An OdinText User Story – Text Analytics Tips Guest Post (AI Meets VOC)
Today on the blog we have another first in a soon to be ongoing series. We’re inviting OdinText users to participate more on the Text Analytics Tips blog. Today we have Kelsy Saulsbury guest blogging. Kelsy is a relatively new user of OdinText though she’s jumped right in and is doing some very interesting work.
In her post she ponders the apropos topic, whether automation via artificial intelligence may make some tasks too easy, and what if anything might be lost by not having to read every customer comment verbatim.
Of Tears and Text Analytics
By Kelsy Saulsbury
Manager, Consumer Insights & Analytics“Are you ok?” the woman sitting next to me on the plane asked. “Yes, I’m fine,” I answered while wiping the tears from my eyes with my fingers. “I’m just working,” I said. She looked at me quizzically and went back to reading her book.
I had just spent the past eight hours in two airports and on two long flights, which might make anyone cry. Yet the real reason for my tears was that I had been reading hundreds of open-end comments about why customers had decided to buy less from us or stop buying from us altogether. Granted eight hours hand-coding open ends wasn’t the most accurate way to quantify the comments, but it did allow me to feel our customers’ pain from the death of a spouse to financial hardship with a lost job. Other reasons for buying less food weren’t quite as sad — children off to college or eating out more after retirement and a lifetime of cooking.
I could also hear the frustration in their voices on the occasions when we let them down. We failed to deliver when we said we would, leaving the dessert missing from a party. They took off work to meet us, and we never showed. Anger at time wasted.
Reading their stories allowed me to feel their pain and better share it with our marketing and operations teams. However, I couldn’t accurately quantify the issues or easily tie them to other questions in the attrition study. So this year when our attrition study came around, I utilized a text analytics tool (OdinText) for the text analysis of our open ends around why customers were buying less.
It took 1/10th of the time to see more accurately how many people talked about each issue. It allowed me to better see how the issues clustered together and how they differed based on levels of overall satisfaction. It was fast, relatively easy to do, and directly tied to other questions in our study.
I’ve seen the benefits of automation, yet I’m left wondering how we best take advantage of text analytics tools without losing the power of the emotion in the words behind the data. I missed hearing and internalizing the pain in their voices. I missed the tears and the urgency they created to improve our customers’ experience.
A big thanks to Kelsy for sharing her thoughts on OdinText’s Text Analytics Tips blog. We welcome your thoughts and questions in comment section below.
If you’re an OdinText user and have a story to share please reach out. In the near future we’ll be sharing more user blog posts and case studies.
Thanks for sharing. While I’m a fan of using AI and Machine learning where it can eliminate tedious tasks, improve quality, and enable more time for strategic thinking, it only works when we couple it with human emotion and understanding. Empathy is what connects people, not data.
I’ve learned about how AI can be used to do more than analyze text. It also can be used to decide what topic to ask about next during an exchange with a respondent. (instead of a rigid survey, AI can allow for a “conversation” with a respondent) I look forward to hearing what anyone’s experience has been with using AI this way. thoughts?
Lisa, I really like your comment “empathy is what connects people, not data.” I’m sure to start saying things like that. thanks
If we could code Kelseys and others emotions (on similar tasks) as data.. we have created what i read somewhere Thick data. If we can code sensations such as taste temperature as data..why not the emotions?
At Brandtrust, we are Odin text users and enthusiasts. We’re also experts in understanding human emotions via applied social and behavioral science. I’m fond of reminding our teams that it’s hard enough for one human being to understand another let alone think that machines will can do it completely.
@Thanks Travis, yes I know BrandTrust are also doing some very interesting work with OdinText, and what I love about your firm is how you take it to the next level and also use it strategically, not just to save time but to do more advanced analysis that humans could not do with the data. This is of course why your clients love what you do, the additional above and beyond strategic analysis. Not just fast and cheap.
I think that’s the key point here perhaps, I agree that totally unsupervised AI has many problems, but OdinText allows for supervised Ai, i.e. the best combination is a delicate mix of machine and human.
