AI Mines Product Reviews for Packaging Insights: Packaging World

How does PackSense, an AI-powered system for product quality insights, address the gap between controlled testing and real-world experience?

Editor-in-Chief Matt Reynolds sits down with Michigan State University (MSU) student Jordan D'Amario and MSU professor Dr. Euihark Lee. They discuss how AI is turning online reviews into actionable packaging data. MSU research shows how brands can pinpoint failure points, quantify consumer frustration, and prioritize redesigns using real-world feedback.

Read the full featured article on Packaging World.

Transcript

SPEAKER_00 0:03

Welcome to Through the Line, the podcast exploring innovations and information across the packaging and processing landscape. From topics impacting consumer packaged goods and healthcare packaging to the latest technologies in food processing operations.

SPEAKER_02 0:21

Hi, Matt Reynolds, editor of Packaging World magazine, back with another edition of Through the Line Podcast. Have a listen as I speak with Michigan State University's Dr. UHARK Lee and student researcher Jordan DiMario about PacSense. That's an AI-driven tool designed to mine consumer reviews on sites like Amazon for real-world packaging insights. From identifying failure points to quantifying consumer emotion, their work offers a new way for brands to understand how packaging actually performs once it's out there in the wild. As you can imagine, AI plays a central role in turning unstructured feedback into actionable data that brands and CPGs can use. So welcome, Dr. Lee and Jordan. Why don't you introduce yourselves?

SPEAKER_01 1:01

So I'm Jordan DiMario. I'm a student at Michigan State University. So go green. I have been working with Dr. Lee for about the past two years or so on this research, and now it's coming into fruition. And I'm presenting it here today at ISTA.

SPEAKER_02 1:16

Good. And Dr. Lee, introduce yourself first. And what's the genesis? This has to have been going on longer than revealing it today. So what's going on behind the scenes?

SPEAKER_03 1:24

Okay, so my name is Lee. I'm a Sun professor from the Scroll Packaging at MSU. So what I usually do is I try to bring some new aspect of the packaging evaluation tool that the traditional method couldn't catch up. So originally I was thinking about like a computer simulation in Dev Lab, but I realized there's some challenges because computer simulation is not fully represent what the packaging are facing. So then I moved to the database one. So that's how this accents project starts.

SPEAKER_02 1:55

Okay. So what problem are you solving with this data? What is the issue in this world that is going to help address?

SPEAKER_03 2:03

My understanding is actually if you find out the problem clearly, then finding solution is easy. Most of the case, we don't know what's the problem. That's make finding solution very challenging. So packaging issue is so diverse, so variations from the material, from the supply chain method, and then the way people handle. And then traditional methods cannot catch up those information well. So what we're trying to do is, because it's my personal experience, because whenever I love traveling, and whenever I go traveling and then finding a hotel, I always check the reviews. And then, okay, because we I got some very meaningful information from the hotel reviews, and why don't we grab similar ideas from the packaging reviews? So that's how this project starts. And then what we try to do is, because the older reviews are actually real cases. So by extracting some meaningful information from those reviews, we can actually catch what really happening into the real world that traditional method cannot catch up. So that's what we're trying to do throughout the review analysis to finding something traditional method cannot find it.

SPEAKER_02 3:13

Okay. So you're filling a gap that was there between traditional lab testing or field testing and the real world, which we all know behaves differently. But at the same time, you're these customer reviews have they vary from good, bad, thumbs up, thumbs down, to extremely nuanced to using foul language to having photos and not having photos. So what are you extracting from these very, extremely varied and disparate reviews?

SPEAKER_01 3:38

Okay, Carl. Yeah, so like you mentioned, consumer reviews are so varied. Some people say their entire life stories in their reviews, other people are very vague or just basic with their reviews. And what our framework aims to accomplish is to use AI to extract the packaging-related reviews. Because believe it or not, consumers do talk about packaging whether they realize it or not. So when we use generative AI and also our rule-based AI models, we can extract specific components, conditions, severity, and emotions from online reviews. So what that means is we can figure out which components of packaging were most damaged, which components of packaging were most affected by the consumer emotions, and how strongly were they damaged. So that it's to tone in the way where we can understand what kinds of packaging were damaged, and we can also understand how consumers react to packaging.

