Lisa Painter: Welcome to the CXNow podcast. I am Lisa Painter and I am joined today by the head of customer care at Ipsy, TJ. TJ, it's so nice to sit down with you. Today we're gonna talk all about how better contacts leads to better AI. So welcome.
TJ Stein: Awesome. Thanks for having me. I appreciate you inviting me to to join you today.
Lisa Painter: Well, you are a thought leader when it comes to context. I've had a chance to interact with you in our thought leaders community. And so let's get right down to it about context. So Ipsy is a subscription business. And so of course, all customer interactions kind of build upon the last one. And so I would love for you to share with our listeners today, in your world, what is what does it really mean to have good or really great customer context? And why do you feel that it has become so important now that AI is more and more integrated into customer experience?
TJ Stein: Yeah, no, it's an awesome question and I don't know if this is specific to Ipsy, but I think rather like all mature sort of AI deployments in that, you're right, customer context is a requirement for that sort of true agentic or like conversational experience. So for Ipsy, in order for us to make really informed decisions about how to best support members, we need to know, you know, who we're supporting, what are their beauty preferences, their their purchase history. And so the way I sort of think about the maturity model within you know these sort of AI deployments is that you know tier one deployments are often like you know a glorified knowledge base, like I'm just giving you shipping information. It's not specific to you. Tier two is where I think a lot of teams are today in that you can give someone relevant or helpful information about their tracking details or how much store credit they have. I think tier three is where we most teams wanna get to and we're starting to get to as well is like we're actually taking action on behalf of customers, right? So using all that relevant context and detail about who they are and and Now we're able to process a return, give them a refund. so in order to get to that spectrum of AI maturity, like you have to have really deep integrations with a lot of internal systems to get to that level of context.
Lisa Painter: Mm-hmm. I feel like we hear that a lot. And I love how you compare it to a maturity model. I feel like we've been doing that for many years in customer experience. This is just another way that we need to really take a good check on our maturity. So let's talk about misconceptions or maybe actually not that. Something companies may miss. Let's let's start with misses. So it's easy. Organizations may miss something that they really kind of need to have in regards to customer context before they start doing things like you were talking about before you start to change how your team is using ai or how AI is using the context. So what what is best? What is needed there?
TJ Stein: Okay. Yeah, I actually I don't think it's so much like the metadata stuff like so order history or any of that sort of stuff. That's helpful, of course, but I think one of the things that causes the biggest misses is around the repeat contact for the same resolved issue. So I think the way most support systems work today is that they treat every conversation as its own sort of unique event. So when someone emails you you know three times About a broken feature or a missing order, you know, your AI agent on the conversation three opens with You know, I'm so sorry to hear that, let's get this sorted. And then it's the exact same tone and approach
Lisa Painter: What?
TJ Stein: they took in the other prior to you know treatment. So you know, for for the customer it's not just the third interaction, that's strike three, right? So in the early days at at Miyundi's I had a CX agent who would respond to missing items or missed packs with something to the effect of my gosh, I'm so sorry, this literally never happens. I'm gonna personally pick an pack your replacement order at no cost to you and this should never really happen. The intent was very pure. Be overly empathetic, take ownership, and just get an enormously great CSAT score. The problem is that for some members, it happened again. So that gesture of empathy and goodwill that you demonstrated last month comes in pretty empty on month two when it's just kind of pulled from that script, right? And the same sort of motion applies to AI programs, right? This is one of those experiences that feels awful to navigate. because as a consumer it just comes off as like very robotic, right?
Lisa Painter: Hmm. Yeah, that's a huge miss. And like you said, support agents are just wanting to be helpful. And in this siloed world of tickets versus just picking up a conversation where you left off is very different. So let's talk about how better AI happens through better context. And so I would love to know how you have seen AI be successful. And do you feel like it's related more to the model or Do you think it's more related to customer data? Is it related to the workflows or maybe even the platform that you're leveraging? Would love
TJ Stein: Okay.
Lisa Painter: your just experience there. I know you've worked for many different brands and you've been there done that.
