Numbers and Narratives
Numbers and Narratives bridges the gap between the marketing/customer experience and data - come listen to marketing and CX experts talk about how to use data to better engage with your customers and provide a great experience.
Numbers and Narratives
CX and AI: Pitfalls and the Future Ahead - Disha Gosalia, Career CCO
In the latest episode of Numbers and Narratives, we engaged in an insightful discussion with Disha Gosalia about the transformative impact of AI on customer experience (CX). Gosalia, drawing from her extensive experience in the customer experience space (most recently as Gladly's Chief Customer Officer), shared valuable perspectives on the current state and future of AI in CX.
Key takeaways from the conversation:
- AI adoption crucial is inevitable in the CX space
- Generative AI actually lowers the barrier for personalized experiences
- New AI-specific roles are emerging
- Privacy and hallucination concerns are a threat to adoption
Gosalia emphasized the importance of starting AI implementation now, while being mindful of potential challenges. She advised CX leaders to focus on efficiency and be cautious of unrealistic vendor promises. The discussion highlighted the evolving nature of CX roles and the need for companies to adapt their strategies in this AI-driven landscape.
You can find Disha here: https://www.linkedin.com/in/dishagosalia
I'm Amy Syed, I'm Sean Collins and this is Numbers and Narratives.
Speaker 3:Okay, Disha, in the length of a tweet or a tweet thread, what did we talk about today?
Speaker 1:CX and AI pitfalls and the amazing future ahead.
Speaker 2:Thanks so much for coming on, Nisha. Before we get started, do you mind reintroducing yourself for us really quick?
Speaker 1:Yes, of course. Well, thanks for having me, sean and Nibi, it's exciting to be here chatting with you both today. So I have almost a couple of decades of experience in customer facing roles. I've ran, you know, large global support teams, customer experience teams in large enterprises like GESAP, but also at, you know, startups and sort of midsize companies, and my last two roles have been in customer experience software companies, running customer success or sort of whole customer life cycle from onboarding, implementation, renewal, retention, and through that I've also got a lot of experience talking to leaders who run customer support teams, who own customer experience at large and small and medium-sized organizations. So, yeah, I guess that's what we are here to talk about. Organizations. So yeah, I guess that's what we are here to talk about. And now, with AI purely coming in and disrupting, I would say, the space, there's just a lot to talk about here.
Speaker 2:Yeah, I'm pretty stoked. So there's a couple of interesting things here. We have a lot of retention, customer experience, people on from both the B2C side and a little bit of the B2B side. What's hilarious is you kind of encompass a lot of these personas. You've got the fact that you've worked for a company that does customer experience. You have customer success ie retention and growth experience as well, as well as a couple other bits. So I'm super stoked for this conversation. I kind of want to set the stage by starting off in defining kind of what AI is in this space. Can you give us kind of a brief overview and a lay of the land as to what is? How have you used AI? How are these platforms that you've worked for incorporating AI? What does it actually like kind of mean? What are the customers looking for? What problems are they trying to solve? What does kind of the AI landscape look like today and what are people kind of investing more into?
Speaker 1:Yeah, totally. So just this is an interesting time we're talking about this Because I have been really curious about this as well, right, like, I think, with generative AI coming in really strong, there's just so many players out there now, so I was interested in collecting data real time. In the last couple of months, I've spoken to probably about 50 CX leaders and in all types of businesses, like big, large retail, b2b, b2c and yours sort of some of my readings. So, first of all, bearing maybe very few players, which is very high margin, high touch everyone is eager to either themselves jump into, like, using AI to perform their CX service or improve their customer experience, or they have a mandate from the leadership, like they may not know what they can do, but they're like, oh, my CEO is telling me we need to use AI for customer experience. So I would say, at this point, of all industries, customer service is seeing the most immediate and tangible value from AI. You know we always had the traditional. Since several years, we've had the traditional chatbots and the ML models right, they've been there for years, but it came with a set of challenges in terms of, like how, the maintenance of it, and then, you know, the implementation of it right. You had to create those workflows. You have to think of every use case out there, so it was harder for every company to implement without large investments.
