Latest / The Adults In The AI Room / Earley AI Podcast - Episode 68 - Composable Intelligence: Rethinking Customer Data with Seth Earley and Abhi Yadav

Transcript

Earley AI Podcast - Episode 68 - Composable Intelligence: Rethinking Customer Data with Seth Earley and Abhi Yadav · The Adults In The AI Room

• 0:01
  SPEAKER_00: This is the Early AI Podcast hosted by Seth Early.
• 0:04
  SPEAKER_00: Join us as we delve deep into the passions, expertise, and experiences of thought leaders and practitioners to talk about what's possible with artificial intelligence.
• 0:14
  SPEAKER_00: The Early AI Podcast is sponsored by Early Information Science, your digital transformation journey with design and deployment of innovative technology solutions, as well as vector.
• 0:26
  SPEAKER_00: Now enjoy the show.
• 0:28
  SPEAKER_01: Welcome to today's Early AI Podcast.
• 0:32
  SPEAKER_01: Mine, your host, Seth Early, and today we're exploring how AI, data, and the enterprise of maturity are converging to unlock new levels of decision intelligence.
• 0:40
  SPEAKER_01: From customer graphs to real-time journey orchestration.
• 0:44
  SPEAKER_01: Our conversation today is going to unpack the components that make intelligence systems truly work at scale and what most organizations are still getting wrong.
• 0:52
  SPEAKER_01: Joining us today is Abi Yadev, founder of iCustomer.ai, a composable decision intelligence company that's reshaping how brands connect with customers.
• 1:02
  SPEAKER_01: Abi has a rich history of building enterprise technology from pioneering the CDP category, customer data platform, that is, to now developing advanced multigraph frameworks that power customer intelligence with minimum minimal data disruption.
• 1:16
  SPEAKER_01: Abi, welcome to the show.
• 1:18
  SPEAKER_03: Yeah, thanks for having me.
• 1:19
  SPEAKER_01: Excited.
• 1:20
  SPEAKER_01: Yeah.
• 1:21
  SPEAKER_01: So, you know, when you start looking at the current landscape and you start thinking about the MDM era, the CDP wave, and now you're talking about a knowledge era and a knowledge engineering era.
• 1:34
  SPEAKER_01: What's the driving, what's driving that shift?
• 1:36
  SPEAKER_01: And how do you see AI uh accelerating this transition?
• 1:41
  SPEAKER_03: Yeah, no, absolutely.
• 1:42
  SPEAKER_03: I think this is a very important question because when the MDM AI existed, the basic driver was compliance, actually.
• 1:52
  SPEAKER_03: And a lot of companies were trying to create this central source of data repository so they can do whatever they want subsequently in the downstream, which is yet to be defined.
• 2:03
  SPEAKER_03: And then came CDP with a very specific pre-package use case built-in, which is CX, marketing, personalization, and personalization kind of drove all this.
• 2:15
  SPEAKER_03: But now we are almost going back to the basic, which is like, hey, you know, with CDPs and then point-based solution, we cannot have too many data platforms.
• 2:26
  SPEAKER_03: We need one data platform, and that requires multiple graphs, where the knowledge level should be all encompassed because you know these agents uh in AI, which are now the accelerant for having an AI-ready knowledge layer or data layer, is becoming like super critical.
• 2:49
  SPEAKER_03: Because we have suddenly built out so many agents, so many use cases, and people are like, wow, this is awesome! The use cases are like you know, getting you so excited.
• 3:00
  SPEAKER_03: The business has started to see, like, see the impact in the whole demo.
• 3:04
  SPEAKER_03: But when you go back and try to see how we implement, you just don't have a back end, you just don't have this one source of truth, which um, and you have a source of truth, you know, with these, every buzzword has taken, so it's kind of hard to sort of say, like, you have this source of truth, but it's like limited.
• 3:25
  SPEAKER_03: What companies are today looking for is a multi-graph, multiple source of identity, intent, you know, interest graph.
• 3:35
  SPEAKER_03: Like in the world of like CX, marketing, customer intelligence, we have all that data available.
• 3:41
  SPEAKER_03: Either it is not unified at an entity level, and even if it is unified, there are a lot of like ETL pipelines running, which also doesn't work well for some of these agents.
• 3:51
  SPEAKER_03: But then more importantly, like the sheer missing, you know, taxonomy across these data silo, you know.
• 4:00
  SPEAKER_03: So there's no like domain-specific or company specific ontology which exists.
• 4:05
  SPEAKER_03: So I I feel all the kind of you know, uh complicated engineering kind of required, and suddenly uh people are embracing this because they want to make their data AI ready.
