AI That People Actually Use: Zoher Karu on Personalization, Trust, and Building AI at Scale
Everyone is talking about AI. Far fewer people have spent decades actually building AI and data systems inside some of the world’s largest organizations.
In this episode of Signal & Noise, Brett House and Rio Longacre sit down with Zoher Karu, Head of AI at Taelor, to separate AI hype from what it actually takes to create measurable business value.
Zoher brings an unusually broad perspective. His career has taken him through McKinsey, Sears, Citi, eBay, Blue Shield of California, and now Taelor—an AI-powered men’s clothing rental service attempting to combine machine intelligence with human styling expertise. Across those very different businesses, Zoher argues that the same lesson keeps resurfacing: the technology is rarely the hardest part.
The conversation starts with one of enterprise AI’s least glamorous truths: bad data doesn’t disappear because you put an LLM on top of it. As Zoher puts it, AI can simply give you “bad answers faster.” Data governance, business processes, organizational knowledge, and change management remain foundational.
From there, the discussion gets practical. Zoher explains how Taelor is attempting to teach machines something surprisingly difficult: taste. Matching clothes to a person requires understanding not just size and style, but weather, occasion, context, individual preferences, previous feedback—and even whether two individually appropriate pieces of clothing actually work together.
That becomes a window into a much bigger conversation about the future of personalization. Generative AI dramatically expands the amount of customer context businesses can process, how quickly they can respond to new signals, and the number of individualized experiences they can create. Instead of choosing among three versions of an email, brands could theoretically generate an almost infinite number of variations for individual customers.
The discussion also tackles the uncomfortable economics of enterprise AI. Companies are spending enormous amounts on models, infrastructure and tokens—but are they actually redesigning the business processes required to capture the ROI? Zoher argues that automating pieces of an existing workflow may deliver incremental efficiency, while the much larger opportunity comes from asking whether that workflow should exist at all.
Finally, the conversation explores what may become one of the most important issues in enterprise AI: context. Agents can access data, but data alone doesn't contain all the rules, judgment and institutional knowledge humans use to make decisions. Capturing that tacit business knowledge—and making it available to AI systems—could become a critical source of competitive advantage and intellectual property.
In this episode:
Why dirty data can derail even sophisticated AI
Why AI transformation is really organizational transformation
The gap between AI spending and measurable ROI
Why simply automating existing processes isn't enough
How AI is changing personalization and recommendation systems
How Taelor combines human stylists with machine intelligence
Why context and business knowledge matter as much as models
Whether AI is actually eliminating jobs or simply changing them
Why change management may be the biggest barrier to enterprise AI
The continuing importance of human judgment in increasingly autonomous systems
The companies that win the AI race may not be the ones with the most sophisticated models. They may simply be the ones that figure out how to build AI that people actually use.
#ArtificialIntelligence #AI #EnterpriseAI #GenerativeAI #AgenticAI #Personalization #CustomerExperience #DataStrategy #DataGovernance #MachineLearning #DigitalTransformation #AITransformation #ChangeManagement #MarTech #RecommendationEngines #FutureOfWork #SignalAndNoise #Podcast
Read the full transcript below.
Brett (00:01.123)
Hey everybody, welcome back to Signal and Noise. This is Brett House, joined by my co-host Rio Longacre, and today's guest is Zohar Carew, the newly appointed head of AI for Taylor, which is a Silicon Valley-based menswear clothing rental subscription service. So I thought that was really interesting. I certainly did a little bit of homework on those guys. And tech startups, so it's just seed funding, I think, at this point, right?
Or or are you in it well into series A or series B funding at this point?
Zoher Karu (00:32.17)
No, it's pre s it's pre series A right now, but you know, growing growing quickly. So
Brett (00:34.861)
Yeah.
But it it was founded a while back though, right? Fa I think I I read that it was founded in twenty twenty.
Zoher Karu (00:43.54)
It was founded a few years ago. Yeah. I know the f founder or one of the founders actually. she used to work for me many years ago, so that's how we kept in touch. So
Brett (00:51.241)
there we are. Yeah, I think you may be referring to Anya Chang, right? And Phoebe Tan. so so yeah, so and for those that don't know who who Anya Chang is, she's a published author. I know, I know that for a fact. And she worked at Meta as a product lead where she helped build and launch Facebook and Instagram shopping, and has been a TED X speaker, right, and written a couple of books on on career development and in in
Zoher Karu (00:54.766)
That's right.
Zoher Karu (01:18.614)
Yeah. No, she's she's great. she and I met at Sears Holdings when I was there. So sh I hired her on hired her on my team there and we've kept in touch over the years, like I said. So
Brett (01:24.616)
nice.
Brett (01:29.943)
Yep, and so so Taylor for those that don't know also uses they use a mix of artificial intelligence and human stylists to curate everyday outfits for busy professionals. Sounds like something I need for a monthly subscription of ninety-five dollars. Right. So certainly interested to hear more about that and what you're doing from a data and AI perspective on that side of things. But before Taylor, you were the VP in chief data and analytics officer at Blue Shield of California.
where you led a global team of a lot, 200 people, 200 data scientists and engineers, responsible for things from like personalization, analytics and AI. and and so so we really think that that trajectory from sort of large company to kind of founding your own consulting business into kind of an early stage startup is a really interesting one and we'd love to hear a lot about that. And then prior to that you were you were at City, Citibank.
where you ran a a AI and analytics organizations across seventeen countries. you were a VP of in and chief data officer at eBay. and you also worked at Sears back in the day. So certainly an interesting career trajectory, an iconic American brand, right? And you just recently got recognition, I don't know how recent it was, but data IQ's top one hundred global data leaders. you got a PhD from MIT, right, in electrical engineering and computer science and
Rio (02:38.466)
It's an iconic American name, yeah.
Brett (02:52.595)
And it's along with Carnegie Mellon, a bachelor's so you definitely are better at math than either Rio or I were liberal arts majors. So I I did my best, but I realized I was stronger in other areas of of academics.
Rio (03:07.67)
I was a computer science major for my first year of college and then just I I I I was told that I should find a new profession.
Brett (03:10.337)
Yeah.
Zoher Karu (03:10.742)
Good thing, okay.
Zoher Karu (03:15.062)
Yeah.
Brett (03:15.361)
Yeah. So so we thought, hey, this is this is this will be a good conversation. I think you've got you know, sort of AI success and f failure stories and data strategy stories across a bunch of different industries. And so I think, you know, we'll dig into that and certainly get your perspective and so Zarya, welcome to Signal and Noise. Pleasure to have you.
Zoher Karu (03:34.478)
Thank you so much. Yeah, thanks for having me. I appreciate it.
Rio (03:37.27)
Did Brett miss anything in your background? Did he get anything wrong? Anything you want to add?
Brett (03:37.483)
So so
Zoher Karu (03:41.977)
No, I I've done sort of a variety of different companies, a variety of different industries. Frankly, the common theme has sort of been interesting ways to use data. you're right, I did do a technical education, but my first job, I sold my soul and went over to McKinsey, did consulting for many years, and then several smaller startups. in fact, some of the startups are also interesting. They were tracking customers inside of retail stores, using video cameras or analyzing phone conversations in call centers.
Brett (03:58.242)
Yeah.