Speaker 1:Generative AI is changing the game now. So the large language models are good at a lot of tasks in customer service, especially reading your already existing data and knowledge base and providing quick answers for repetitive type of questions. That's the phase one. That's almost everyone today should have that or are working towards having that right. So that's I want to say where it is. With generative AI, it's very easy to point it to your database and it also has the model right and you know it gives you the quick answers, which really doesn't require you know the empathy piece or the complex piece that humans bring. I mean, for example, if I want to reset a password as a customer or I want to know for a B2C company where my order is, like I really expect at this point zero response time, Like I don't care about how are you, what are you doing, where you're to help you. I would rather just have you tell me the answer and that is what, like today, ai is really good at.
Speaker 1:I think where the next phase is and maybe where there is some sort of I want to say discouragement that is coming for these CX leaders in my conversation is the software vendors are promising them a lot. They're promising that 50% resolution day rates, 50% reduction in their cost structures, and all of that is I don't think we are there yet, because what AI has not been able to solve today is really the long hold type of use cases. It's not we're not there yet, but I will tell you it is coming, it is coming really fast. So you know I would, if I have to talk to a CX leader, I would tell them like, hey, you start, you know, at phase one, right Understanding on how you can.
Speaker 1:What are your repetitive questions? What do you want AI to do without compromising customer experience? And then the scale is going to come, because today AI is also expensive, right, so you're still paying for each resolution, you're paying for a tool, you have to pay for whatever large language model you're using, for all of that, but that all it's going to. Technology is going to keep getting cheaper. I think there's a law which kind of says that in the short term, you know, we always overestimate the effect of technology, but in the short term, we underestimate it. I think that's kind of the state of AI in the CX world today.
Speaker 3:I like that and I like that line a lot that we overestimate in the long run and underestimate it in the short. Overestimate in the long run and underestimate it in the short. When you talk about, you know, starting with phase one of really looking at those repetitive tasks, what is the easiest way and like kind of least intrusive way for a team to figure out the tasks and the questions that are eating up so much of their time? Right, like I think one thing that I have found is everyone underestimates how long it will take them to actually do any one thing and I think they overestimate how often they do any one thing. So how is it like some sort of time tracking tool or some sort of counter that you can do to see where you're wasting time or where you're doing repetitive stuff? Or like other than just like a query into you know, like plugging in every chat transcript into an LLM, like how can you figure out where you need to automate first?
Speaker 1:Again, I think it is based on your business. But I think I would say, like in my conversation, what I found is this the CX leader start more with hey, I'm going to reduce X percentage point of my cost or I'm going to reduce X percentage staff. I don't know that that's the right way to start right now. The way to start right now is what is the job that I want AI to do and what is the job AI is good at doing? So I think understanding your business like if you start with this core understanding and you ask yourself how do I make my solution faster, cheaper and more efficient? Right, what is AI going to be good at helping that? And what my humans are going to be good at? Right, they're going to be better at certain use cases. They may be slower, more expensive, but you know you need humans for that. So I think putting together those use cases you know where I think AI can help, you know, make it more smarter and efficient, and where I still need to rely on the humans is kind of that first step. So I mean, for a retail B2C, it's a lot of like where's my order, you know? How do I process a return, kind of that's step one. Step two is how do I integrate with the systems and create that workflow? So now, where's my order? Can I call the API of my shipping company and, you know, provide that tracking right there?
Speaker 1:Now the third stage. There is kind of that action and which is where I think we still need a little bit more work. I heard, you know, one of these podcasts where, like this text to action is a big thing. That's, you know, today, you know we're working on in the LLM world and that's kind of what is going to help now that if my package, if my shipping company is still showing that my package is on the way, but it's been two weeks, so of course it's lost, right, what is the action that I'm supposed to take right now?
Speaker 1:At this point, you know you would have to create a workflow to hand it over to a human so they can make a call to the shipping company and do something. But in the future I know you know that's coming as well right where there will be a use case where AI is able to do that right make a call or talk to the system and locate the parcel and, you know, get you more. But yeah, I think as you think of, like, where do you start? First, I think of use cases, easier use cases, and then more workflow, algorithmic use cases, and then, of course, you'll get to just more complex ones. In time. That does get caught.