• 4:16
  SPEAKER_03: And that's that's what is exciting, I think.
• 4:21
  SPEAKER_01: Yeah, no, that's fine, that's great.
• 4:23
  SPEAKER_01: And I think when you start talking about single source of truth, it's it's it's a source, right?
• 4:28
  SPEAKER_01: It doesn't mean it's a single repository of that.
• 4:31
  SPEAKER_03: Yeah.
• 4:32
  SPEAKER_01: And when you think about it, there's always going to be lots of things that are in motion, right?
• 4:37
  SPEAKER_01: We have many applications that are dealing asynchronously or synchronously with customer interactions, and so those are constantly picking up signals from the customer, those that digital body language, and that has to be reconciled somewhere and then acted upon, right?
• 4:53
  SPEAKER_01: So there are multiple mechanisms for doing that, and they're never going to be completely in alignment, but we need that those sets of organizing principles that form that layer across those applications, that semantic layer, that common taxonomy, the common ontologies, the knowledge graphs that will contextualize the information from those sources and allow other systems to operate on them, right?
• 5:17
  SPEAKER_01: Because it's always about activation of the insight.
• 5:19
  SPEAKER_01: You get an insight, maybe it's a it's a data point or set of data points.
• 5:23
  SPEAKER_01: Well, what do you do with it?
• 5:24
  SPEAKER_01: It's the so much question, right?
• 5:26
  SPEAKER_01: How do you orchestrate?
• 5:27
  SPEAKER_01: So I think what we're saying is that you know, you're always going to have like lots of data sources, information sources, knowledge sources, insight sources, but then we need a unified source of reconciliation, so to speak, to be able to do that stuff.
• 5:44
  SPEAKER_01: And that becomes from the the multiple knowledge graphs that are able to describe this information effectively.
• 5:49
  SPEAKER_03: Yeah.
• 5:50
  SPEAKER_03: I mean, you you've been doing this work for I don't know how many decades.
• 5:54
  SPEAKER_03: I learned a ton from you.
• 5:56
  SPEAKER_03: You you wrote some of the most amazing book, right?
• 5:59
  SPEAKER_03: AI Powered Enterprises.
• 6:01
  SPEAKER_03: Such a good treat about it.
• 6:03
  SPEAKER_03: And you know, what thank you for the kind words.
• 6:06
  SPEAKER_01: I'll I'll take them.
• 6:08
  SPEAKER_03: Yeah, no, absolutely.
• 6:09
  SPEAKER_03: It's totally deserving.
• 6:10
  SPEAKER_03: And I learned a ton personally.
• 6:13
  SPEAKER_03: And I I'm trying to now kind of tone it down for some of my sort of CMO, CDO, kind of more retail B2B kind of industry audience, that making correlation between MDM, CDP, and then now this knowledge engineering kind of in a way they see the value and they see that running after AI and agent is great, kind of a North Star, but we gotta catch along, you know, our data readiness in the same way.
• 6:43
  SPEAKER_03: What you just expressed, you know, I I came up with this framework for kind of, you know, erstwhile CDP audience, that what you in sort of system of action, system of source, what you actually now sort of require is kind of, you know, in your unified data fabric, which is a lot of people have spent a lot of millions of dollars on creating like unified data, unified data, either in a warehouse, either in an MDM, either in a like a multi-graph system.
• 7:14
  SPEAKER_03: Like I'm a I'm I'm a I'm an advisor in a company called RELTO, uh, which started with an MDA.
• 7:20
  SPEAKER_03: Is it called what is it called?
• 7:22
  SPEAKER_03: It's called RELTO.
• 7:23
  SPEAKER_01: How do you how do you spell that?
• 7:24
  SPEAKER_03: Reltio is R-E-L-T-I-O.
• 7:28
  SPEAKER_03: Okay, just like Reltio, uh and you know, like Salesforce just acquired Informatica, right?
• 7:35
  SPEAKER_03: So so in the magic quadrant of Gartner, there are only two companies, Informatica and Reltio.
• 7:41
  SPEAKER_03: And Reltio basically pioneered this multi-domain, multi-cloud, MDM solution more on the cloud situation, while Informatica started more like on-prem.
• 7:52
  SPEAKER_03: And what they're seeing, and I'm kind of proud to associate it with those folks as well, has really bringing an organization-wide multi-graph approach into a sort of production and building ontology and some of those things, you know, for some of the kind of work required to solve multiple problems.
• 8:13
  SPEAKER_03: Not just me and I customer, we're like laser focused to solve like customer experience, getting operation, marketing activation, ad activation.
• 8:22
  SPEAKER_03: But then organization has this multi-org, multi-agents kind of thing.