Zoher Karu (04:10.13)
but from there I went to bigger companies like the ones you mentioned, and running data and analytics teams and working across the enterprise to help people take advantage of all of the data they have to really drive business performance.
Rio (04:23.884)
Sold you Sultan McKinsey by selling expensive PowerPoints. I I'm a former consultant as well. I guess I still kind of am, but but I I definitely know the business. I my first first consulting gig was a Cap Gemini consultant.
Zoher Karu (04:28.238)
Yeah. Okay, there you go.
Brett (04:29.226)
Mm.
Zoher Karu (04:36.29)
Okay, got it.
Rio (04:38.37)
But Zohair, it's great to have you on here. What we'd really liked is not only this like work you're doing specifically in AI right now with Taylor and other companies that we definitely want to get into that we thought was very interesting, but I think it's cool that your career spans healthcare, banking, e-commerce, retail, and now this this fashion startup, right? And I I I think you look at those things and you think, well, what are they what do they really have in common? But like across all of them, at least the interpretation Brett and I had is you've been using data and now AI and machine learning and other
Zoher Karu (04:44.834)
Yeah.
Rio (05:07.65)
Other technologies to really make better decisions, create better consumer experiences, right? So I kind of saw that as a common thread. And like looking at AI and lessons learned and how it's being applied, we thought this would be a cool one to talk about like how like where are you seeing successes? So maybe just start there, like in your current role, like looking at like AI that's being used by people today, that's but like out that's not just in these pro pilots, right? I I I think it it was a death by pilot, right, with AI the last few years.
Zoher Karu (05:13.368)
Right.
Zoher Karu (05:37.121)
Yeah. Yeah.
Rio (05:37.452)
Love to maybe start there. What are you seeing as succeeding?
Zoher Karu (05:40.909)
Well, I think, you know, people have basically realized the same problem that they've realized decades ago that the core of the issue is largely your data. Where is your data? How clean is your data? Do people know where it is? Do people know what it means? because just because you point powerful tools at it like some of these large language models et cetera, you just get better you get bad answers faster if you're not careful, right? And so
You need to understand what you're working with. Certainly AI is transformational, no doubt, and able to do a lot of things very quickly that used to take, you know, weeks or months sometimes. But I think people have sometimes tried to jump jump too far in that neglecting that you know data governance is sexy again, right? and you really need to think about what you're working with. And
You know, all of these large language models that you see out there, Chat GPT, et cetera, they're trained on all public data. Public data is probably only I don't know, I I don't know the number exactly, but maybe ten percent of all the data, right? There's a lot of data behind corporate firewalls that and there's long standing issues of silos and duplication and single source of truth and things like that. And so some of that needs to get addressed. it's not that it all needs to get addressed before you can use AI. You need to be
iterative in its nature, but you can't just neglect and say, well, we'll just use Chat GPT and all of our problems are going to be solved, right? and so I what I've those who have succeeded are the ones who really have taken the time to understand their data, understand their business processes, and not you know just try things for the sake of trying them. I mean, in a way it's okay to dip your toe in the water, but things won't scale otherwise, right? And and
Largely it runs into the data problem, the business process problem, but also a big one is a change management problem because you're asking people to do things totally differently.
Brett (07:45.314)
Yeah.
Yeah, and that and that's probably a big part of the the the hardest thing to solve, right? Is how do you get the organization to change how they behave to to better operate some of these systems. Right. And and so what do you think when you talk about governance, right, what does that look like at a pre seed startup versus these very large, very you know, arguably siloed or matrixed organizations that you worked at previously?
Zoher Karu (07:52.237)
Yeah.
Yeah. Yeah.
Zoher Karu (08:12.344)
Yeah. well, you know, in the big ones, everyone seems to want their own version of the truth, right? They and and it's basically human nature. They'll use whatever data they need to accomplish the task in front of them, right? But data is one of these things, like the analogy I use it's it's like blood in your body. Yes, there are multiple organs. Each organ has a different function, but they all use the same blood, right? There's not multiple versions of the blood, there's not dirty blood, the blood's not leaking, like whatever, right? And and so
Brett (08:35.169)
Yeah.
Brett (08:40.246)
Yeah. That's a better analogy than data is is oil, right? I think that's a much better analogy, right? 'cause 'cause data's largely the operating layer and without like a healthy data ecosystem, like a healthy blood, you know, the the entire system breaks down and all the different functions attached to it, right? Right.
Zoher Karu (08:41.058)
That's important. and
Okay. Yes. Data's the lifeblood. That's right.
Zoher Karu (08:52.782)
That's right.
Zoher Karu (08:59.456)
Yeah. So in a s in a smaller environment, it's a little bit easier to control some of those things because you don't have all of those crazy separate silos. But what is challenging about a smaller environment is setting things up for scale. Yes, you can set things up and capture things in a way that sort of solved the problem of today, but you if you don't think about it a little bit further ahead, you will outgrow that methodology very quickly, right? and that is the
That's the that's the challenge behind putting that through. So it's a balance of knowing what you need now versus what's it without getting into sort of bu bureaucratic overkill, what's the way it needs to be, you know, in the future. Right. and so trying to run run that balance is part of the art, I'd say.
Brett (09:42.4)
Yeah.
Rio (09:46.623)
Yeah. Well it's interesting you say that. I there's been a lot of talk that we've had in NISP pod and and Brett, I wrote that article recently about how AI is impacting the marketing organization, creating new roles. And I like the one thing that we've heard again and again and has become really clear working with clients in AI, right? Is that more than anything else, AI really is org transformation. It's change management, as you mentioned. It's changing how you how how the org operates and and adopting AI necessitates that. So
We've seen a surprising am amount of that. And I think initially you look in professional services, a lot of the AI billings were actually AI training, funny funny enough, but I think the last year or so that's really pivoted and a lot of AI work now was actually change management. Love your thoughts on that.
Zoher Karu (10:30.306)
Yeah, for sure. I mean, your, you know, companies before all these technologies, all these companies existed. They were you know, moving right along, doing their thing, maybe not as efficiently or as effectively as they could, but regardless, they they were operating, right? And so now you have a new tool and you're like, okay, now what do we do? There are basically two, in my mind, two categories of where to start, right? One is you have an existing process.
Some parts of it are a little bit manual, like, okay, a human has to read this claim and type in some numbers into this database. So, like, okay, you can start automating some of that with AI and take existing processes and streamline them. That will yield some improvement. it does of course run into the change management problem because now you're changing people's jobs. Like, okay, I used to be the number typer, now what am I doing again? Right. and so that that does run into some challenges, but that that will get you some improvement.
maybe whatever, 10, 20, 30% efficiency improvement. or I'm gonna, you know, summarize the notes after a call center call for me instead of a call center agent having to type it up. Great. But the real value of AI in my mind is changing your business processes entirely. So on one hand, you could say, yes, we're gonna help this agent type up the notes after a call. Maybe the real question you should be asking yourself is why am I taking a phone call to start with? Right. and so, and really thinking through how your business operates
given the capabilities of some of these tools, right? And so that's I think that's what's going
Rio (12:00.642)
Yeah, if you're if your job is like the c the call note taker or the number inputter, then y I think you're you're probably gonna be you probably will be looking for a new job in the near future, that'd be my guess. Yeah. I mean it's yeah.