TJ Stein: Yeah, look, I mean, you know, I I think specific to CX in our space, I think customer context definitely has a more pronounced impact on the success of an interaction versus like whatever model or platform you're using. I there's a lot of variances, of course, but I think for most platforms and vendors, you don't have a choice of like the the model you're using. You just get whatever they're using. And I think you know, a lot of the frontier model advancements that we read about. About an experience are for like general purpose use cases. So which is kind of crazy, right? We have access to like the most powerful intelligence, arguably the most sophisticated models, and I'm like this weekend I'm asking, like, how how long do you t how long does it take to caramelize onions? So you know, for a lot of what we do in CX, these sort of straightforward customer support interactions using AI, I don't need a model that solves like complicated physics problems or like concode an app. I just need very reliable reasoning. So in fact I think a lot of the more advanced tools on the market today have sort of rolled their own models very finely tuned for s support And so forth. So for the past year or so at IPSY, you know, we've been on this fairly persistent journey to expand like the breadth of context that we send into our platform. And despite all the model and vendor advancements and everything else, what's made the biggest difference is giving it an additional API endpoint, giving it the ability to take action on the customer's behalf. So what we've been doing is analyzing What are the gaps in our service model? Like what's leading to the most handoffs? Where do customers get frustrated? What dispositions lead to greatest handoffs and so forth? So from there, we just started kind of chipping away at those problem areas. And we've seen this kind of like slow and steady improvement. But just to answer your question succinctly, it's primarily been that line of work versus like anything that you know this happened on like the Model layer.
Lisa Painter: Mm. And I you know your customers, I'm sure, appreciate it as they continue to see that you understand them more. 'Cause that's all we want. when we interact with humans in our, you know, personal life, we just wanna know that they're they've got context about us and that they understand us and they're hearing us and we we also look for others to help us. And so I think that's really awesome. Let's talk about foundational groundwork. I think that a lot of the things you've already said kind of speaks to that, but would love for you to share maybe where you think organizations may be underestimating what it takes to have the true foundational groundwork that you need to kick off your AI journey and customer experience.
TJ Stein: Yeah, so I I think I'm sure both you and I have read about like unsuccessful AI deployments and how like these you know BCGs and Gartners of the World have said, Hey, you know, a lot of a lot of a companies are investing in these programs, but they're not necessarily like paying dividends quite yet. I think it's not so much there isn't like the right technology or like even like capital resources to make it work, but It often fails and again specific to CX deployments, it's failing when a lot of times there's just so much discretion in your service model. So I've often said like if your SOPs have a lot of it depends,
Lisa Painter: Yeah.
TJ Stein: that's gonna create some problems for your your deployments. So I think it's difficult for you know, some organizations to make that leap when their service model is predicated on judgment and discretion. Right. Hey, I I used to one of my prior teams you know is all in house. I used to say, I trust your judgment, you know, to to make the right call and so I think when you're bringing AI into your service model, those sorts of philosophies need to adapt to this new kind of environment. So I think, you know, well structured
Lisa Painter: Yeah.
TJ Stein: SOPs, escalation paths, all that stuff to just give AI the sort of you know, understanding of what to do when where to go for help. I think the other thing is like some sort of like governance or guardrails. you know, I think the risk of hallucinations is like overblown a little bit at this point, but most
Lisa Painter: Yeah.
TJ Stein: platforms ship with some type of like coaching or quality framework. And then you know having somebody with like a clear mandate, you know, c clear ownership. We have at Ipsy we have two full-time team members that focus on strategy implementation, conversation design. And they have a very clear mandate around what success looks like. it's very ambitious, difficult work, but I think that unglamorous and underappreciated groundwork really sets the foundation for a lot of success for these programs 'cause without it, you know, I I don't think the results are gonna be as as significant.
Lisa Painter: I agree with you. I think we've seen this in the history of of work, right? If you don't have governance, lacking strategy, you are going to run into all the things you just mentioned. And so AI, it being such a pivotal big change in how we work. I mean, if there's any time where it's calling for that sort of role or roles in your organization, it is now, right? Yeah,
TJ Stein: Yes.
Lisa Painter: absolutely. So maybe we could talk a little bit about misconceptions that leaders may have about. what it takes to implement AI into customer experience. I know that you've done this, you just shared some of your, you know, kind of groundwork roles and the way that you kind of look at it, but what are some misconceptions we should be aware of? Because some of our listeners may be showing up to to hear just that.