• 8:28
  SPEAKER_03: And where I was going with that was we basically, so you have a unified data fabric, which companies like you know, Railtier and some of them kind of already created.
• 8:38
  SPEAKER_03: But now, on top of a unified data fabric, according to this framework I built, we we now need a data uh quilt on top of that.
• 8:49
  SPEAKER_03: I just use the word data quilt better than the fabric.
• 8:53
  SPEAKER_03: You do require fabric, which like how you express the source of truth and unification and all that, but you do need that all the ontology and knowledge engineering on top, which is some of the work you're doing, and you're doing amazingly well, like creating this data quilt on top of it.
• 9:10
  SPEAKER_03: I like that.
• 9:10
  SPEAKER_03: I like that to extend the metaphor.
• 9:13
  SPEAKER_01: So you've you've introduced the idea of combining identity, context, and action graphs.
• 9:20
  SPEAKER_01: And how does this multi-graph model support real-time decisioning?
• 9:23
  SPEAKER_01: How is it different than traditional customer data strategies?
• 9:26
  SPEAKER_01: We've touched on a little bit of that, but maybe we can expand on that a little more.
• 9:30
  SPEAKER_03: Yeah, I know it's such an important question.
• 9:32
  SPEAKER_03: So, customer data was primarily about whatever customer definition the business has.
• 9:40
  SPEAKER_03: Mostly it was coming from loyalty database in the retail B2C wall.
• 9:44
  SPEAKER_03: In B2B, it was like, okay, whatever is in my CRM or SNOP, like let's just call start calling them customer and prospect.
• 9:53
  SPEAKER_03: But then that's the kind of what I would say, like my customer golden profile, like this, right?
• 9:59
  SPEAKER_03: You unify the new people or contact or person or a household or company, and then we just call them as golden profile.
• 10:08
  SPEAKER_03: But what is now happening, and this is not a technological chain, this is actually the business chain, which we call now this is an evolution of like where people will start questioning, you know, well, target is calling SAID as a customer, but the individual said doesn't know that I'm a customer to Target.
• 10:29
  SPEAKER_03: I may have purchased, you know, five times, who cares?
• 10:33
  SPEAKER_03: But what is the definition of customer again?
• 10:36
  SPEAKER_03: So some of us are like trying to educate industry on both zero-party data and then trying to bring this notion, let's just call everyone as audience.
• 10:45
  SPEAKER_03: So, iCustomer, we still say decision intelligence platform, but it is primarily to unify your audience data and to do audience interoperability because my audience lives in different places.
• 10:58
  SPEAKER_03: Like if I'm a brand, my audience spends a lot of time on Meta, spends a lot of time on LinkedIn, spends a lot of time on these Amazon, and that's where I'm like activating a lot of ads and campaigns and things like that.
• 11:11
  SPEAKER_03: Well, I can send email for my opt-in audience.
• 11:16
  SPEAKER_03: You know, obviously, I can send that email directly one-on-one.
• 11:20
  SPEAKER_03: And that's what some people say customer lifecycle management, and that that companies are still doing, and most of the customer CDPs were actually helping do you know campaigns or customer lifecycle management.
• 11:36
  SPEAKER_03: But that now, going back to your question, that data is very limited.
• 11:41
  SPEAKER_03: That's the only data in the loyalty system, which I'm only taking those events, it's not even complete, it's not even like unified touch points out in there.
• 11:51
  SPEAKER_03: I don't even know what they're doing beyond my first party reel.
• 11:56
  SPEAKER_03: And now, with the availability of third-party data set, there's a lot of information you can get out of like if you're doing ads on Meta and LinkedIn, things like that.
• 12:06
  SPEAKER_03: What really happening to my audience right there, if I'm doing so though, those systems are still wall garden.
• 12:12
  SPEAKER_03: Don't take me, but if if I'm spending so much money on ads and I'm engaging that audience, I can still get a lot of those insights and events.
• 12:23
  SPEAKER_03: And this whole process makes the whole system very kind of open standard oriented instead of very close, like, okay, I have this tiny set of customers who are in my loyalty, but then the world is my playground, right?
• 12:38
  SPEAKER_03: And my audience keeps changing, right?
• 12:41
  SPEAKER_03: You know, and their topics keep changing.
• 12:43
  SPEAKER_03: And that's when I know where if I'm tracking a certain set of audience, which are male between 25 to 35, what podcasts they listen to, where they hang out.
• 12:54
  SPEAKER_03: So I want to learn all that, right?
• 12:55
  SPEAKER_03: So and and and humans evolve, audience evolve.