Zoher Karu (12:06.541)
Yeah.
Zoher Karu (12:12.364)
Yeah, that's right. Yeah, yeah. Well, yeah, those those agents, I mean, it does save them time. You know, you hang up from the phone, but every time you call a call center, they'll type up some notes after the call, right? And so you can pre-type some of that stuff, get some efficiency, yes. And you know, a lot of these AI starting points are labor takeout, largely, right? and y you know, sometimes people aren't realizing the ROI from AI because
Brett (12:28.8)
Yeah.
Zoher Karu (12:35.192)
They're not doing the labor takeout, right? Like they're yes, they're spending a lot on AI and they're doing all of these things and the costs went up, not down, and you know, that's where a lot of CFOs are, you know, raising question marks basically.
Rio (12:37.398)
Right.
Rio (12:45.206)
Are you seeing jobs getting like eliminated or being automated away by AI? I mean I mean I'm just not seeing it. I mean in either in anecdotally or or th in the numbers. It's just not
Zoher Karu (12:53.108)
Yeah.
Zoher Karu (12:57.14)
Well, okay. There's certainly a lot of headlines in the news, you know, layoffs in the tech especially here in the Bay Area, layoffs at a large large large tech companies and the party line being given is because of AI, we don't need as many people, right? and so they are certainly saying that. Is it really happening? It's a little hard to say, like you said. and there you know, maybe that's just the the the cover story. But it is true that, you know, coding and
you know, whatever, even marketing, material creation and all of these things have gotten a lot faster. That is true, right? But is it really eliminating or just reshuffling, you know, what people are doing? I I don't know. It's like every new technology disruption does eliminate some jobs, but it also creates other jobs. Like who's doing AI governance now or, you know, security or whatever else, right? And so yeah. Yeah.
Rio (13:49.187)
Yeah, it creates jobs and it changes jobs. Like the like the way that you know roles stay there, but like they might just be doing certain activities differently during the day, using AI for certain things. Even like every technology does that to your point. Like even when when when computers came out or word processors and computers over typewriters, even typewriters over writing by hand. I mean, still people still act as journalists and and do things, but clearly like the the the output of work they're able to produce has gone up quite a bit. I'd argue the quality's probably gone up too.
Zoher Karu (13:58.894)
Yeah.
Rio (14:19.423)
but those jobs still exist and maybe even more people do it in some instances.
Zoher Karu (14:19.864)
Yeah.
Yeah. I think, you know, the breakthrough in AI, which we may even be already on the cuspon, frankly, is AI, or at least recent version of AI rather, is is trained on existing data, right? You give it a bunch of existing data, it learns patterns and it goes from there. But can it really make brand new patterns from scratch, right? Can it really connect dots that were never connected before? And that's why I still think there's the need for human partnership, if you will, right? with with technology.
Will AI reach whatever generative AI type or general AI, sorry, general intelligence? I who knows, right? and will it come up with patterns on its own? Okay, maybe, right? You can do it by trial and error, perhaps, and you know, so yeah. Yeah.
Rio (15:04.054)
Yeah, I it's it's pretty creative though. I mean, like you ask it things, I mean it's and even like hallucinations are an output of creativity, right? I think you could make that argument, right? That it's it'll make up some wild stuff sometimes, right? It'll lie and then like you're like, Wow, that's hallucination. You're like why'd you say this? And say, Well, I said that because you wanted this and I couldn't find it anywhere else and this is the best thing I could come up. It'll actually it's it's I think that's actually wild, right? It's actually
Zoher Karu (15:14.572)
Yeah, yeah.
Zoher Karu (15:23.854)
Yeah. What what is completely what is completely fascinating to me is okay, back in undergrad days, I programmed neural networks, right? Like programmed one to play tic-tac-toe. And I I know what's involved in something as basic as that, and it's not so easy, right? And now I watch these things spitting out entire paragraphs of coherent information with PowerPoint and whatever else it spits out. It's like magic to me, frankly, right? Like I can't believe it. And all it's doing is predicting the next word.
And the and the and nothing else, right? and yeah. That's right.
Brett (15:54.069)
Yeah. Yeah, it's fundamentally a sti a stochastic system, right? That's predicting the next word. It's it's just a super advanced prediction machine and Rio talks about this all the time. Whether or not human brains are really just that but but firing
Rio (16:06.594)
Yeah, our are human brains just prediction machines with a lot of compute and a lot of inference and a lot of storage, right? I mean so
Brett (16:10.815)
Yeah.
Zoher Karu (16:13.366)
Yeah, that's right. well the number of neurons is still go ahead.
Rio (16:15.904)
And the illusion and it and the illusion of consciousness. I mean you could make you that's one I don't buy it, but like that is one that's one argument.
Zoher Karu (16:24.556)
Yeah. Well, you know, what's the definition of consciousness or what's the definition of, you know, intelligence? These are all, you know, I think we're way past Turing test kind of thing, almost like I don't know what it is now, right? So yeah. Yeah.
Brett (16:35.381)
Yeah.
Rio (16:35.468)
Yeah, we blew past it. Well, I think I think that's what's so w so crazy, Zohir. Like we blew past the touring tests, you know, initially a few years ago. Then I think list last year, like it beat it eighty percent of the time, right? When with when you had like multiple parties. So it's clearly these things are able to pass off as humans, right? And people but yet people it hasn't changed the world. Like, you know, we haven't had nuclear Armageddon, people's jobs have like it hasn't changed the world as radically as people maybe thought it might be when we had these capabilities and the can keeps being kicked down the road. like
Zoher Karu (16:47.266)
Yeah.
Rio (17:04.758)
It'll happen when we have AGI or when we have ASI, but it already had but s the so the wholesale changes haven't happened. But I think day to day within jobs like there are some really big like the way I work has changed.
Brett (17:18.494)
Yeah, yeah for sure. Yeah.
Zoher Karu (17:19.692)
Yeah, for sure. Yeah. Like, you know, creating show notes for your podcasts or whatever, right? And like it it can spit it out pretty quickly. Right. So
Brett (17:27.444)
Yeah. So so l so tell us a little bit about what you're doing with Taylor. I'd love to hear sort of an i an actual example. And it's a it's a recommendation engine. It's trying to understand taste, right? It's choosing from huge, you know, catalogs of of options in terms of of the cl you know, so tell us what that model looks like. you know, the service that you're delivering, you know, a box of tailored clothing s you know, for for professionals, I guess, right? And then and then some of how the technology is powering.
Zoher Karu (17:46.296)
Yep.
Zoher Karu (17:49.634)
Yeah.
Zoher Karu (17:54.072)
Yeah.
Brett (17:57.493)
that decisioning process so that when I get something in the mail it's actually tailored to my needs and it seems like t it's trying to to kinda lean into this notion of of taste and how do you actually predict what's most likely gonna appeal to an individual's taste. I mean how does that whole can you can you explain that? How that works? What is the t what is the technology behind it?
Zoher Karu (18:20.352)
Yeah, for sure. I mean it's a actually a really hard problem, in terms of trying to match clothing with people, right? and be if you just if you if you th first of all, Taylor, like you said, is a men's subscription rental service. It's designed for people who don't have the time or the skills or the inclination to really want to figure out what to wear every day, right? so and
Brett (18:48.362)
Yep.