TJ Stein: Okay. Yeah, I think there are a few. I mean, I th I could probably spend a whole episode on to talking about the this s this particular question, but I think there's a really obvious one and that I think you know, most CX leaders still approach AI as something that just gets kind of tacked or bolted on to the existing operation, right? So I'm gonna pick a vendor, I'm gonna connect it to my help desk or my CRM and I'm just gonna start deflecting, you know, tickets. And so in reality I think AI implementation often requires like kind of a complete service model overhaul. So we've been reimagining things, goals, strategy, staffing model, even things that impact like our agent personnel, like career development, all that stuff
Lisa Painter: Mm.
TJ Stein: has been kind of updated. so you know, misconception one is is it's it's it's not just a chat bot you connect in reality, it's like a whole new labor service layer. Number two I think is You know, that it's kind of a plug and play kind of set it and forget it solution.
Lisa Painter: Right.
TJ Stein: and I th I think I've seen a lot of folks sort of overestimate maybe the the software or the model and underestimate all of the surrounding data process and like oversight stuff you need to like make it successful. so I I sort of think about these platforms as the Lego set, right? They're gonna give you all the pieces, all the instructions, but at some point like somebody's gonna have to put together That Lego said to make the whatever you're building. So and I think some of that comes from you know, a lot a lot of consumers interact with AI now through Chat GPT or Claude or something, and you'll
Lisa Painter: Mm.
TJ Stein: be lured into thinking like AI can solve any problem. So
Lisa Painter: Yeah.
TJ Stein: but that's just not how it works inside like the enterprise, right? So it takes a lot
Lisa Painter: Yeah.
TJ Stein: of work to get right, and we're still very much in like early days, so like I'm skeptical of anyone that says like they've figured figured it all out. Like everybody's still learning right now. So a lot of misconceptions to navigate for for sure.
Lisa Painter: Yeah, that is true. And that first one that you mentioned, I mean, hey, l we can't we can't blame a lot of these organizations because we've had a lot of vendors just saying, It's easy. You can just do this
TJ Stein: Yes.
Lisa Painter: quickly. We're gonna have things run up and running. It's gonna be magic. And, you know, that was kind of early on as people really started to think about implementing AI. And then the plug and play thing, we hear it often in our thought leaders community that This actually exposes some really bad stuff. So, you know, where you think you may have really great data or really great knowledge or your workflows may make sense, i AI is gonna expose that that may not be the case. And so the fact that you're reimagining and really kind of thinking through, I feel like most organizations are gonna be forced into doing that.
TJ Stein: yeah. I don't blame the vendors. Like they they need to sell their stuff. So like their job is to say it's gonna be really easy, but I'm here to say it can be challenging sometimes.
Lisa Painter: Good. Well, your peers appreciate you for that. So let's talk about some areas or maybe an example of where AI underperforms. So again, let's revisit is it the model? Is it the underlying data? Is it the knowledge workflow? Would just love for you to like maybe just pin right in on that and talk about where you've seen it underperform and why.
TJ Stein: Yeah, look, I mean this is maybe a a a theme upon a theme at this point, but it it's usually the inputs, not some like you know, model or vendor feature that's missing. So that and that usually shows up in in my my experience in you know, missing knowledge, a gap in some maybe messy workflow. And so the information may technically exist, but It's usually like scattered across help centers, macros, and like loose policy documents. So I think those like missing edge cases, sequencing and exceptions, those are the sort of things that cause underperformance in the AI program, right? So for for me, I think what I've also seen is AI is really good at exposing like ambiguity and the procedures, right? So if I'm a customer and I ask you, I'm like, hey, I'm trying to use this promo code, but it doesn't work, why? Inside, you know, that playbook or that SOP, your AI agent needs to understand what are the conditions that determine promotion eligibility. What happens
Lisa Painter: Uh-huh.
TJ Stein: if you have multiple sources and they disagree? what happens when you need to make an exception? What's the monetary authority that that AI agent has? So there's probably like twenty different questions. So I guess like That's a very simple question for most like customer, you know, tier one customer support reps to answer. But without those you know, definitions and playbooks in place, the A the AI agent is gonna struggle every time. So that's where I see underperformance more often than not is when you haven't really sat down and defined like here's exactly what you need to know to navigate the most complex
Lisa Painter: Perfect.
TJ Stein: issues and also like your most common issues. Yeah.
Lisa Painter: I think the organizations that for have traditionally been really great about onboarding their new people to the organization or giving them the information, the context, the SOPs are the ones that are going to do better in this AI journey. Again, it goes back to it's going to expose if it's really bad. Your AI is going to just really suck.