• 12:59
  SPEAKER_03: Yeah, I just don't have a way to keep up with all the changes going, and that's where like iCustomer is laser focused right now, is keeping your audience or behavior.
• 13:09
  SPEAKER_03: We have like taste profile on customers and things like that.
• 13:13
  SPEAKER_03: So we we keep up not just with your first party, but also like third-party signal out of thing and kind of evolve, help you see the evolvement of your customer behavior.
• 13:25
  SPEAKER_01: That's great, that's terrific.
• 13:26
  SPEAKER_01: Thank you for the explanation.
• 13:28
  SPEAKER_01: Can you just for level setting and for getting people on the same page define zero party, first party, second party, third party data, right?
• 13:37
  SPEAKER_01: Because we many times you talk about first party, third party, we forget about what second party is, what's zero party, right?
• 13:43
  SPEAKER_01: There's all sorts of parties here.
• 13:44
  SPEAKER_01: We just want to go have a party.
• 13:46
  SPEAKER_01: So, um, so tell me about the no, I love it.
• 13:49
  SPEAKER_03: Uh, the question is very relevant, and I'm sorry, I I've been living in this world for two times.
• 13:54
  SPEAKER_03: I know, I know.
• 13:54
  SPEAKER_01: You gotta take a step back, and yeah.
• 13:56
  SPEAKER_03: So I'll start something basic, right?
• 13:59
  SPEAKER_03: Third party.
• 13:59
  SPEAKER_03: So third party is like any third party selling database, like Zoom Info, DNB, Axiom to Experian, you name it.
• 14:10
  SPEAKER_03: Any third party data vendor, if you go to like any marketplace, data marketplace, and you see the list provider, you'll get hundreds of thousands of data vendors.
• 14:19
  SPEAKER_01: You know, it's astounding.
• 14:21
  SPEAKER_01: Many of them just buy from each other and just resell, and it's like, what do you, you know?
• 14:26
  SPEAKER_03: Yeah, that's that's the topic for another day.
• 14:29
  SPEAKER_03: On and on and that, yeah.
• 14:30
  SPEAKER_03: So that's third party.
• 14:32
  SPEAKER_03: The second party is, you know, where some people have come up with a cooperative, which is like, okay, you know, you share your some fields, I share some fields.
• 14:41
  SPEAKER_03: So, like crossbream, you know, is a B2B player.
• 14:45
  SPEAKER_03: Uh, and there were like other players too, like which used to share like who's buying at a certain time and certain like a syndicate and a PIA production.
• 14:53
  SPEAKER_01: You say cross stream or cross stream, cross streams, cross stream, okay.
• 14:58
  SPEAKER_01: You said cross stream is a vendor, cross beam.
• 15:03
  SPEAKER_03: I'm sorry, the name is cross beam, cross beam, yes, beam is the cross beam, like B E A M.
• 15:11
  SPEAKER_01: Okay, yeah, and it and it handles cross stream.
• 15:15
  SPEAKER_01: Yes, that's the concept.
• 15:17
  SPEAKER_01: So maybe explain that a little bit more.
• 15:18
  SPEAKER_01: Explain that keep that more.
• 15:20
  SPEAKER_03: Thanks for that call out.
• 15:21
  SPEAKER_03: Okay, so cross beam is the vendor, is an example, second party, and like that.
• 15:26
  SPEAKER_03: There are my there are many.
• 15:29
  SPEAKER_01: First party is one I know, but but explain again just what does a second party data vendor do?
• 15:35
  SPEAKER_01: So it's saying that you're getting information, but you're you're you're you're not selling it as a third party, but you're sharing it as a partner, perhaps, yeah, cooperative or some mechanism where exchanging data that seems like a little subtlety there that we need to understand.
• 15:52
  SPEAKER_03: So you could say like second party is like shareable data with just like third-party data vendor is just one such, right?
• 16:00
  SPEAKER_03: But in second, you're just connecting two or three sources, yeah, as a syndicate, and everyone's getting the complete piece of puzzle.
• 16:10
  SPEAKER_03: So let's say there's one data about the same entity is coming from a different source and the other, and then the other, and you unify all the three, four, it just becomes a cooperative, right?
• 16:22
  SPEAKER_01: It could again could an example be so I'm having a webinar, and then I have a sponsor that webinar or a partner in that webinar, and I say, check the box to share the information with them.
• 16:33
  SPEAKER_01: That's actually second party data because we're cooperatively sharing it, I'm collecting it, but then it's we have a relationship to do to share that with, and people opt in for that.
• 16:42
  SPEAKER_01: So that's I'm not selling it as a third party, I'm sharing it as a second party.
• 16:46
  SPEAKER_01: Okay, right.