SPEAKER_02 4:36

Okay. Now translate that information, that new data, to Procter and Gamble or Unilever or whomever craft. How do they then make that action?

SPEAKER_01 4:45

Yeah. So if you're from a consumer or from a company standpoint, all you would need is like your Amazon review page link. And from there, what insights you can get is you could understand which components of your packaging were breaking the most. So whether that's if you have a bottle with a pump, we can understand whether it was the pump that was breaking, the closure, the seal, the actual bottle body that was breaking. And then we can also understand how that how your consumers reacted to that packaging damage. So from there we got a scale of consumer emotion. So quantifying the emotion cracked in the reviews. And when you match that with the components and the conditions that the components experience, then we can run targeted calculations to really understand where your design priorities should be.

SPEAKER_02 5:33

Okay, your redesign priorities specifically. Okay. Now you had on stage today, but you had, I can't, I don't know if this is a real brand or just like a pilot brand or like a test, like a beta test kind of thing. I think it was Tea Tree. I know that's a true brand too. So okay. What were your findings from this? This is a real world, this was something that consumers are actually opening up to varying degrees of satisfaction, let's say. And what did you find? What were you able to glean and extract from reviews and from the most sophisticated to the least sophisticated reviews? Photos, non-photos, that sort of thing. What did this case study tell you?

SPEAKER_01 6:04

So this case study really demonstrated the importance of having a package that is one functional two and two durable, because those are the components that we realize that consumers care about the most. Those are the components. So specifically, like with the case study, it was the closures, pumps, and the bottles. So again, the components that reduce or impact the functionality, those are the ones that drive the most negative consumer sentiment. So those are going to be the ones that you want to prioritize for redesign.

SPEAKER_02 6:34

Okay, and what does that look like? What does what's the action? Does it is it making keeping the seal unbroken and taking and maybe shipping with the entire closure and pump closure and everything separate that the consumer then does it himself?

SPEAKER_01 6:46

Yeah, so that's one of the ideas. Our program also is able to produce a packaging recommendation that's generated from generative AI, but we engineered a prompt to give those kind of high-risk areas. So in that case, you can understand, okay, if we have this many occurrences of the seal breaking, maybe you need a stronger, maybe you need to increase the adhesive temperature to make sure it's on there better. So things like that, the generative AI can pull out and give recommendations to. And then it's also able to provide engineering references like ISTA documents, other case studies that people have done in the past to then help with that ideation and redesign.

SPEAKER_02 7:29

Okay. Now, what it sounds like though is there it could be a simple fix that might not require a total redesign, but just moving from adhesive base to ultrasonic ceiling or something like that. Okay, good. Now, one of those are the things that you were hoping to find, but extraneously to the study, I think you said there are certain failures or breakages or leakers or so on, that really creates a certain emotional response that maybe you weren't expecting. So why don't you walk us through some of the unexpected things on consumer emotion that leaky packages give us all?

SPEAKER_01 8:00

Yeah, so I would say from our case study, we noticed that even though there's some damages with shipping boxes, those typically weren't as common or as frequent. And however, they still drove negative reactions. There's no big surprise that if your package gets damaged, consumers are going to be angry about it. But it's that certain components drive certain or certain failures drive certain emotions or stronger emotions. So we found, like I mentioned earlier, but things that affect the functionality. So for example, if a pump isn't working, all of a sudden the consumer has to take the bottle, transfer the product to another package. And that just adds a step of inconvenience that consumers honestly don't want to deal with. So I would say that was the biggest takeaway is that you really want to focus on certain components that drive the strongest negative reactions to make sure that you're optimizing the most consumer satisfaction.

SPEAKER_02 8:58

20% of the failures are causing 80% of the angst or something along those lines.

SPEAKER_01 9:02

Yes.

SPEAKER_02 9:02

Okay, our audience, packaging world's audience, largely brand owners, e.g., some contract packagers, some private label folks. So if they're watching, what's one packaging change that they could realistically make? You mentioned pumps and closures, that's mostly for one category. But what are some, if you can think of some of the your findings, what are some areas that they should really zoom in on?