TJ Stein: Yeah.
Lisa Painter: Yeah. All right. So let's talk a little bit about value. So we all know, or you know, I know, and probably most of our listeners know that CX is a value center. But it's really important that we're able to measure it. And so I'd love for us to share, you know, ways where how do you know that you're getting value out of AI versus just automating tasks? And then I think it would be good to share with our viewers maybe a couple metrics that really matter to you to you today in this AI world.
TJ Stein: Yeah, I think you know, earlier in the conversation I mentioned that maturity model. And I think those brands and and teams that do exceptionally well, they're having an impact on the actual business performance. And not every business operates the same way. We're a subscription business, so we we talk a lot about you know, churn and lifetime value and those sort of retention based metrics. But regardless, I think for every company there's some essence of like business performance that needs to Materialize. So you know, for us, retention lifetime value. For other brands, it might be order value and conversion rate. But again, it's it's it cannot be just a simple like vanity metric for CX programs. I think you know, beyond that, there are a few core dimensions that sort of like define value within most AI programs, including ours. I think usually there's some component of like the economics, right? So what is your fully loaded cost per resolution? So tha again that that is defined differently by many teams, but I think there's like an economic component of like how are we performing on a per resolution basis? Of course, some dimension of like sentiment, right? Or loyalty. So if it's CSAT and a promoter score or effort score, you gotta be able to like capture how people are feeling as a result of that interaction, whether or not it was contained or escalated. And then like a true resolution rate, you know, I I know it's quite common for us to talk about why maybe deflection isn't a great barometer for success. I think think you know, we know some customers may get so frustrated they just leave, right? So that's not
Lisa Painter: Yeah.
TJ Stein: success for most CX leaders, right? But a resolution is, right? We it means that your your program, your your bot or your AI program, was able to resolve the issue fully. And so that's ultimately the best outcome, but it surprisingly that's not always the easiest to like grasp or understand. But so I again I guess maybe to summarize, I think success can be many things and, you know, using that success from a KPI standpoint, brands may choose to like optimize for different things, but for us like, you know, loyalty and retention is number one.
Lisa Painter: Yeah. Well I know we're gonna have a lot of people listening to this particular podcast that's going to relate to what you just shared. And it went so quickly because we're to my last question for you. And so what I like
TJ Stein: Yeah.
Lisa Painter: to do is leave our listeners with some advice. It's some good food for thought. So say there's another head of customer care that's listening today or in a role that's similar to yours, and especially in a retail brand. what would you say as, you know, one thing you know, one piece of advice you would give them if they were about to start their AI journey before they ever, you know, enter one single prompt. Like what's something they should know?
TJ Stein: I think it's very noisy right now in this space. You know, I I've been fortunate to be involved with a few great brands in the past and I think a lot of CX leaders are getting squeezed. so just as an example, every, you know, head of CX probably right now has a dozen or so LinkedIn messages from you know prospective vendors that would like to do business and solve all their problems. You know, your your company leadership is likely pushing for more automation, greater cost efficiencies. Your team is probably asking you for tools to improve. their quality of life, right? And
Lisa Painter: Mm-hmm.
TJ Stein: most importantly, your customers probably want like faster, more efficient, more friendly responses. So yeah, I guess like being squeezed is like the best way to describe this kind of position that we're in right now. And so I'm in that same seat. I feel it. I think my guidance is for anyone in that position is to be very Ruthlessly focused on great customer outcomes. it's very easy for you know, clarity to become very murky in terms of what's most important to me right now in this present moment. And so I always have to remind myself, even though it feels, you know, somewhat natural at this point after a few a few years of doing this, like what does success look like? Let that guide your decision making around what's best for members and what's best for customers. 'cause I I think it's very easy to lose focus and and be pulled in many different directions.
Lisa Painter: Yeah, I completely agree with you. I love ruthlessly focused on outcomes and success. That's a great way to leave the CX Now podcast for now. T J I just really enjoyed our conversation. I enjoy it every time I see you pop into our community meetings. Can't wait to meet you in real life. It's
TJ Stein: Yeah, likewise.
Lisa Painter: yeah, awesome. Well, I just look forward to just seeing this, you know, get out and the thought leadership to be shared with others that need it. and for those that are listening, we hope you join us again on the CX Now podcast. Bye now.