Zoher Karu (18:50.314)
Or, I'll just wear what I wore in college or as opposed to really, you know, what what I want what I should be wearing or could be wearing. To really you know, it's really about elevating your own self confidence and self esteem and helping you achieve whatever you wanted to achieve, whether that's a different job or a date night or whatever, right? But clothing is a piece of that puzzle, right? And so that's what's behind it. And you write it's a subscription service where you sign up
For a monthly subscription, every month you get a box of clothes that were selected with a combination of human stylists and machine algorithms. And if you like you could buy some of you could just re or you you don't like them, you you wear them, do whatever, return them, you get a new box, right? and so it's it's an interesting concept.
Brett (19:33.771)
Yeah. And and how is how is it personalizing? Because I mean I it's that notion of like the human stylist that but being codified. Like all of the things that might be seem like a gut or just something that you've y you've you've been in this industry, you've been a clothier or you've been a design fashion person that's been trained, you've gone through some sort of apprenticeship, and you sort of build all of this knowledge, but some of it's just kind of finger in the wind, you know, it's kind of like the artist's sort of
Zoher Karu (19:48.514)
Yeah. Yeah.
Brett (20:02.236)
I I can't explain exactly why and all the all the permutations of that goes into making this decision of why this is right versus why this is right. but how does how does that happen from a predictive analytics perspective?
Zoher Karu (20:07.566)
No no, it's
For sure.
So it's a it's a it is a hard problem as you're alluding to, right? Because not only do you have to understand the clothes themselves, okay, what is this piece of clothing? Is the what's the material? Is it stretchy? Does it run long? Does it run short? Like does it like and the small is not a small is not small as anybody who's gone shopping knows, right? and so it's it's very complex and just getting all the right metadata on the piece of clothing, like what
Temperature weather is this thing good for? What this, what that, right? And capturing all of that data about every piece of clothing is hard. so I mean
Brett (20:47.316)
Yeah. And it's an interesting concept to think about data the metadata associated with clothing, right? Which we we're I don't think you you'd think metadata associated with video and content, you know, in digital ecosystems. But to actually start to apply that same sort of digital logic to clothing itself, it's a it's an interesting concept.
Zoher Karu (20:52.586)
Yeah, that's right. Yeah.
Zoher Karu (21:00.703)
No, there's
Zoher Karu (21:06.092)
Yeah, there's a lot of things to capture and some of it are not that straightforward captures. Like sometimes people are like we're taking out a measure tape and just measuring the chest size ourselves, right? As opposed to just taking whatever is comes in the manufacturer's specs or something on the clothes. And so we're so you're capturing all this metadata about clothing, fine, which is hard in itself, like you said, but then you also need to capture all the metadata about a person, right? Who is this person? What is their occupation? Of course, what is their height, weight, you know.
Brett (21:30.58)
Yeah.
Zoher Karu (21:36.015)
size kind of information, but what what what kind of tastes do they have? Like are they conservative? Are they streetwear? Are they edgy? Are they formal? Are they casual? Are they like there's this what is how do you try to understand somebody's preferences? And the third dimension that makes it even more complicated is the context matters.
Like, you know, Rio going to a party or Rio relaxing at a barbecue on the weekend are not the same person, right? and so how do you think about what you need for each of the different contextual things? And even the weather is is a simple example of context. The weather changes, right? And so what do you really need? and so you take all of these things together and it turns into a massive matching problem in terms of I've got all this metadata about this human.
And we get that by the way by getting people to start by answering a style quiz, right? When you sign up you answer Yeah, so yeah.
Brett (22:32.379)
Yeah, yeah, that was my question, is is where are you gathering is
Rio (22:33.665)
How many questions are in the qu in the quiz? How long's the quiz? Out of curiosity.
Zoher Karu (22:37.396)
It's like ten minutes, fifteen minutes. I mean, I can't ask I mean you of course you'd love to ask more, but you can't ask everything other people get bored, right? and so
Rio (22:39.179)
Okay, it's not that bad.
Brett (22:39.625)
Okay.
Brett (22:43.015)
Is is it is it pulling stuff from public profile as well around the individual where it's sort of r no. It's all it's all based on that quiz and that data entry from the consumer. And then do they have the option or ability to customize and personalize further? Meaning, you know, it's sort of like nudge the system to say, Hey, you got this wrong or this is off.
Zoher Karu (22:47.776)
No, we don't pull anything. We start from scratch, although
Yeah. You know.
Zoher Karu (23:01.62)
Yeah. Well, y for sure, right? and so the way it operates is you start with a style quiz, because we do need a starting point. The cold start problem is very hard, right? Like we have no idea even how tall you are, right? and so you you need you need to know basic things like, Okay, what size do you typically wear, what brands do you like, you know, what is your occupation, where do you live?
Brett (23:12.563)
Yeah.
Yeah.
Zoher Karu (23:26.666)
You know, it would show you a bunch of pictures and say which of these pictures appeals more to you, right? And so you get metadata from that, like, okay, this style looks better to me. and so you're trying to build out that pattern of what is it that this person appears to like. And then you make some, of course, historical judgments like, okay, you're a software engineer working in the Bay. Okay, I can probably guess the types of clothes you wear, but you know, so you you make some starting assumptions. But as you said, if there is a feedback loop.
Brett (23:35.038)
Yeah.
Zoher Karu (23:54.541)
continuously in terms of two ways you can sort of get through go through the catalog and sort of favorite some items if you will and we start picking up signals from that like that shirt you you like that shirt or jacket or whatever. You can also get signals from after you receive the box where encouraging people to leave reviews about what you received. Okay, this I loved it. This ran
you know, felt tight on my shoulders, this, that, whatever. And we tr we're trying to understand what it what it is that can alter recommendations in the future. Right. And so to the point we were making earlier about how to use AI, part of it is, okay, let's just take the no machine version. A human stylist tries to understand you, picks out clothes. Like you could walk into Nordstrom and I want a stylist to help me. They'll search around, they'll find some clothes, you go try them on, whatever, and they'll try to iterate, right? and
The art behind this, honestly, is that it
Rio (24:50.957)
And that's an underrated service by Nordstrom, by the way, for those who've not done it. Is that like you go and you can actually they most of them are pretty pretty knowledgeable. They know their stuff. You can actually go and ask them and they'll they'll they'll recommend stuff for you and it's usually pretty good, so
Zoher Karu (24:54.595)
Yeah.
Zoher Karu (25:02.176)
Yeah, they're they're pretty good. so so the the challenge here is it's a that's a labor intensive process and there definitely is a variation in skill. I mean they're all pretty good, but there's some are better than others, right? And so and so and and so that's like
Rio (25:14.563)
Huge variation, yeah, for sure.
Brett (25:15.709)
Yeah. Rio's a Nord a Nordstrom brand loyalist, I think, is is what yeah, me too. Yeah.
Rio (25:19.767)
I love Doorsstrap, yeah. There's there was one right up the street which helps too, but it's convenient.
