dashboards for their customers that help them understand and apply the data that has been collected.
Text Analysis in Action: Mining Taco Bell’s Voice of Customer Data
Taco Bell created a voice of customer website that received around 2.5 million responses over a two-year period. They wanted to leverage AI to help analyze the sentiment of the responses and improve their customer experience.
The voice of customer website included data fields that allowed customers to rate their overall satisfaction, indicate whether they would recommend Taco Bell to a friend, and determine service metrics associated with a quick service experience. Taco Bell also gave customers an opportunity to send responses in an open comment field, which was used as the focus of the AI analysis.
Taco Bell worked with Megaputer Intelligence to build and train algorithms for a taxonomy, or broad topic categories that aligned with key performance indicators (KPIs) measured in the structured data. Categories included service, staff, product, facilities, and suggestions.
The data was also passed through a sentiment engine and scored as positive or negative so Taco Bell could see if a customer was satisfied or not. Taco Bell was particularly concerned with how to improve the customer experience by looking at unsatisfied customers’ responses. While the majority of sentiments were positive, the analysis revealed that customers were frustrated mainly by rude staff, slow service, or not-fresh products.
“The outputs gave management clear priorities of what to focus on improving, and they were able to see how different elements affect customer satisfaction,” Ananyan explains.
How it Works: The “Magic” Behind Text Analysis
Text mining typically involves three types of analysis:
- Linguistic: When people talk, Ananyan explains, there are many dimensions to how we communicate meaning to another person, and it takes time to teach this to computers. These dimensions include terms, noun and verb phrases, and sentence structure.
- Semantic: Texts also have varying meanings that computers also have to learn. Humans can use different words and still understand each other, Ananyan says, because there are different semantic relations that we are taught.
- Statistical: Text mining includes checking the frequency of different words and seeing if they are used in a particular domain, or if there are deviations that reveal interesting patterns.
When you complete text analysis frequently, Ananyan says, it can be hard to know where to start. He suggests two steps:
- Unsupervised analysis: Unsupervised analysis involves teasing information out from the data. First you detect the most important words and phrases and the relationships between them, then elicit data-driven hierarchical structure for document classification.
- Supervised analysis: Next, you reorganize the data depending on what’s important to your business by building taxonomy queries for accurate document classification. These queries are based on linguistic, semantic, and statistical information.
Uncovering Insights from Product Reviews: Analysis for a Printer Company
A company that sells printers recruited Megaputer Intelligence to use text analytics to complete a product review analysis. AI was used to analyze thousands of product reviews from customers who commented on four different brands’ printers, which saved the first company time and resources while allowing them to get a thorough picture of the sentiment of the reviews.
The technology was able to compare customer reviews for four popular printers in the same price segment to determine what product features are important to customers, and what brand or model fairs best with respect to those features.
Megaputer Intelligence conducted a multi-step data analysis for the printer company. They marked each review to identify what customers were talking about, and whether they were satisfied with the product or not, through sentiment analysis and effect extraction. Then, they generated a simple report to compare the printers using different categories, including setup, selling price, and wireless connectivity.
Retailers can use this information to decide which brands they want to carry out of the four printers, and manufacturers can look at the outputs to learn what customers prioritize when they’re buying a printer.
“In this kind of analysis, you’re trying to look outside of your company to listen to the entire world, which wasn’t possible before,” Ananyan says. “If retailers and manufacturers are smart enough, they can listen. The data is out there.”
Leave a Reply