Speaker 2:That makes a ton of sense. I'm curious from your perspective. There are a ton of you know. One thing that we've kind of talked about the last couple of minutes that I find interesting is like there's a ton of use cases. They're very specific kind of to the industry and you mentioned this towards the beginning of when we started chatting but a lot of this is the immediate benefits are clear, right. The reason that this has been sort of taken by storm is that there's this big chunk of use cases that AI can just kind of take off people's plates. It means that people have to spend less on these teams. What are some other benefits, what are some other improvements that you've kind of seen in the CX space from being able to implement this AI technology? But also, like, where do you think it doesn't come to play? Because you and I had a conversation before this on sort of the harms and benefits, like what are some of the other sort of like benefits that you've seen? And then we can jump into the harms in a sec.
Speaker 1:Yeah, yeah. So I think in terms of benefits, of course, that is the use cases we talked about on how you can, you know, bring efficiency by answering questions that are easy, that are kind of there in your knowledge base. You know that does not require a lot of complexity and several steps. Number two is, you know, really using your data in the right way. This was something that wasn't easy to do to understand your user preferences, your customer preferences. Ai has made that easy right. So it kind of we know today that companies, especially today in the B2C world, they're doing some fantastic things to learn about their customer preferences, to whether it's past buying history, social media posts, likes and all of that they do, and now they can recommend their product and services to improve that engagement, get those repeat sales, et cetera.
Speaker 2:And then there was also some limitation, just a little bit. My company is part of one of those companies. We do that. So yes, 100%.
Speaker 1:Yeah, totally. And then you know there was just some limitation there on how you could contextualize and customize that content to your buyer. Right Now, generative AI is changing the game there as well, right, it's like I mean, I'll just give you an example, like if I bought a pair of pants online and the person would just get suggestions on, ok, what top would go with it and with the options of how I can pair it, and you know some stylist telling me what I can do. But now, if I can add the context on there, like here's my specific body type, or I'm only five foot one, right, generative AI now has the ability to look at other people like me and provide, like more customized recommendations. So I think that's kind of just from a marketing standpoint, there's a lot that is happening.
Speaker 2:So kind of what you're saying there, just so I understand, is kind of taking more data from what you, that you get from your customers right, like things that they might write in a support channel, things that they might give us context around other areas of your business, not just during the buying experience, and using that to say, alright, we know that this customer likes pants. We also know or likes pants and whatever, and we also know that they live in Michigan and we know that you know they like things that are generally more slim fit than classic fit because they wrote asking about something like that in a support call. Let's put them in a bucket of people that are very, very similar, that we know about and cross-reference. Hey, if somebody else who lives in a cold climate and prefers slim fit things also buys this specific type of hoodie, let's, you know, sell the thing because we know that they have similar interests. Like let's sell what the other person bought to each other. Is that is that kind of what you're? Kind of what you're?
Speaker 1:exactly, yeah, yeah. And then I know that's where the the combination of generative AI and then the data science models which companies like Europe sell something see, science models which companies like Europe so I'm assuming are doing, you know is solving for these more complex problems. Right, when you can become more industry, specific, more domain, more personalized these, what do you call as a last mile of the long haul problems? Because I guess data science can study more of those complex problems in that more controlled environment. It's just, I just I'm very excited, I'm optimistic about the future of where this can take us right, Like I just feel there's so many possibilities of use cases that I can't even imagine today. Right, Like I know, when Steve Jobs said that every household will have a personal computer, you know, people couldn't believe it because those were the days of supercomputer, it was so expensive to own that and people couldn't think of because those were the days of supercomputer, it was so expensive to own that and people couldn't think of it beyond a fast calculator.
Speaker 1:But look what it can do today, it's completely revolutionized, like how we work, buy, educate ourselves, do business, etc. So I know that we're kind of there at the cusp of this here in the next few years of what LLM and data model and all of that kind of coming together can do.