TJ Stein: Thanks for having me.
TJ Stein: Awesome. Thanks for having me. I appreciate you inviting me to to join you today.
Lisa Painter: Well, you are a thought leader when it comes to context. I've had a chance to interact with you in our thought leaders community. And so let's get right down to it about context. So Ipsy is a subscription business. And so of course, all customer interactions kind of build upon the last one. And so I would love for you to share with our listeners today, in your world, what is what does it really mean to have good or really great customer context? And why do you feel that it has become so important now that AI is more and more integrated into customer experience?
TJ Stein: Yeah, no, it's an awesome question and I don't know if this is specific to Ipsy, but I think rather like all mature sort of AI deployments in that, you're right, customer context is a requirement for that sort of true agentic or like conversational experience. So for Ipsy, in order for us to make really informed decisions about how to best support members, we need to know, you know, who we're supporting, what are their beauty preferences, their their purchase history. And so the way I sort of think about the maturity model within you know these sort of AI deployments is that you know tier one deployments are often like you know a glorified knowledge base, like I'm just giving you shipping information. It's not specific to you. Tier two is where I think a lot of teams are today in that you can give someone relevant or helpful information about their tracking details or how much store credit they have. I think tier three is where we most teams wanna get to and we're starting to get to as well is like we're actually taking action on behalf of customers, right? So using all that relevant context and detail about who they are and and Now we're able to process a return, give them a refund. so in order to get to that spectrum of AI maturity, like you have to have really deep integrations with a lot of internal systems to get to that level of context.
Lisa Painter: Mm-hmm. I feel like we hear that a lot. And I love how you compare it to a maturity model. I feel like we've been doing that for many years in customer experience. This is just another way that we need to really take a good check on our maturity. So let's talk about misconceptions or maybe actually not that. Something companies may miss. Let's let's start with misses. So it's easy. Organizations may miss something that they really kind of need to have in regards to customer context before they start doing things like you were talking about before you start to change how your team is using ai or how AI is using the context. So what what is best? What is needed there?
TJ Stein: Okay. Yeah, I actually I don't think it's so much like the metadata stuff like so order history or any of that sort of stuff. That's helpful, of course, but I think one of the things that causes the biggest misses is around the repeat contact for the same resolved issue. So I think the way most support systems work today is that they treat every conversation as its own sort of unique event. So when someone emails you you know three times About a broken feature or a missing order, you know, your AI agent on the conversation three opens with You know, I'm so sorry to hear that, let's get this sorted. And then it's the exact same tone and approach
Lisa Painter: What?
TJ Stein: they took in the other prior to you know treatment. So you know, for for the customer it's not just the third interaction, that's strike three, right? So in the early days at at Miyundi's I had a CX agent who would respond to missing items or missed packs with something to the effect of my gosh, I'm so sorry, this literally never happens. I'm gonna personally pick an pack your replacement order at no cost to you and this should never really happen. The intent was very pure. Be overly empathetic, take ownership, and just get an enormously great CSAT score. The problem is that for some members, it happened again. So that gesture of empathy and goodwill that you demonstrated last month comes in pretty empty on month two when it's just kind of pulled from that script, right? And the same sort of motion applies to AI programs, right? This is one of those experiences that feels awful to navigate. because as a consumer it just comes off as like very robotic, right?
Lisa Painter: Hmm. Yeah, that's a huge miss. And like you said, support agents are just wanting to be helpful. And in this siloed world of tickets versus just picking up a conversation where you left off is very different. So let's talk about how better AI happens through better context. And so I would love to know how you have seen AI be successful. And do you feel like it's related more to the model or Do you think it's more related to customer data? Is it related to the workflows or maybe even the platform that you're leveraging? Would love
TJ Stein: Okay.
Lisa Painter: your just experience there. I know you've worked for many different brands and you've been there done that.