• 16:47
  SPEAKER_03: And all the parties will you'll know which all parties are like is aware of the same information, uh, more less.
• 16:55
  SPEAKER_03: Yeah, absolutely.
• 16:56
  SPEAKER_03: Great example.
• 16:57
  SPEAKER_03: And then the first party is really about your own, like brand specific, where you own the IP, this is your network, this is your under your firewall, whatever has happened, right?
• 17:09
  SPEAKER_03: That's all first party is.
• 17:11
  SPEAKER_03: And then when the whole privacy lock started to come in, actually, it came through the whole GDPR thing, and I was remember, like I was running my last company, which is a CDP leader, their whole GDPR became like a kind of tailwind to invest in CDPs and data warehouse and data platform in general, and have like unified profile.
• 17:37
  SPEAKER_03: But then there is a catch.
• 17:39
  SPEAKER_03: The catch was that a lot of consumer data you were capturing, whether your explicit customer or even a prospect who may have touched, visited your website, may have given you a call, maybe that going back to Target example, right?
• 17:56
  SPEAKER_03: I might have purchased and I may not be in a loyalty system of Target.
• 18:01
  SPEAKER_03: Um, I in a way, yes, I am a Target customer, but I never give you give Target explicit consent that you can store my data and use it for any other marketing.
• 18:15
  SPEAKER_01: Like I have never given that, and that drove to this new initiative with analysts started calling zero party, and that is the data that they explicitly opt into giving you for the purposes of our interactions, right?
• 18:31
  SPEAKER_03: That's that's absolutely right.
• 18:33
  SPEAKER_01: So, again, zero party is stuff that you've explicitly given me.
• 18:37
  SPEAKER_01: You've logged on and you've given me your name and your address, and you've filled out some stuff.
• 18:41
  SPEAKER_01: First party is inferring that or implying that or deriving that from lots of different touch points, yes?
• 18:49
  SPEAKER_02: So there's no explicit.
• 18:50
  SPEAKER_01: I mean, you explicitly permission, but it's data that I've collected about you from your interactions, yeah.
• 18:57
  SPEAKER_01: And I and I can act on it because you've given me access to your interactions, but I've certainly can't sell it, and I certainly can't share it, right?
• 19:06
  SPEAKER_03: Yeah, totally.
• 19:08
  SPEAKER_01: And I mean, so it's a subtle difference between zero and first.
• 19:12
  SPEAKER_03: I mean, you basically start with first party data collection, which is your brand.
• 19:17
  SPEAKER_03: You start unifying so you know who's what, and you do identity resolution and all those things, but then you start, and this is the way how enterprise are like all these things are afterthought, right?
• 19:29
  SPEAKER_03: They're not built in.
• 19:30
  SPEAKER_03: We have seen in our career like all these things, a lot of these things keep piling up and they become afterthought.
• 19:37
  SPEAKER_03: So zero party came as an afterthought where now that we have this first-party data, but I don't think we have like double opt-in, or we don't have explicit consent.
• 19:48
  SPEAKER_03: So either we run a campaign or we do a gamification, we call them as our community.
• 19:54
  SPEAKER_03: Yeah, there are interesting ways to make it zero party, but you got to have a first party ecosystem level.
• 20:02
  SPEAKER_01: Understood.
• 20:02
  SPEAKER_01: Um, I think that that's been great.
• 20:04
  SPEAKER_01: So thank you for explaining that and expanding on that because they're subtle differences, right?
• 20:09
  SPEAKER_01: And but they're all very important.
• 20:10
  SPEAKER_01: But when you've worked with large organizations, like you've worked with Fortune 1000, Fortune 500, Fortune 100 companies, um, what are the recurring challenges that you're seeing at the enterprise level?
• 20:21
  SPEAKER_01: What keeps decision intelligence from scaling?
• 20:24
  SPEAKER_01: And how do you break through?
• 20:25
  SPEAKER_01: How does your approach break through?
• 20:27
  SPEAKER_01: Yeah, this is such not a commercial for iCustomer, but no, no, absolutely philosophically discussing it, you know, you know, how do you break it?
• 20:35
  SPEAKER_01: Absolutely.
• 20:36
  SPEAKER_03: I look at this from the last decade.
• 20:38
  SPEAKER_03: What has happened is people didn't even have basic data infrastructure.
• 20:43
  SPEAKER_03: So they invested in these MDM, CDPs, data warehouse, data lakes, whatnot, to start collecting this and then start adding sophistication like these compliance and activation kind of plans and things like that.
• 20:58
  SPEAKER_03: But I do still see that there is a lot of challenges in activating these databases from these data platforms.
• 21:07
  SPEAKER_03: Because that's where you know use cases drives all this.