Speaker 3:We had a guy named Paul Meinshausen on shoot, probably like 15 episodes ago at this point, but he's a brilliant data scientist and one thing he said is stuck with me and like I haven't been able to get out of my head for five months now. He said too many companies rely on data being like provided to them and like very explicit event based data, and he said that what you really need to get to is realizing that you can create your own data and use that to fuel your model. And that's exactly what I kind of took away from that Like using chat transcripts, using review transcripts, using browsing data all of that to build a more complete customer interest profile and affinity profile and then being able to leverage that through your recommendation engine, through automations. Right, if you now have a profile of all these traits, like you know you were describing, it'd be that that fictional person in Michigan uh, when you launch a new product, they should that group of people should explicitly get a separate message being like we know you liked this and so you like you.
Speaker 3:We think you'll like this, right, and so it's not just the recommendation in like that. You might also like kind of three item carousel at the bottom. It is redefining the way you think about go to market and product launches. It's sort of like even you know I'm always blown away when I see Shopify brands that don't have a back-in-stock notification feature on Like. That's a default or not default, but that is a simple and no more cost feature from Klaviyo. I assume you're on Klaviyo and so it's crazy me you wouldn't use that no-transcript.
Speaker 1:So you know, hopefully that's going to open up, you know, create more competition and just open up the world to a lot more players.
Speaker 3:Yeah, it almost. It almost feels like we're back in, like I mean, obviously it's a completely different space. But 20 years ago, 15 years ago, how much you kind of just needed one super scrappy, hardworking, smart person who's just like kind of quote, unquote, figure the internet out. And now it's like, yeah, maybe I don't need a. Hey, here's the five years of experience I expect you to have. You should know all these tools. It's like I want you to be smart, curious and willing to piece things together and figure it out. And that is like the SMB job description right there.
Speaker 2:Sean, one thing that it kind of reminds me of is like, imagine if you could in, like, imagine if you could like. I mean, look, you've worked at a ton of different CX platforms. One thing that I find interesting and, sean, you work with Braze One thing that I'm kind of thinking of here is there are so many touch points where you could basically pass data about a customer to a model and say, hey, your job right now is to create the perfect tagline for this email. We've got an email. So, going back to the Michigan person, right, you're creating, either they log in and they like, open up a chat window and the chat window has context on who's like, writing in, like for a customer service platform.
Speaker 2:Or you're Sean and you're you know you're working for this, you're the head of marketing for this. You know jeans company, your jeans company, and you're trying to write the perfect tagline that you know slim fit jeans are back in stock. And it's like hey, like you. For each individual customer. Now you have this model, now you have this data, craft the perfect intro line to say, hey, you know that you really love are like back in stock. You can kind of like start to personalize every single touch point and make that sort of brand experience so much more cohesive, because you have so much information about a customer that you can pass to some sort of model.
Speaker 3:Is all I's going to to like figure out the exact tone, uh, keywords, structure for each user, so that, like it's not like right now, every ESP uh that I can think of and that I've touched in the last six months or a year has some sort of like AI powered, you know, subject line generator or text generator, like that's. That's become you know table stakes now, because every, every brand had to sprint to get AI features so they could say, look, we now have AI, even though they were already doing some sort of AI beforehand. So now it's all it's like subject line and text recommendations. But this would be like not even just hey, write a prompt, get a subject line. This would be personalized subject lines for each person to each person's preferences at scale, which is pretty sick to think about.
Speaker 1:Yeah, like providing a personal stylist to all of their customers.
Speaker 2:Yeah, yes, that is a great, great example, tisha. Along that same train, though, there are some things that AI maybe shouldn't do, and I'm curious, if you have any, you don't have to name names, any stories, any case studies, any anything around. Hey, like it can actually have pretty large consequences and like what are some of those consequences? What are some of those consequences?
Speaker 1:look like, yeah, and and actually I I'll talk about a couple of public stories which everybody should know about, so I can can name names, because AI does come with its pitfalls. Right, it can frequently hallucinate. There are definitely governance and privacy concerns, there's recency and contextualization issues today, and so the data can become stale that it's learning from. There's like all of these factors. The big story that had gone viral was about air canada. Right, like I think I'm assuming you both read it, but for those who didn't, uh, like I think the air canada chatbot told a person that, oh, you can't get a refund for your flight because it was like a very meant type of a reason and they just it just made it up like nowhere on the website that was there. That's like that. That's the hallucination. Right, and generative AI can do that. If they can't find a response, it's going to make it up.