TJ Stein: Yeah, look, I mean, you know, I I think specific to CX in our space, I think customer context definitely has a more pronounced impact on the success of an interaction versus like whatever model or platform you're using. I there's a lot of variances, of course, but I think for most platforms and vendors, you don't have a choice of like the the model you're using. You just get whatever they're using. And I think you know, a lot of the frontier model advancements that we read about. About an experience are for like general purpose use cases. So which is kind of crazy, right? We have access to like the most powerful intelligence, arguably the most sophisticated models, and I'm like this weekend I'm asking, like, how how long do you t how long does it take to caramelize onions? So you know, for a lot of what we do in CX, these sort of straightforward customer support interactions using AI, I don't need a model that solves like complicated physics problems or like concode an app. I just need very reliable reasoning. So in fact I think a lot of the more advanced tools on the market today have sort of rolled their own models very finely tuned for s support And so forth. So for the past year or so at IPSY, you know, we've been on this fairly persistent journey to expand like the breadth of context that we send into our platform. And despite all the model and vendor advancements and everything else, what's made the biggest difference is giving it an additional API endpoint, giving it the ability to take action on the customer's behalf. So what we've been doing is analyzing What are the gaps in our service model? Like what's leading to the most handoffs? Where do customers get frustrated? What dispositions lead to greatest handoffs and so forth? So from there, we just started kind of chipping away at those problem areas. And we've seen this kind of like slow and steady improvement. But just to answer your question succinctly, it's primarily been that line of work versus like anything that you know this happened on like the Model layer.
Lisa Painter: Mm. And I you know your customers, I'm sure, appreciate it as they continue to see that you understand them more. 'Cause that's all we want. when we interact with humans in our, you know, personal life, we just wanna know that they're they've got context about us and that they understand us and they're hearing us and we we also look for others to help us. And so I think that's really awesome. Let's talk about foundational groundwork. I think that a lot of the things you've already said kind of speaks to that, but would love for you to share maybe where you think organizations may be underestimating what it takes to have the true foundational groundwork that you need to kick off your AI journey and customer experience.
TJ Stein: Yeah, so I I think I'm sure both you and I have read about like unsuccessful AI deployments and how like these you know BCGs and Gartners of the World have said, Hey, you know, a lot of a lot of a companies are investing in these programs, but they're not necessarily like paying dividends quite yet. I think it's not so much there isn't like the right technology or like even like capital resources to make it work, but It often fails and again specific to CX deployments, it's failing when a lot of times there's just so much discretion in your service model. So I've often said like if your SOPs have a lot of it depends,
Lisa Painter: Yeah.
TJ Stein: that's gonna create some problems for your your deployments. So I think it's difficult for you know, some organizations to make that leap when their service model is predicated on judgment and discretion. Right. Hey, I I used to one of my prior teams you know is all in house. I used to say, I trust your judgment, you know, to to make the right call and so I think when you're bringing AI into your service model, those sorts of philosophies need to adapt to this new kind of environment. So I think, you know, well structured
Lisa Painter: Yeah.
TJ Stein: SOPs, escalation paths, all that stuff to just give AI the sort of you know, understanding of what to do when where to go for help. I think the other thing is like some sort of like governance or guardrails. you know, I think the risk of hallucinations is like overblown a little bit at this point, but most
Lisa Painter: Yeah.
TJ Stein: platforms ship with some type of like coaching or quality framework. And then you know having somebody with like a clear mandate, you know, c clear ownership. We have at Ipsy we have two full-time team members that focus on strategy implementation, conversation design. And they have a very clear mandate around what success looks like. it's very ambitious, difficult work, but I think that unglamorous and underappreciated groundwork really sets the foundation for a lot of success for these programs 'cause without it, you know, I I don't think the results are gonna be as as significant.
Lisa Painter: I agree with you. I think we've seen this in the history of of work, right? If you don't have governance, lacking strategy, you are going to run into all the things you just mentioned. And so AI, it being such a pivotal big change in how we work. I mean, if there's any time where it's calling for that sort of role or roles in your organization, it is now, right? Yeah,
TJ Stein: Yes.
Lisa Painter: absolutely. So maybe we could talk a little bit about misconceptions that leaders may have about. what it takes to implement AI into customer experience. I know that you've done this, you just shared some of your, you know, kind of groundwork roles and the way that you kind of look at it, but what are some misconceptions we should be aware of? Because some of our listeners may be showing up to to hear just that.