• 21:11
  SPEAKER_03: And now suddenly with agentic architecture and agentic AI coming in place, the earlier world where like CDP used to say sell-service activation, you can segment the data in a self-service way, and you're a marketeer, you don't have to do coding, and you can activate a segment into an ad platform, an email platform, that suddenly has shifted to a more built-out agentic infrastructure.
• 21:38
  SPEAKER_03: And this is what is going on.
• 21:40
  SPEAKER_03: But there is still the single biggest challenge in enterprise, or in general, I would say even in mid market, which is all these people, all these leaders, and I'm seeing that evolution and market in enterprise across all level, there are a lot of new leadership which are coming.
• 22:00
  SPEAKER_03: In or folks are coming in, or now they have a decade years of experience of data-driven something.
• 22:06
  SPEAKER_03: You know, they're all getting comfortable with how data and pipes and activation kind of work.
• 22:12
  SPEAKER_03: And so you can always automate this through a new technology or agents and things like that.
• 22:18
  SPEAKER_03: But the biggest challenge is who takes the decision?
• 22:22
  SPEAKER_03: Is it the human?
• 22:23
  SPEAKER_03: Is it the agent?
• 22:24
  SPEAKER_03: Or is it some kind of decision agents will take, some kind of decision human will do.
• 22:30
  SPEAKER_03: And that's where the whole like right now, the struggle, or I would say figuring it out going on.
• 22:39
  SPEAKER_03: And this is what we call decision intelligence, because we think, you know, the whole dogma of like, okay, single source of truth, data activation was kind of the level one.
• 22:51
  SPEAKER_03: But the real interesting problem is now this the second layer, which is the decision intelligence.
• 22:57
  SPEAKER_03: And decision intelligence, whether human or agents taking, requires trusted inputs, context.
• 23:06
  SPEAKER_03: And that's what, like in the AI was like, oh, we need context.
• 23:09
  SPEAKER_03: If I had to take a decision here, then I need a context from where why are we doing here separately?
• 23:16
  SPEAKER_03: And there's another data rule on the other workflow separately.
• 23:20
  SPEAKER_03: Context varies by workflows.
• 23:22
  SPEAKER_03: This is where decision intelligence is very critical.
• 23:25
  SPEAKER_03: Going back to your original question, this is where I think the biggest struggle is because decision is normally taken by operators and not technical people.
• 23:37
  SPEAKER_03: And now we have this thin line between where is the technology nuances and where is the business operator taking.
• 23:46
  SPEAKER_03: And that's where we are trying to build machines which are more friendly to like business operators.
• 23:53
  SPEAKER_03: So they can decide, they can delegate, just like employees, that I want you to work for me as my loyal agent, and you can take all these kinds of decisions.
• 24:05
  SPEAKER_03: But for these kinds of decisions, escalate it to me or things like that.
• 24:09
  SPEAKER_03: I think that's where the world is moving.
• 24:12
  SPEAKER_03: That's where we are personally kind of inclined and trying to focus on laser focus on those workflows.
• 24:18
  SPEAKER_03: And it varies by vertical by vertical, domain by domain.
• 24:22
  SPEAKER_03: That's why domain knowledge is important, that's why vertical nuances is important.
• 24:26
  SPEAKER_03: We at iCustomers laser focus on marketing operations, you know, GTM operations, we call it, you know, because there the entire buyer journeys has hundreds of tools and hundreds of decisions to be taken.
• 24:41
  SPEAKER_03: You know, in B2B context, when a lead comes on a website and that person is in my audience data, the workflow would be different, the decision would be different.
• 24:51
  SPEAKER_03: The whole one size fit-all approach was still like some sort of rule engine with like Zapiers and kind of these kind of automation tools, you could do that, which is now going to a level where somebody got to take an actual decision at that node level.
• 25:08
  SPEAKER_03: And this is where we kind of laser focus.
• 25:10
  SPEAKER_03: And I feel this is an interesting problem yet to be solved.
• 25:14
  SPEAKER_01: So, yeah, and and what's interesting about this is when you talk about the fact that there's hundreds of tools and hundreds of decisions.
• 25:22
  SPEAKER_01: How do you begin mapping that, especially when you're trying to build a gentic approaches, right?
• 25:28
  SPEAKER_01: Because we still have to say, if we're building an agent, an agentic approach, we still have to map the process as though a human were doing it in some way, shape, or form, right?
• 25:36
  SPEAKER_01: We need to know what the inputs are, we need to know what the outputs are, we know to need to know what the objectives are.
• 25:40
  SPEAKER_01: We can then kind of put some guardrails around it in terms of what the capabilities are, what data sources, what the actions can be.
