Recently I backed into a hotel parking spot after returning from a customer dinner. It was dark and rainy, and I was tired from traveling. My mind wandered until I heard a shrill “BEEP BEEP BEEP” coming from my rental car. I looked down at the dashboard’s rear-view camera, and saw that I was on a collision course with a light pole. I stepped on the brake and eased the car into place. “No bumper damage tonight,” I thought. I have to admit, I felt like a smaller, luckier version of Edward Smith, captain of The Titanic.
My fatigue-induced parking near mishap makes me think of any analysis involving text data:
• I benefited from a camera that saw something I missed.
• I found something that was there all along, but I didn't initially see it.
Statistician and author Nate Silver said: "Every day, three times per second, we produce the equivalent of the amount of data that the Library of Congress has in its entire print collection, right? But most of it is like cat videos on YouTube or 13-year-olds exchanging text messages about the next Twilight movie.”
Nate’s comment addresses the myriad of information that’s available. From a business viewpoint, organizations struggle to sift through and perceive what their customers are really saying when a messy, unstructured database is involved. In short, text data can cause headaches.
Our manufacturing customers tell us that studying text data helps them face challenges more successfully. Consider the following questions aided by text analytics:
Product quality engineers:
- How can we get a more nuanced read on product defects before they hit our warranty system?
- What circumstances lead to a product failure?
Digital marketing analysts:
- What keywords could be used on our product pages to improve organic search results?
- How can we improve our search engine optimization (SEO) strategy?
Vehicle launch teams / product managers:
- How do people feel about our new vehicle on social media?
- Should we modify the hashtags on our Twitter campaign?
- Which comments from our annual satisfaction study represent a majority versus a small, yet vocal employee population?
- How do employee comments posted on certain online company review sites impact the risk of losing good people?
Corporate aftermarket teams:
- What are our dealers writing in the notes section on service or parts orders that reflects why a customer ordered a part?
- Can we improve the SKU mix for our dealers so they serve their customers better?
What’s the common thread in these questions?
- Data is there for the taking. Comments from warranty claims, vehicle repair orders, new buyer surveys, employee satisfaction studies and online posts or remarks give further color to numeric analyses. The data is certainly there!
- Data analysis fuels proactivity. Digital marketers want to plan ahead to best position their products with relevant content while spending the least amount of money for paid/sponsored ads, etc. HR executives want to know the nuanced meaning behind employee comments so they can avoid losing high-performing staff. Aftermarket teams want to improve their parts demand forecast for their dealer network.
- Data leads to financial gain. Reducing the volume of warranty claims or avoiding product failures will directly impact the balance sheet.
Text for prediction
Early in my career, I worked in market research. I was an analyst and consumer surveys were my data source. I usually designed and analyzed attitudinal studies for automotive clients.
For example, I answered questions such as “what is the sentiment of first-time vehicle buyers?” or “what do online car shoppers expect of a dealer once they become an online lead?” Back then, I wish someone had taught me that comments from surveys could actually be used to predict outcomes as opposed to just reporting topical themes, or citing verbatims from open-ended survey questions.
Compare the impact between these two statements. For illustrative purposes, assume the data came from a consumer survey that studied online shopping experiences:
Insight using text data to report descriptive themes:
“Numerous comments from respondents indicate a high distrust level of the online shopping process.” (So what, right?)
Insight using text analytics to create predictor variables for a logistic regression model:
“When online shoppers mentioned ‘it takes too long’ or ‘they didn’t return my phone call,’ their odds of switching to another dealer increased by 12 percent.” (Now we’re on to something useful.)
Text analytics is a powerful tool that adds predictive value. Consider these two scenarios:
- Customer treatment decisions during live chat sessions: Scoring comments during a live chat session brings attention to escalation needs. A customer service representative can contact their manager or invoke a new script to best resolve a certain discussion or transaction based on real-time chat scores.
- Reducing unnecessary employee attrition: Employee comments from annual satisfaction studies can be correlated with attributes implying possible departure risks. By transforming key words or topics into independent variables, they serve as indicators to help predict the outcome of employees leaving the company. This gives the employer the opportunity to address any concerns and potentially keep valued employees.
The technology available today improves our data mining results and predictive modeling. If your listening platforms don’t include the ability to utilize messy, unstructured text data, you’re missing a chance to better serve your employees or customers. To this end, I encourage you to watch this on-demand webinar showing how machine learning (which makes data mining and predictive modeling more robust and productive!) is aided by a contextual analytics approach.
Disciplined approaches to understanding text data is no longer a nice-to-have. The Titanic sunk for a reason: The crew didn't see what was there. Don’t be the Captain Smith of your organization.