Of course OdinText has emotional analysis, and in fact Anderson Analytics (from which OdinText came) was the first in the marketing research field to offer automated emotional analysis. It literally takes a second to identify ALL comments containing ‘sadness’ or ‘fear’ etc. in OdinText, The question is if, how and when you choose to use that feature.
Of course some wholistic human understanding that would come from reading every single comment verbatim might be lost, but then again this is usually an impossibility if data is larger and speed is of the essence. Also, the analytical capability of the application far exceeds what a human reader/coder could achieve.
I understand Kelsy is speaking/thinking more about where she sees Ai going in the future as it is used more often and becomes more and more unsupervised. An analyst may stop at seeing that say 5.5% of comments contain ‘Sadness’, but should they stop there or do they drill into those comments further?
In the past you didn’t know what those were automatically. So you HAD to read each one not knowing what they were. Now you know immediately. Now what do you do? Do you include a section on Sadness in your report? That’s up to the analyst 😉
Hopefully in the future, we won’t stop and/or ignore these and accept higher level statistics only.
Great post Kelsy! You’ve nailed the potential risk of replacing human coding with text analytics. I’m a technology guy so I lean toward trusting the steady march of progress in a machine’s ability to detect and highlight emotion. As a result, the analyst should use the time-saved to dive into the emotional pools and eddies in the data AND coming to grips with the emotion should still require reading a more select group of comments in verbatim form (and , increasingly, listening to audio…watching video.) In my opinion, that is the key to activating such insights, as Lisa’s comment underscores: using that emotional valence to connect your organization with the heartfelt need states of consumers in order to stimulate empathy and trigger action.
I can see her point about automation removing the analyst from the personal detail in the data. I do agree that data analysts and business people should not treat data like it is just a bunch of numbers and letters. It’s important to incorporate subject matter knowledge, secondary sources and personal contact into analytics and business process.
Still, there is another angle on the role of automation. When I worked in software, I routinely asked what clients what they did with open-ended responses. Often, the answer was: nothing. Not just no analysis. People were collecting data and not even looking at it.
One of my students put it very well. She said, “That’s a violation of trust.” People take the time to answer the question because they hope the response will motivate action.
If automation means that people will do something with open-ended responses rather than nothing, it’s an improvement.
Well said, and a great article. Thanks for sharing Tom
Kelsey is handling expressions of dissatisfaction in precisely the way they ought to be handled. If Kelsey’s the only one doing so therein lies the difficulty. The problem isn’t so much about lack of feedback as it is will to fix the issues the feedback uncovers. Actionable insight has been the watch word for a good decade yet customers still run into bad processes; one-sided policies; and under- supported front line people. Unless and until the senior leadership of a firm deal directly with verbatim comments , feedback remains dismissable as statistically in the noise. Indeed it has been the only thing that drove change in capital investment and operational support – aka putting your money where your mouth is.
Great post. Kudos to Tom, also, for providing the space for this!
There’s a lot to think about here from multiple perspectives. Kelsy is justified in her concern about the potential for AI to increase the distance between human beings. I share it.
But there is opportunity! Meta’s comments about ignoring comments offered in good faith from customers and Carol’s about dismissing them are excellent points! What good is promoted by collecting information that is ignored or dismissed? This post features a professional, Kelsy, who is deep into the open text but runs into processing roadblocks that make it hard to fix the problems she is uncovering. OdinText and similar offer opportunities to very quickly surface and class customers that might be facing the most difficulty and move their situation to the next point in the “fix-it” process. The AI text processing value chain done well, by people with compassionate motivation and effective operational processes, can be a good thing!
Automated analytics provide insights much quicker than data analysts can manually do. However, actions must be taken by humans, not a software. For example, once the data collected through clicks or surveys is analyzed, and it is discovered why users are happy or not happy with the product or service, you could interview these users in person to dig deeper and determine what works and can be duplicated, and what must be avoided. This will provide the happy users more reason to be happy, and the not happy users will at least feel listened to (of course as long as privacy is not breached and this is all done with user permission)