Speaker 3:So can a disgruntled employee to say, hey, you can't. Just, it's not an AI problem, this is just a hiring problem.
Speaker 1:Yeah, and I think where Air Canada actually busted it up is they fought that and they said, no, sorry, our AI was wrong and they should have just given it.
Speaker 1:But they're like, no, we are not liable for our AI, which really didn't help their brand. Another thing which also came across in one of my former customers who were trying to sell our AI solution to it brought that up was and this was only, I want to say, a month or so ago where Patagonia, like one of their customers, actually sued them for you know, reading her buying history and then making it public or talking about, hey, because you bought X, now we'll do you know, we can sell you this. So they were sued that I wasn't told that you're like reading my data and you're going to use it to market. So I think, again, that's where governance and data privacy comes in, and because it had become sort of a, a code case. Now there are brands who are like, you know, we don't want to get into this unless we have all our checkboxes from, like a privacy and, you know, governance checkbox. So it's suddenly slowing things down on on that side and and definitely I think, requires some more work interesting this is super.
Speaker 2:I actually didn't know about the patagonia one, but it's that's really really interesting. Apparently, at value, it might not be legal in the state of California to do this, because, huh, that's super interesting. That makes that makes it. That makes a ton of sense. Wild, wild, wild, wild. Very cool, very cool. Yeah, that's that's. That's a great example. The other one that I like to kick around on the marketing side is the.
Speaker 2:There's a grocery store chain in New Zealand. I'm pretty sure I've brought this one up on the podcast before, but it's like my favorite example of this. It's kind of horrifying, but the thing that they did is they had a recommendation engine. So, going back to what we were talking about earlier, they had a recommendation engine that looked at a customer's buying history and then, after they had made a purchase, it would take the receipt and it would send them an email with the receipt. And it would send them an email with the receipt and it would say here's a recipe based on what you just bought. And one of the ones that went viral was a customer had bought fertilizer and flour and sugar and so it sent them a recipe for poison cookies and it was like hey, cook your fertilizer and boil it down into poison and then you've got flour and eggs and sugar. So make some, make some cookies and put some fertilizer in there for poison cookies and like it was just like uh, what, um? So yeah, definitely, definitely.
Speaker 3:I mean algorithmic stuff is is always had you know holes, right, like there's the old, the old one. I can't remember when this happened, right, but Target some point in time, like 10 years ago maybe, used its recommendation engine and realized that a woman was pregnant before she did because of the other things she was buying, and so they recommended like pregnancy tests or something, and she was young. So it became like a whole thing. I mean like meta, what you know. They have a toggle to prevent using like protected fields if you're in like housing and credit and lending and stuff like that. But a lawsuit a few years ago found that even though it wasn't explicitly being told to optimize for protected fields, it was still finding correlative things and so it was taking those into account. And it's like that.
Speaker 3:That, to me, is always going to be the scariest part about anything. That's like black box. It goes into an AI and just kind of happens and there's no, no governance, there's no monitoring, there's no way to kind of see what what factors it's taking into account is. It becomes pretty hard to unravel or even say like this wasn't correctly done, our mistake, because you it's, you can't't tell why it spit out, what it spit out, so I think that is pretty scary.
Speaker 1:Yeah, yeah, and I mean I think the way I know Meta and the providers are augmenting this is having you know large, you know data privacy teams or reviewing content and you know saying this is taboo, etc. And so I think that is kind of just coming back to CX space. As I talk to, you know some of these companies and the leaders. That's the recommendation, like the way you get around some of these hallucinations and you know AI kind of making stuff for you is you have targeted roles whose only job is to make your AI successful and powerful. So I think content creators, you know, or you know, ai knowledge base administrators those are roles that are coming up and you know it's their really their sole job is to make sure that the data that your you know AI is learning from is the correct one.
Speaker 1:You know. Obviously, you also get so much insight as you start using this. So look at those insights and you know whenever there's a mistake made, like, how are you going to correct that? And you know you you will need humans at this stage to to have that continuous improvement happen. So, yeah, I think, from a stage perspective, that's where we are and then it's only going to keep getting better.