TJ Stein: Okay. Yeah, I think there are a few. I mean, I th I could probably spend a whole episode on to talking about the this s this particular question, but I think there's a really obvious one and that I think you know, most CX leaders still approach AI as something that just gets kind of tacked or bolted on to the existing operation, right? So I'm gonna pick a vendor, I'm gonna connect it to my help desk or my CRM and I'm just gonna start deflecting, you know, tickets. And so in reality I think AI implementation often requires like kind of a complete service model overhaul. So we've been reimagining things, goals, strategy, staffing model, even things that impact like our agent personnel, like career development, all that stuff
Lisa Painter: Mm.
TJ Stein: has been kind of updated. so you know, misconception one is is it's it's it's not just a chat bot you connect in reality, it's like a whole new labor service layer. Number two I think is You know, that it's kind of a plug and play kind of set it and forget it solution.
Lisa Painter: Right.
TJ Stein: and I th I think I've seen a lot of folks sort of overestimate maybe the the software or the model and underestimate all of the surrounding data process and like oversight stuff you need to like make it successful. so I I sort of think about these platforms as the Lego set, right? They're gonna give you all the pieces, all the instructions, but at some point like somebody's gonna have to put together That Lego said to make the whatever you're building. So and I think some of that comes from you know, a lot a lot of consumers interact with AI now through Chat GPT or Claude or something, and you'll
Lisa Painter: Mm.
TJ Stein: be lured into thinking like AI can solve any problem. So
Lisa Painter: Yeah.
TJ Stein: but that's just not how it works inside like the enterprise, right? So it takes a lot
Lisa Painter: Yeah.
TJ Stein: of work to get right, and we're still very much in like early days, so like I'm skeptical of anyone that says like they've figured figured it all out. Like everybody's still learning right now. So a lot of misconceptions to navigate for for sure.
Lisa Painter: Yeah, that is true. And that first one that you mentioned, I mean, hey, l we can't we can't blame a lot of these organizations because we've had a lot of vendors just saying, It's easy. You can just do this
TJ Stein: Yes.
Lisa Painter: quickly. We're gonna have things run up and running. It's gonna be magic. And, you know, that was kind of early on as people really started to think about implementing AI. And then the plug and play thing, we hear it often in our thought leaders community that This actually exposes some really bad stuff. So, you know, where you think you may have really great data or really great knowledge or your workflows may make sense, i AI is gonna expose that that may not be the case. And so the fact that you're reimagining and really kind of thinking through, I feel like most organizations are gonna be forced into doing that.
TJ Stein: yeah. I don't blame the vendors. Like they they need to sell their stuff. So like their job is to say it's gonna be really easy, but I'm here to say it can be challenging sometimes.
Lisa Painter: Good. Well, your peers appreciate you for that. So let's talk about some areas or maybe an example of where AI underperforms. So again, let's revisit is it the model? Is it the underlying data? Is it the knowledge workflow? Would just love for you to like maybe just pin right in on that and talk about where you've seen it underperform and why.
TJ Stein: Yeah, look, I mean this is maybe a a a theme upon a theme at this point, but it it's usually the inputs, not some like you know, model or vendor feature that's missing. So that and that usually shows up in in my my experience in you know, missing knowledge, a gap in some maybe messy workflow. And so the information may technically exist, but It's usually like scattered across help centers, macros, and like loose policy documents. So I think those like missing edge cases, sequencing and exceptions, those are the sort of things that cause underperformance in the AI program, right? So for for me, I think what I've also seen is AI is really good at exposing like ambiguity and the procedures, right? So if I'm a customer and I ask you, I'm like, hey, I'm trying to use this promo code, but it doesn't work, why? Inside, you know, that playbook or that SOP, your AI agent needs to understand what are the conditions that determine promotion eligibility. What happens
Lisa Painter: Uh-huh.
TJ Stein: if you have multiple sources and they disagree? what happens when you need to make an exception? What's the monetary authority that that AI agent has? So there's probably like twenty different questions. So I guess like That's a very simple question for most like customer, you know, tier one customer support reps to answer. But without those you know, definitions and playbooks in place, the A the AI agent is gonna struggle every time. So that's where I see underperformance more often than not is when you haven't really sat down and defined like here's exactly what you need to know to navigate the most complex
Lisa Painter: Perfect.
TJ Stein: issues and also like your most common issues. Yeah.
Lisa Painter: I think the organizations that for have traditionally been really great about onboarding their new people to the organization or giving them the information, the context, the SOPs are the ones that are going to do better in this AI journey. Again, it goes back to it's going to expose if it's really bad. Your AI is going to just really suck.