• 25:47
  SPEAKER_01: You know, you don't want your agent to decide that the best way to increase, you know, volume sales of a product is to give it away for 10 cents, right?
• 25:56
  SPEAKER_01: When it costs$100.
• 25:57
  SPEAKER_01: So you have parameters, you have guide, you know, guardrails.
• 26:01
  SPEAKER_01: But but how do you begin to structure that?
• 26:04
  SPEAKER_01: Like, do you take it on a use case by use case basis and then extrapolate?
• 26:09
  SPEAKER_01: Do you have classes of use cases?
• 26:11
  SPEAKER_01: Um, because again, it seems to me that it's going to provide a lot of uh uh you know uh kind of a uh a uh combinatorial explosion, right, of possibilities.
• 26:22
  SPEAKER_01: So how do you begin to factor that down, filter that down, collapse it down to a reasonable number of human understandable use cases that then can be extrapolated, I guess.
• 26:33
  SPEAKER_03: Yeah, no, it's very challenging right now.
• 26:36
  SPEAKER_03: So what we've done is we looked up, you know, why people were using these proven strategies in vertical.
• 26:44
  SPEAKER_03: So in e-commerce, everyone was using CDP to do segmentation and activation for personalization.
• 26:51
  SPEAKER_03: So in order to do that whole journey, you have like bring the data, unify, then do something, and then you create playbooks, you know, in journeys.
• 27:02
  SPEAKER_03: And we kind of map that whole process from the data source to all the way to like activation, and we brought in this concept in iCustomer called Oda Loop, you know, as you can imagine, observe, orient, decide, and activate, which is basically an always-on audience.
• 27:24
  SPEAKER_03: So if I have a unified data fabric of my audience in my data platform, then there are a bunch of agents which are like observing all the events happening for the customer profiles I'm concerned, or prospect profile, I'm concerned.
• 27:41
  SPEAKER_03: And I'm I'm seeing all their journey, I've automated their journey in a kind of optimized way.
• 27:48
  SPEAKER_03: We have continuously like deciding, you know, what is optimized decision making for there, and then there are guardrails, there are like escalation paths to humans come look at it.
• 28:00
  SPEAKER_03: Like you want to activate an ad campaign, have a certain budget set, and you wanted to tweak it every time.
• 28:07
  SPEAKER_03: If it is not working, you want it to stop.
• 28:10
  SPEAKER_03: Messaging, adjusting messaging.
• 28:12
  SPEAKER_03: We we try to focus more on like data observability and events observability and customer and nuances in a in a more you know bespoke situation for each tenant, even though this is multi-tenant kind of a scalable way.
• 28:28
  SPEAKER_03: But ODA loop is a learning loop, it kind of self-learn, it has a learning loop, so you can see ODA observe.
• 28:36
  SPEAKER_03: What was the what is the actual ODA loop is uh is a military style technique uh was started by somebody kind of did a observe decide addressing yes, it's called double ODA, like observe, orient, decide, and activate.
• 28:52
  SPEAKER_03: So you observe your audience data, uh, you orient it, which is means this is my high value customer, or things like that, and you know, whatever the basic rules setting you have done, and then you have decision loops and things like that, and then you actually activate whatever needs to be activated.
• 29:11
  SPEAKER_03: So, going back to your question, one is this way we've been able to do that, but it is way more complex in other use cases, industry.
• 29:23
  SPEAKER_03: Not that this is not complicated or what we've done, but like I was talking to a CIO of a healthcare company the other day, and they are super excited like about having decision intelligence in their place because they feel that because of compliance reason, people have been talking about data lineage all this while.
• 29:46
  SPEAKER_03: But what about decision lineage?
• 29:48
  SPEAKER_03: They want to know whether human or machine or whoever has taken set of decisions, when the auditors come in, they could see at what level these decision lineage exists.
• 29:59
  SPEAKER_03: So this would also bring a whole new kind of, I would say, excitement or a new category around like creating decision intelligence or decision lineage.
• 30:08
  SPEAKER_03: I was interacting with an analyst, I believe Gartner has suddenly started recognizing this as a new category.
• 30:16
  SPEAKER_02: Yeah.
• 30:16
  SPEAKER_03: Coming up with a new hype cycle or whatever report.
• 30:21
  SPEAKER_01: New acronyms.
• 30:22
  SPEAKER_03: Yeah, no, no acronyms.
• 30:24
  SPEAKER_01: Make sure it's buzzword compliant.
• 30:26
  SPEAKER_03: Yeah, totally.
• 30:28
  SPEAKER_03: So um, so I think we are at a cusp where people were super excited to see how agents can be wired together and do a workflow.