Zoher Karu (25:24.814)
no, it's like, you know, are all the desi interior designers equally good? No. Okay, some are better than others, right? And so, but how do you capture the essence of what makes a good designer or a good stylist? I mean, there's a sort of you're bad and you're good sort of black and white, probably, then it gets into the gray of like, okay, is this one better than this one? Well, then it gets a little bit, you know, gray. Yeah. So yeah. Yeah. I mean, so you you know, all you
Brett (25:45.513)
Well yeah, then you have you have things like pattern pattern and color matching based on skin tones and and and all this other stuff.
Zoher Karu (25:53.891)
What you want to do is try to scale at least a a good stylist, right? and so we, you know, one of the measures of success at Taylor is how fast the styling is getting done, right? because if machines are assisting in augmenting the styling process, then you know, that's that's one definition of success. And of course, customer satisfaction from that. Like we're not picking random items and sending them to you, right? And so so that's the that's the feedback loop.
Brett (25:58.773)
Yeah.
Brett (26:17.726)
Yeah.
Zoher Karu (26:21.42)
But can we continue to grow the customer base without adding any more humans to the process, right? By by augmenting them in some way. And so that's the, you know, let's let's take humans and what they were doing. They were filtering through inventories of clothes and picking out items and and make that easier algorithmically, right? Like, okay, there's the basic stuff like let's just automatically filter for the right size and the right weather, fine. But then you can get into more sophisticated okay, okay, what's the, you know, vector matching.
Optimal between these metadata of clothes and the metadata of you, right? And start sorting and ranking and saying, okay, these are likely to work based on what you've liked before, based on the you know, information you've given me, all the context that I know, whatever. And so we're picking that off and augmenting the stylist choosing clothes. Now you take you can take it one step further and say, don't just you know, filter for me, which is fine, but actually, you know, select for me.
Like make me a sh make me a shipment, right? and you can maybe edit it after that, but that saves a huge amount of time if you can just automatically put together a shipment. You know, what in addition to the complexity I mentioned earlier about the metadata of the clothes and the metadata of human and the context. There's also does this shirt even go with this pant, right? Like there's met there's there's metadata between the two, right? and so just because you understand the shirt and you understand the pant doesn't mean should wear them together, right? and so there's another layer of complexity as well. And
Brett (27:37.266)
Yeah.
Zoher Karu (27:45.913)
You know, this is a it's a it's definitely a challenging problem, one that I think certainly has a big market, one that needs solving, but it's it's not it's not so niche that you can identify the exact customer base who would benefit, right? Lot lots of people would benefit. And so that's been part of the challenge is really thinking through how do we find the people who want this or need this service, right? So
Brett (28:08.242)
Yeah, that's an I an ICP an I C P challenge, right?
Rio (28:09.313)
Well yeah. W question, how much of this the success of this like based on what we've seen so far is just the recommend the the actual recommendations for them to close themselves versus the experience people have actually interacting with the virtual stylist and and how good that is. I mean, or is it a combination ultimately that's gonna drive s success?
Zoher Karu (28:12.045)
Yeah.
Zoher Karu (28:31.094)
So you said how the two things are the virtual stylistic Yeah.
Rio (28:34.743)
Just a clothe just just a clothing, right? Like you know, and and then and then and versus actually like the ability to interact with the with the virtual stylist, like g talk to it and like work through kind of recommendations, or is it or is it a combination of those two together?
Zoher Karu (28:49.006)
No, it's definitely a combination because, you know, if it's just the first one, choose your, you know, choose the clothing. It's almost like going to Nordiswim.com and trying to choose yourself, right? and so it you need you need some guidance. like if I just said, you know, go pick out some cabinetry for your kitchen, like, all right, well, it would help to have some guidance, right? So you need to be able to interact. yes, it can be human, it can be machine, but you need
a little you know, you need some interaction where you know that there's some expertise expertise on the other end, not you know
Brett (29:19.772)
And and there's a conversational agent aligned to this as well, 'cause you talked about the quiz. Are you actually interacting with a virtual stylist?
Rio (29:26.721)
Or is it just giving you like these are clothes?
Zoher Karu (29:26.742)
So it's right now it's well, no, there's a there isn't interac well it's not shopping in the sense that, you know, go out and pick your own clothes yourself, right? although we are introducing some maybe self-pick options. But it is interacting often right now it's like with the human stylist, like, hey, I'm going to a wedding, I need the that that, like but you can imagine that quickly turning into a an AI conversational bot as well.
to translate that into your needs. Right. And so we are we there's sometimes a little bit of interaction, frankly, but largely it's, you know, clothes show up. Right. And and you could book, you know, I I've been a customer for, you know, a little while and I'm pleasantly surprised. Like, I would not have picked that if I was in a store looking around, but then I put it on like, okay, it looks pretty good. Okay, fine, I'll buy it. Right. and so the part of part of the
Brett (30:16.498)
Yeah. Yeah. So you guys you guys are it sounds it's early stage where it's you take the quiz, that's basically the data input. You're not actually interacting with a chat bot where you're saying, Hey, and I'm assuming part of this process is uploading a photo, giving your size dimensions, your weight dimensions, all that sort of stuff so that it can match g the color matching is probably the biggest problem, right? Style is one thing, but you know, if
Zoher Karu (30:29.944)
Yeah, of course. Yeah. Yes. Yeah.
Zoher Karu (30:38.866)
Yeah, and you can introdu you you can do yeah, skin tone matching and you can say I don't wear pink and I don't wear purple and whatever you can say, those kinds of things and it'll avoid those as well. Although like I said, we are trying to push the boundaries a little bit, right? and you can I mean we use large language models.
Brett (30:45.084)
Yeah.
Rio (30:52.269)
Pink I'm gonna know. Maybe purple purple purple I could say maybe. If it was if w it was done right.
Zoher Karu (30:56.27)
Yeah. Well no and then you
Brett (30:59.228)
Yeah, hey, there's certain certain certain people that certain yeah, certain men can pull off pink really well. Yeah.
Rio (31:00.683)
I actually I do have a pink shirt, now I think I have a pink dress shirt. It goes great w goes great with grey pants.
Zoher Karu (31:05.944)
That's right. and so
Brett (31:06.812)
Yeah. See Rio Rio may not Rio may not be in your ICP because he is his own personal stylist. In fact, Rio's given me advice. He's quite he's quite for forward with that. And it's actually ch I'm like, Rio, that's a good point, right? 'Cause I just don't cognitively think enough about it. But Rio's you should buy you might want to try this, you might want to do this, you might and I'm like, okay, those are these are good, you know, but you have to be practiced at it. And Rio's definitely he's well experienced.
Rio (31:07.595)
How it so
Zoher Karu (31:22.851)
Yeah.
Zoher Karu (31:29.548)
Yeah, yeah.
Rio (31:31.299)
So but so so so but Zoe, how is this different? So I know there was a bunch of these like subscription services like Trunk Club and a bunch of others that started maybe around the same time, maybe a little earlier. Seems like that was a big thing and it kinda fizzled out. Like what is what is Taylor doing that you think is gonna is an improvement upon the previous attempts to do this or a new way of doing it altogether?
Brett (31:38.309)
Yeah.
Zoher Karu (31:40.742)
Yeah.