TJ Stein: Yeah.
Lisa Painter: Yeah. All right. So let's talk a little bit about value. So we all know, or you know, I know, and probably most of our listeners know that CX is a value center. But it's really important that we're able to measure it. And so I'd love for us to share, you know, ways where how do you know that you're getting value out of AI versus just automating tasks? And then I think it would be good to share with our viewers maybe a couple metrics that really matter to you to you today in this AI world.
TJ Stein: Yeah, I think you know, earlier in the conversation I mentioned that maturity model. And I think those brands and and teams that do exceptionally well, they're having an impact on the actual business performance. And not every business operates the same way. We're a subscription business, so we we talk a lot about you know, churn and lifetime value and those sort of retention based metrics. But regardless, I think for every company there's some essence of like business performance that needs to Materialize. So you know, for us, retention lifetime value. For other brands, it might be order value and conversion rate. But again, it's it's it cannot be just a simple like vanity metric for CX programs. I think you know, beyond that, there are a few core dimensions that sort of like define value within most AI programs, including ours. I think usually there's some component of like the economics, right? So what is your fully loaded cost per resolution? So tha again that that is defined differently by many teams, but I think there's like an economic component of like how are we performing on a per resolution basis? Of course, some dimension of like sentiment, right? Or loyalty. So if it's CSAT and a promoter score or effort score, you gotta be able to like capture how people are feeling as a result of that interaction, whether or not it was contained or escalated. And then like a true resolution rate, you know, I I know it's quite common for us to talk about why maybe deflection isn't a great barometer for success. I think think you know, we know some customers may get so frustrated they just leave, right? So that's not
Lisa Painter: Yeah.
TJ Stein: success for most CX leaders, right? But a resolution is, right? We it means that your your program, your your bot or your AI program, was able to resolve the issue fully. And so that's ultimately the best outcome, but it surprisingly that's not always the easiest to like grasp or understand. But so I again I guess maybe to summarize, I think success can be many things and, you know, using that success from a KPI standpoint, brands may choose to like optimize for different things, but for us like, you know, loyalty and retention is number one.
Lisa Painter: Yeah. Well I know we're gonna have a lot of people listening to this particular podcast that's going to relate to what you just shared. And it went so quickly because we're to my last question for you. And so what I like
TJ Stein: Yeah.
Lisa Painter: to do is leave our listeners with some advice. It's some good food for thought. So say there's another head of customer care that's listening today or in a role that's similar to yours, and especially in a retail brand. what would you say as, you know, one thing you know, one piece of advice you would give them if they were about to start their AI journey before they ever, you know, enter one single prompt. Like what's something they should know?
TJ Stein: I think it's very noisy right now in this space. You know, I I've been fortunate to be involved with a few great brands in the past and I think a lot of CX leaders are getting squeezed. so just as an example, every, you know, head of CX probably right now has a dozen or so LinkedIn messages from you know prospective vendors that would like to do business and solve all their problems. You know, your your company leadership is likely pushing for more automation, greater cost efficiencies. Your team is probably asking you for tools to improve. their quality of life, right? And
Lisa Painter: Mm-hmm.
TJ Stein: most importantly, your customers probably want like faster, more efficient, more friendly responses. So yeah, I guess like being squeezed is like the best way to describe this kind of position that we're in right now. And so I'm in that same seat. I feel it. I think my guidance is for anyone in that position is to be very Ruthlessly focused on great customer outcomes. it's very easy for you know, clarity to become very murky in terms of what's most important to me right now in this present moment. And so I always have to remind myself, even though it feels, you know, somewhat natural at this point after a few a few years of doing this, like what does success look like? Let that guide your decision making around what's best for members and what's best for customers. 'cause I I think it's very easy to lose focus and and be pulled in many different directions.
Lisa Painter: Yeah, I completely agree with you. I love ruthlessly focused on outcomes and success. That's a great way to leave the CX Now podcast for now. T J I just really enjoyed our conversation. I enjoy it every time I see you pop into our community meetings. Can't wait to meet you in real life. It's
TJ Stein: Yeah, likewise.
Lisa Painter: yeah, awesome. Well, I just look forward to just seeing this, you know, get out and the thought leadership to be shared with others that need it. and for those that are listening, we hope you join us again on the CX Now podcast. Bye now.
TJ Stein: Thanks for having me.