Speaker 3:There's no reason to be afraid to jump into this. Yeah, I think that's a big. There's no reason to be suck compared to your peers. Or you can start to try and like see what you can do and learn and just like make it part of the way you work, and I think that that's going to be critical.
Speaker 2:Yeah, adapt or die no-transcript half of them, but I have one last one that really like kind of underlines your expertise. I'm curious about two things. One is what you see as the future of sort of like the human touch, because we talked about this before the podcast but, like you're like a big proponent of you know, sometimes a human needs to be in the loop. Like there are some experiences that, like an ai really won't be able to do. What do you think is the breakdown between kind of like the human side of customer experience and the what an ai is going to be able to do, and for the in, like you know, five to ten years in the future? And then, lastly, for companies that are starting to explore putting this as part of their tech stack, what advice do you give them on, like where to start?
Speaker 1:yeah. So I think, if you're looking at the future, right and I don't know if it's two years or five years or 10 years, but I do believe the human jobs and materially change, Humans will still be good at what we are good at today, Right, Like we have deep understanding of you know the context of our product or our company and you know we're good at multitasking and solving complex use cases. But if you are in this kind of space, whether it's CX or any job, I think and this is something Eric Schmidt told in his Stanford interview to actually programmers, right I think he was talking to a computer science class that it's important to think about. You know what you are doing at today, how you describe your job and if AI can take, like, just imagine the possibility, if AI is there and can take away your job, what all can it take away? So I think I would say in the CX world that that's important, right, Like AI is going to become better, faster, more cheaper, and so you need to now think about in that context, like you're not going to win with AI, but you know, from a human perspective, how can you be more creative, how are you going to assist AI. So I think, like the type of jobs that will be there in CX with material change. Actually, I looked at a service now like Vision Demo or something, and they were showing that, like in a couple of years from now, you won't need any support agents, they will just be. They will not be talking to customers, but they'll be monitoring and taking care of bots that are serving customers. Like that's showing this big dashboard and yeah, well, you know, I think everybody's sort of like dreaming of this, but yeah, as long as you keep in touch with what's happening and how you're involving yourself, I think that's kind of going to change, I think, in terms of also, so your job is then to give smart instructions right to your AI, to your company. You know, today, I would say, in sales and marketing, you know the whole role of you know how do you give good prompt. I think prompt engineering is a thing, right. Two-generative AI is a thing. So, like, how are you going to get better at? That is important.
Speaker 1:I think the advice I would give CX leaders who are just beginning to explore AI is, first of all, the time is now, you know, kind of like, as I said earlier, to be more dramatic, like it's literally you adapt or die, and then focus. Number two, focus on the problems you want to solve. Start with the ones that need speed and efficiency that's simple sort of your level one type of issues and then go to the next level where you can create more workflows and then think about how you can augment human work with AI. So, even if AI can't do all the complex decision making for you, where can it take you? So now your humans have a lot more information than if they had just started on their own. They had right.
Speaker 1:So how you're going to measure, you know your main KPIs is all going to change, right, Because humans are not now going to do more complex work. So your me time to resolve is going to like actually increase, you know, but it's important, kind of you look at that in that context. I will say lastly, there is a lot of smokes and mirrors out there. There are a lot of vendors, new players coming kind of promising a lot. That's probably not even possible. So the current stage is confusing, but I would say there's no reason to fear to at least get started right. So start with your phase one, phase two, and you will figure out on the way as technology evolves, so sort of stay ahead of the curve there.
Speaker 2:I think that's awesome. Those are two very good social snippets right there, disha. Thank you so so much for coming on. Really appreciated, you know, getting to chat with you. And yeah, before we jump off any sort of last things that you want to pump work in, our listeners kind of find you where can they hear more about anything that you want to sort of give a shout out to?
Speaker 1:Yeah, I mean, I am on LinkedIn. You can look for me at Disha Gosalia and yeah, other than that, I'm on LinkedIn, you can look for me at Disha Gosalia, and yeah, other than that, I'll be out there. I'll be seeing most people in CX and SaaS conferences. But, yeah, thank you for both of you for having me here. I think it was an engaging and good conversation. I'm always eager to learn more from others, so, yeah, I'll keep an eye on what else you guys have coming up there.
Speaker 3:Thank, you awesome, great chatting with you.