Zoher Karu (31:50.829)
I think a lot of them were largely you know, like people places like Rent the Runway, for example, the women you have to choose your own clothes, right? And you get And so you can get access to much of clothes you wouldn't normally get, but you're doing the work yourself. some of these other ones are very, you know, human oriented, right? and not as algorithmically oriented as we're trying to achieve, right? so that's a difference. Plus what's also turning into interesting is that we're collecting a lot of data.
about which customers like which types of clothes. So brands are sort of also very interested now in the sort of the aggregate data monetization type story. Like, you know, yes, North Face knows what North Face sold, but does North Face know which type of customers really love their clothes, right? and so we have a piece of information around what you know who is literally who is actually buying the stuff or renting the stuff and what are they liking about it or not liking about it. So
That's another angle that I think we're trying to put together.
Brett (32:50.268)
Yeah. It's a it's a buyer profile that you're building based on these on these personalized sort of recommendations that you sell that you sell to Yep. That's interesting. And
Zoher Karu (32:56.064)
That's right, right? Like you know
Yeah. So that go ahead.
Brett (33:02.886)
yeah, no, no. And it sounds I mean, like if if you think about it from a business problem perspective, and and we talked about this we've talked about this in a couple of pods, how are you removing friction from a process that that has friction built into it, right? And so when when you think about ICP, you know, you're kind of making the bet that men don't like to shop. right? You know, and there's some friction there in terms of the going and spending time
Maybe men outside of Rio. Rio loves to spend time in Nordstrom You know talking to stylists. Okay, so Rio Rio's a d I could have sworn he was there on a weekly basis based on his recommendations. But no, but but point you know, like the the friction of actually having to go to stores, spend time, right? We're all we all have busy lives, mortgages, you know.
Rio (33:35.304)
I go t I go twice a year. I mean I mean, I know what I'm gonna get.
Zoher Karu (33:38.762)
Their annual annual sem semi annual sale. All right. Yeah.
Brett (33:54.255)
because I'm assuming the ICP is probably gonna be people that are a bit older, professionals, you know, young working professionals, right? And so so the bet is that you remove friction from this process and you get it right algorithmically, and then people are satisfied and they never have to go, which you know, put it is a bit of it puts puts a little disintermediation pressure on the you know, the Macy's and the Nordstroms and the Malls that attract these audiences.
Zoher Karu (34:16.318)
Yeah. Yeah. I mean, you know, look, there'll be people who always love to, you know, browse, if you will, and experience the mall interaction, if you will, and if not for them necessarily. But like if I said, Okay, Brett, here's a thousand dollars. I want you to go pick six pieces of clothing and northern and come back out, like it's overwhelming, honestly, at times, right? You like you don't you don't know where to start. You know, like, well, does this even look good on me? Like, I don't know, you know, and so you need
Brett (34:38.173)
Yeah.
Zoher Karu (34:44.568)
To somehow bottle some expertise. Like, yes, you could talk to the Nordism stylist and they would help you out, fine, but that is not scalable. Yeah.
Brett (34:49.309)
Yeah.
Rio (34:49.453)
But it's like to your point earlier, it's hit or miss as well. Like y yeah, you don't know if like the stylist will be any good, right? Or and or maybe there won't be enough that day. Yeah, but that's a good point. So
Zoher Karu (34:59.711)
Yeah, like can you can you capture the genie in the bottle, right? And scale that is really what we're trying to do is like what makes this good and how do I capture that and and give it to more people, right? and of course it's the time savings and all effort savings and all of those kinds of things also. But you know, you want to be, you know, p you want to look good, you want to feel good, like okay, how do I achieve that, right? and so what's going to help me do it?
Brett (34:59.891)
Yeah.
Zoher Karu (35:27.102)
Even if you have the time and the money, it's not an easy problem, right? to go do it yourself. And so how do I this is a way to help you.
Rio (35:35.469)
So the clothes that they get, are they initially they're renting them and is there an option to buy? How does it function?
Zoher Karu (35:40.941)
Yeah, you so we we're renting the clothes, of course, dry cleaning and washing them every time it comes back. and so the if you like if you get it and you like it, you just keep it, you you know, pay for it and you keep it, you don't have to send it back. If you do want to send it back, you send it back, you get a refreshed set, right? and so I've bought about I don't know, actually about half the stuff I've been sent, right? And so which is which is a pretty good yeah, it's pretty good hit rate. And I'm like, I'm like
Rio (36:04.811)
Pretty good hit rate. Okay, half, yeah.
Brett (36:06.097)
Yeah.
Zoher Karu (36:08.364)
Wow, okay, that is pretty good, actually. Right. I I'll I'll keep that, right? and so initially when I open the box, I'm like, what? What is this? I'm not sure about this. And so, but then you wear it, and then maybe somebody gives you a compliment. You're like, okay, fine, and maybe it does look good, right? and so it pushes your pushes the boundary a little bit, right?
Rio (36:29.613)
Yeah, no, it's nice because I used to sign up for this subscription service for shirts where they would send me, I think three, three shirts, four shirts every you could s you could do it every month or every three month. I think I did every three months or something like that. And for a while it was great. Then I but then I had this closet, like of this massive closet of shirts. And I realized about a third of them I never wore, right? So it's actually that's that's actually smart. If you can send the stuff back, you don't re you're not really that crazy about.
Brett (36:50.481)
Yeah.
Zoher Karu (36:50.776)
Yeah, yeah.
Rio (36:56.291)
You try it on once, 'cause that's the thing. You need to try it to see. And then you're okay, this is good, but I'm never gonna wear it again or or it just be able to send those back that I like that.
Zoher Karu (36:56.748)
Yeah.
Yeah, yeah, yeah, yeah.
Yeah, and you know, I mean, you know, there's a s there's an undercurrent of kind of the environmental impact as well. Like, you know, people just have who like landfills are full of clothes, honestly, right? and so how do you you know, experience some variety, experience things that, you know, push you out of your comfort zone a little bit, you know, without just filling your closet necessarily.
Brett (37:26.929)
Yeah, yeah, you have to it it's it kinda solves the Marie Kondo problem. This does not bring me joy, but it's i i interesting fact about the the landfills is that a large percentage of US sort of thrown away clothing is sent to Ghana. and it fills up
Rio (37:44.494)
yeah, we we've absolutely nuked their textile business in the whole continent of Africa. We've been sending them in fact, in fact, what's crazy is Rwanda tried to end it, right? They actually were like, no, we we want to ban all this, like these crappy, like used clothing you're sending us. And then the clothing the used clothing lobby here prevented Congress somehow from pa from from like they passed some law like penalizing Rwanda. This is crazy, right? The fact that people don't realize the impact to your point. Like
Brett (37:49.488)
Yeah.
Zoher Karu (37:59.225)
Yeah.
Rio (38:13.293)
throwing away these clothing or giving them away, like they go somewhere, right? Either they go into a landfill or they go to like they we send to some country to to
Brett (38:18.705)
Well, it's the it's the it's called the Kantamanta, I'm gonna mispronounce it, Kantamon Market in Accra, right? And so my f my brother-in-law lives there, and they bring in they they employ like two hundred thousand people, and they bring in bil millions of tons of of secondhand clothing from the world, and and then they filter through the crap, basically. They have they've all of these workers are filtering through the stuff that's sellable, and then the stuff that's sellable goes on pallets and then is distributed across the entire continent.