SPEAKER_03 9:23

So I think because of the CPG has a lot of variety of the packaging type, it's quite challenging to pick up one specific component. But one thing I can tell is because this tool, we actually a strength of this tool is quantity of the data. So without this tool, I think company need to hire some like a like a pilot study. I don't know like what they call someone who put the people and then a pilot study, pilot study or at their specific time, I just focus group. I think it's focus group. They have some small focus group and then testing how they are using this packaging, how they apply this type of the features they implement. But in our tool, without any this type of focus group, it focus, having a focus group is time consuming and then very culturally. But we can actually extract hundreds of the thousands of the data, and then from there we can find out some critical issue that connects to the packaging and then consumer usage. So from there, they will identify oh, this is this packaging change has been working or is not working. I have some story with one packaging professional. I couldn't go detail, but they actually change some packaging. But they wonder, and that it's for consumer convenience, but they have no idea the consumer actually using that feature or not. And then this tool, our two packages, can easily capture if they really consumer really are using those type of tools. So from by understanding this packaging component and how these things are working with the consumer, that is one of the strongest benefits of this tool. And then that will be a key point of the or critical point to develop the new packaging.

SPEAKER_02 11:08

Okay, so it's the consumer research or consumer behavior feedback loop that a brand can learn about how the products the product is being used and the package is being used or misused out in the wild because some of these bad reviews aren't the brand's fault. It's consumer mishandling. We're at an ISTA conference. A lot of what we're talking about are more traditional, whether lab-based, field-based study studies, this is by no means meant to replace those, correct? This is a complement or an additional tool. How would you mix those together? How would this augment existing testing?

SPEAKER_03 11:35

Yeah, so industry standard testing standards is very important because one of the most important things as a development stage, because these standards give the consistent condition. So by changing some, if the testing conditions keep changing, then we don't know because if this packaging is working well because of the testing competition change, or if the packaging itself has been some update. So this consistent testing procedure gives some benefit of identifying if when they develop the packaging, if their packaging is working or not. So I think from the product development side, that would be very helpful. Our end, because we need data. We need a consumer review data. So our end is more likely like a backup data once they launch it. Alternatively, we still can use our tool to see if I need to develop the new shop model. Instead of just starting the scratch, I can just scrape in other computer design and then see how different designs are reacting. So that will give some better starting point of that. But to verify that one, still is that or just typical standard testing standards to require.

SPEAKER_02 12:43

Okay. Anything I'm missing, anything that I should be asking about? Again, my audience, mostly brand owners, CPGs, but what am I missing? What is the big picture takeaway that you want to make sure our viewers hear about or read about?

SPEAKER_01 12:55

I think just that one of the biggest takeaways is that there's another case study we did for a product that was more consumer-centrically designed from the start, and that had a significant more increase in consumer satisfaction. Okay. So I think that's one of the biggest takeaways here is when can companies design their packaging with the consumer in mind from the start, that pays off because consumers end up writing good reviews about how satisfied they are with the package.

SPEAKER_02 13:25

And good reviews that actually generate further purchases.

SPEAKER_01 13:28

Exactly. So it benefits the consumer and it also benefits the business as well. So I think this tool just helps amplify that.

SPEAKER_02 13:36

And if any of our readers want to find out more, learn more about Paxense, do I have that right? Uh do they go to the MSU.edu? Where do we find more information?

SPEAKER_03 13:44

Well, actually, we have a separate website. Okay. Currently, demo version is available. So you cannot actually download your Amazon link, but we already embedded one of the demo as an example. So if you go to the pacense.net, then you should be able to, there's a demo button, then you can see already like uh pre-implemented data you can play with that, what kind of data you can see, what kind of control you can have, those type of things.

SPEAKER_02 14:07

Okay, great. Both of you, thank you so much for your time today. We are in the middle of what would be a happy hour downstairs. So we're extracting them from the really fun part to do this. So thanks again for your time. Our readers appreciate it.

SPEAKER_00 14:19

Thank you for listening to Through the Line Packaging and Processing. You can listen to more episodes on all streaming platforms. Be sure to visit us at packworld.com, profoodworld.com, and healthcarepackaging.com for more packaging and processing news. This podcast was edited by Bree Guns.