Right, all of the secondhand clothing that's relatively cheap. The stuff that is not usable or sellable is dumped on the beaches of Ghana. And and they have, you know, like tens of feet deep of used
Zoher Karu (38:49.912)
Yeah.
Rio (39:03.107)
So you think you're donating your clothes and it's going to go to a good use, it might end up like on a beach in Ghana floating into the ocean just you know just
Brett (39:09.489)
Yeah, 'cause it's just the easiest way for them to get rid of it and it's completely destroying what was a pristine environment. So yeah, just an interesting thing, which I didn't know a lot about until my brother in law moved there and I like, isn't that this is interesting?
Zoher Karu (39:13.548)
Yeah. Yeah, yeah. It's it yeah, for sure. It's sad. I think it's a Netflix documentary or something about something like but anyway, but there's yeah, you you think you're you know, you feel good dropping the thing off at Goodwill, like, I'm doing good for the world. Like maybe not, right? maybe you're making it worse, right? and play you know, the the fashion the fast fashion industry.
Brett (39:34.13)
Yeah.
Zoher Karu (39:41.876)
has made it so easy for people to just like, well it's twenty bucks, whatever, I'll just buy the shirt, right? so yeah. So, you know, in it if you can get people to buy better quality but fewer items and use them l like all those kinds of things, right? Like there's a lot of ripple benefits, if you will. So
Brett (39:44.667)
Yep. Disposable clothing in a way, right?
Rio (40:02.701)
So what's some of the most interesting AI stuff you're working on right now? either for Taylor or other places that you think is you know, the easy taking advantage of the new capabilities like coming out from lo from the the frontier models or just like vertical L L Ms, like what's some of the cooler things that you're you're seeing? Yeah e even if it's stuff that's experimental right
Zoher Karu (40:22.207)
Yeah. Well, you know, the big way first the big wave of the word of the day was agents, right? and then people were doing agents and okay, great. And then people realized that even agents need context. Okay, they need data. Fine, we'll cross the data problem that has its own challenges, like we talked about earlier in terms of cleanliness and the blood and whatever. But okay, now I got agents running around, but sometimes the what is missing in many organizations now is
is what I'll call business context or the ontology or the meaning behind the data, right? Like you can have a database, let's say an airport has a database of arrival time and departure time. Well, who told the database that or you know, arrival time is supposed to be after departure time? That's just a rule that somebody knows in their head, right? but it's not captured in a database somewhere. And so capturing this type of business logic and meaning behind the data is really what is
Brett (41:07.975)
Yeah.
Zoher Karu (41:17.479)
Is is coming into focus. So like I went to the Snowflake conference, for example, earlier this year. And you know, you look around at everybody's booths and talks, it's not agents anymore, it's like ontology and context and all of these other things, right? And people are people are really trying to they're realizing that just because you have the data and just because you have an agent to process the data, if they don't know know the rules, like
that's a VIP client who's trading that stock. Therefore, you need to do this process versus this process. Like, I mean, there's this a lot of stuff that's just captured in people's heads or in documents or whatever, you know. And so how do you get that out of those environments and and scale that, right? is really, I think, where a lot of the challenges in AI these days, right? And which is part of the reason maybe AI didn't all the pilots didn't work necessarily, right? They didn't so
Brett (41:52.828)
Yeah.
Brett (42:04.538)
Yeah, we j
Brett (42:08.666)
Yeah, that's interesting. We just had we just had Eddie Drake from Snowflake on the pod, and he did an entire set of research and then wrote an article you know on LinkedIn about the topic of context, right? And how it's kind of critical to brand IP, right? You're build you're building kind of these rules that again, it's information that's in people's heads, it's it's how you interpret.
the data that the company is amassing around, you know, all of its processes. And that is what's unique to the organization. It's like the organizational you know intel. And and so his whole paper was really around that information is critical to protect both contractually with vendors so that even the patterns that AI is l is is building from this context.
Don't get duplicated into a vendor's ecosystem that a vendor then can take and develop solutions for a competitor or for other. You know, it almost goes beyond this notion of data leakage, which was all the talk, you know, 10, 15, 20 years ago. Not not 20 years ago, where you're you're afraid that data's gonna get out, there's gonna be a data breach, or there's gonna be some sort of privacy issue with the with the use of consumer data. Now it's really about no all of this pattern recognition that we've developed, you know, which is kind of our institutional knowledge.
Zoher Karu (43:13.9)
Yeah, yeah, yeah.
Brett (43:30.808)
as an organization. That's what we need to protect. And we've got all these vendor relationships where that have kind of mushy contractual relationships with us where where they may be, you know, training their models on what they're learning from us and then reselling that for you know to a competitor. And so so is that so you're seeing that that's really all the talk at at the Snowflake conferences and some of these other places?
Zoher Karu (43:53.911)
Yeah, like it it's it's really about capturing that business knowledge, I'll call it, if you will, right? Because okay, just because you have the data, and like I said, that's a challenge, you need to reason on top of the data. Like traditional business intelligence was like, Okay, give me this information, okay, here's the information. But you still need to make a decision off that information, right? And you still need to reason on that. And that reasoning often employs, you know, the the the
Brett (44:01.553)
Yeah.
Rio (44:02.104)
Yep.
Brett (44:12.113)
Yeah.
Zoher Karu (44:22.06)
Bunch of rules and logic and even optimization approaches that sometimes humans are doing in their own head, but it's not really scaled, right? Yeah, that's right. That's right. and so, you know, what's the optimal way to deploy my sales force across this region? Okay, that's a hard problem. Now you can query the database and say, give me sales by region, by person, by month. Okay, fine, here's all your data. You still need to make a decision on how you do that, right? And so
Brett (44:29.126)
Yeah, they it may not be documented or codified in any way within the organization, right?
Zoher Karu (44:52.544)
you need business rules, you need reasoners on top of that to really, you know, optimization tools and graph databases and whatever else, but you need to you know, ultimately all of this AI work needs to lead to decision making, right? it's like a agents are action taking, fine, but what's the decision making, right? and so how do we
And to make that decision often requires what you just said, this this tacit knowledge that's floating around the organization of like, No, no, no, no, that's a duplicate claim claim number. Don't you realize that those things are part of the same surgery? Do you blah blah blah how do you teach the system this kinds of stuff, right? Is is what is hard. So
Brett (45:32.528)
Yep. Stuff that people have learned and have have sort of you know can manage their teams to avoid, but once you start to automate some of these tasks, that that tacit knowledge is not being passed through. Interesting.
Zoher Karu (45:41.88)
That's right. That's right. that no, that is where I think that's where the challenge lies, honestly. And though to your point, intellectual property, like don't let the data leak. Okay, fine, yes, data is valuable, but it's on t what the the meaning of the data and the way you use the data is also intellectual property, right? So
Brett (46:02.833)
Yeah.
Rio (46:04.011)
Makes sense. Should we get a quick hits? I know that you do have a stop at the top of the hour, so here.
Brett (46:06.438)
Sure.
Zoher Karu (46:10.031)
yeah, yeah. I got I got eight ten eight minutes or so, yeah.
Brett (46:10.224)
Yeah, you
Rio (46:16.909)
Brett, you wanna start?
Brett (46:18.608)
You go ahead, 'cause I haven't even pulled them up.
Rio (46:20.547)
Alright. All right, cool. why do AI projects succeed or fail in enterprises? Enterprise organizations. So here.
Zoher Karu (46:31.702)
Yeah.
Dirty data and a lack of appreciation of change management.
Brett (46:38.714)
Yep. Yep. So so what th th what's the gap between AI hype and sort of real business the big the biggest gap between AI hype and sort of real business impact?
Rio (46:40.099)
Like that.
Zoher Karu (46:50.639)
the the the business impact will be achieved largely I mean right now spending is outweighing any sort of cost savings. That's my hypothesis at least, right? and so that's right. That's right. and so, you know, even you you see these headlines like I think it was Microsoft or maybe some other companies.
Rio (47:07.048)
Spending on AI, on tokens.
Brett (47:09.382)
Yeah.
Zoher Karu (47:17.389)
We're taking away AI access from our developers because they're just burning through tokens and s we're costing us a lot of money, right? Because they're not really so what is rethinking the business process is what is the gap between the you know, hype and reality, I'll call it, right? In terms of achieving achieving and that's a hard problem, right? Because rethinking business processes is often interdepartmental. Interdepartmental things are always the hardest thing to ever do anywhere.
Brett (47:33.307)
Yeah.
Zoher Karu (47:44.781)
Right, because then that means people have to talk to each other and coordinate and whatever, like, I don't have time for you, I'm doing my thing. Right. and so that's I think that is what is going to be the wall that people need to get over. So
Brett (47:49.777)
Yeah.
Rio (47:56.856)
Yeah, no, I I like that. looking at personalization, is are we reaching an age of true personalization with AI or is it still gonna be one of those things that is always a few years down the road?
Zoher Karu (48:11.605)
No, I mean I've done a lot of personalization in the past based on, you know, predictive model scores of you know, different offers for different people. And like even back in Sears days, we were printing custom offers to each person on the cast register received, right? and so it was built on but it was like daily refresh and we we were pretty sophisticated. We even took into count temperature outside. So it was cold outside, you actually got different offers than if it was hot outside, right? And so it was relatively advanced for the time. But I think what is really the opportunity now.
Rio (48:38.691)
Pretty sophisticated, yeah.
Zoher Karu (48:40.685)
Yeah, what is really the opportunity now is three dimensions. One is there is a lot more data, signals that you can capture about a person, right? It's not limited to the 30 attributes I put in my predictive model, right? there's a lot of things that you can quickly get access to about a person. And number two, it's the velocity. Like you don't have to wait for the batch overnight reupdating, scoring the model, like what are like you can recompute a lot of the stuff on the fly, right? And the compute power
across all that data, the velocity is fast. And also the ability to generate is is massive, right? Like they're not like three versions of the, you know, email and I'm picking the right one for you. Like there's an infinite number of versions of the email now, right? Basically, that I could generate for you. And so I do think that personalization, if people try to take advantage of it, between velocity and the breadth of data and the generation
ability, I think it really can make a difference.
Brett (49:43.622)
Yeah, it it it
Rio (49:43.672)
Yeah, I like that. I mean the ability to like s sift through and find patterns in a lot of data, which would have been very difficult before AI is great at that. You made that point earlier. Then I think the ability to create the hyper personalized content offers creative in order to support that. That was always a big bottleneck, so no disagreement. Brett, I cut you off.
Brett (50:02.223)
Yeah, no, no, no, and to that point is is we've you know, we've seen both in in our career how brands oftentimes struggle with that level of creative variation, ad variation based on, you know, all the segmentation data that in s in in in sort of all audience targeting data that we have, which can be kind of micro targeted, th there was always a gap between that and your actually abil your ability to serve dynamic ads in
thousands of variations and we've had some some folks on the pod and I think it was Felicia Co that said in the gaming industry, so sort of the mobile gaming industry, these companies are accustomed to doing two thousand to three thousand or four thousand variations of a single ad for a single product per month. Right. And so there's this constant you know optimization cycle
And they might have a hundred and fifty different games as a is let's say a game publisher. So the capabilities are there. It's just a matter of of are the tools there on the big brand side to enable them to do that same level of of customization, you know?
Zoher Karu (51:03.747)
Yep. Yep.
Brett (51:06.093)
So so what do you think what why why is AI transformation? You kinda talked about this really organizational transformation.
Zoher Karu (51:15.799)
people people typically, is the human nature, don't like change, right? and how do you get somebody to change? It usually turns into, well, what's in it for me? Right. And if you don't think about that problem of like what's in it for me, you'll always run into resistance. Like people at the top layer of an organization, this is great. We're gonna save so much money and time, we'll be a better company. Let's just do the do it this way. And then it grinds to a halt somewhere in the middle of the
Organization because they're like used to doing it a certain way, right? and so unless there is a value exchange, if you will, it's like do it this way in exchange for doing it that way, this will be better for you. if that doesn't happen, it usually just hits a brick wall, right? Honestly, right? and so yeah, that's right. Not it's not easy to do, honestly, right? All the time, but give people a reason.
Brett (52:01.807)
Yeah. Align incentives. It's critical.
Zoher Karu (52:11.769)
To change and de-risk the change to the point that they're willing to try it, right? Like if you say, I want you to stop doing this and start doing that, just trust me it'll be better. Like, well, okay, I don't know. Right. and so it you how do you take small steps? Any show of change management is like, how do you understand the person you're trying to change? How do you understand the the risks associated with any change? How do you de-risk it? Say, we'll just try this one small thing for this one small area and let's just see if it works, kind of thing, and then just iterate from there. And
making it clear to them what they're getting out of it. Right? and if you're different if they're not getting Yeah. Yeah. So that's simple as that. So
Rio (52:44.779)
Incentives drive behaviors. Could not agree more. Yeah.
Brett (52:47.407)
Yep. A hundred percent. Yeah. Hey well
Rio (52:51.169)
We know you gotta run. This was fun. I mean it was a quick it was a quick one by by Signet Lloyd Standards, but it was fun. We went covered some cool topics.
Brett (52:58.469)
Yeah, and and thanks for joining Zohare and for those that made it this far, visit us at www dot signalinnoise.ai, subscribe to our newsletter and you can find us on YouTube and Apple Podcast as well as Spotify. and now we're on TikTok and Instagram as well. So thanks everybody for for joining. And thanks, Zohare.
Zoher Karu (53:15.224)
Yeah, and the
If people wanna if people wanna try Taylor, they can go to Taylor.style, T A E L O R dot style and check it out, and give it a whirl. or in and yeah.
Brett (53:26.617)
Yeah.
Rio (53:26.691)
Yeah, we'll put it the show notes too, in case anyone wants to check. But yeah, check it out. It's a cool website. I was playing around on it like yesterday when we were preparing for this. It's they got some they got some nice clothes on there. So as Brett mentioned, I I do appreciate clothing, so some good style. So I I recommend checking it out.
Zoher Karu (53:31.981)
Yeah. Yeah.
Yeah.
Brett (53:40.706)
Y you might be their next best customer. we'll see. Well, thanks everybody. Bye.
Zoher Karu (53:41.456)
Okay, hey, thank thanks again, guys. It's been fun. Yeah. Take care. See ya. Bye.
Rio (53:42.851)
We'll see.
Rio (53:48.878)
Right.
Comments