• 30:39
  SPEAKER_03: But all of a sudden, now, you know, when the the investment on data foundation, this data fabric and quilt will be in place, the next thing they gotta focus on is all things about you know decisioning and and then tracking of these decisioning and you know the lineage and aspects, measuring, measuring, and then the learning loop kind of a thing, which is uh what you cannot measure how would you like improvise or optimize, like how in statistic we say like optimization is the the far end, right?
• 31:17
  SPEAKER_01: Yep, you need you need data for that, absolutely.
• 31:20
  SPEAKER_01: Um, one of the things uh that uh I I wanted to, and that's all very, very helpful because again, it does become a very challenging problem to get your mind around, right?
• 31:31
  SPEAKER_01: If you are a CMO or CIO and you have to decide, you know, what where what's in the hands of the operator, what's in the hands of the technology, how do we design this, how do we optimize this?
• 31:40
  SPEAKER_01: And then you kind of have to start with, I like to say the handcrafted artisan workflows, right?
• 31:46
  SPEAKER_01: You have to say, what's an ideal happy path here, and then what are the variables that I can adjust, and then componentize your your system, your your content, make sure that you have the ability to shift in and out, you know, messaging components and hero images and whatever, you know, calls to action, all those things to optimize whatever that messaging is.
• 32:07
  SPEAKER_01: But there's so many of those parameters around different aspects of your campaign, right?
• 32:11
  SPEAKER_01: Ad spend and targeting and where they they are and platforms and so on.
• 32:16
  SPEAKER_01: So there's a lot of variables to it, but it sounds like you can simplify this by looking at a, you know, starting with a core high-value audience and a core high-value campaign, and then start to optimize that and then build on that so that you can begin to see the variations.
• 32:31
  SPEAKER_01: Now, I want to ask a quick question about uh we've talked about the value and the importance of foundational architecture, especially when it comes to knowledge and content systems.
• 32:42
  SPEAKER_01: Why is this still a blind spot for uh enterprise AI efforts?
• 32:47
  SPEAKER_01: Why do you think that's still a uh an area that's that has a lot of gaps?
• 32:51
  SPEAKER_03: Yeah, I wouldn't.
• 32:53
  SPEAKER_03: I mean, if you look at it from a CMO or a marketeer or a business line of business leader, it's fair to say blind spot.
• 33:03
  SPEAKER_03: But if you look at it from a CIO, CDO perspective, I don't think it's any longer kind of blind spot because everything is catalog, index, you know, they can explain to anyone what's the data.
• 33:17
  SPEAKER_03: It's actually a very interesting time right now in the industry.
• 33:21
  SPEAKER_03: And I sometimes notice quite a bit by working with both CIO, CDO, and one side, and then CMO or CRO and kind of business people on one side, where you know, they're the the business people remain very dissatisfied with the state of affairs, with their data is being managed because they haven't seen the R and C DO is in a completely hot spot because you know they had worked over these last decade, few years in inheriting as massive like debt or data debt, you know, and they have put all that things in an order now, so where anyone can even understand from this lake or savannah, like what is what, you know, so you know what that is.
• 34:08
  SPEAKER_03: They're not in the way like domain expert, or sometimes the use case or the nuances of this language doesn't work.
• 34:18
  SPEAKER_03: Where when business wants, can I activate my journeys, customer journeys?
• 34:24
  SPEAKER_03: And then CDU and CIA office kind of wonder, what is that?
• 34:28
  SPEAKER_03: Like, how do we do this?
• 34:30
  SPEAKER_03: And they have a certain way of working, you know, the business folks, and they want it in their favorite tools, or they want it in their favorite workflows in a certain way.
• 34:39
  SPEAKER_03: And those are some of the challenges where they they don't get appreciated for all the hard work.
• 34:45
  SPEAKER_03: But I feel like there's a lot of infrastructure today exist.
• 34:51
  SPEAKER_03: And once this middle layer, which is what we are talking in technical terms, like ontology and knowledge engineering and all the like natural language, because now natural language has taken over from business guys, they don't have to be in SQL expert.
• 35:07
  SPEAKER_03: Once that problem gets solved, and it is solving rapidly, or people are suddenly started talking about, you know, I won't be surprised that CMO will start going and reading your book, you know, very quickly.
• 35:20
  SPEAKER_03: It should be.
• 35:21
  SPEAKER_01: And by the way, even though it's five years old before L, everything in that book is still valid today.
• 35:27
  SPEAKER_01: Totally.
• 35:27
  SPEAKER_03: Yeah, I mean, I attended a meeting uh in an advisory where we we went back to the list and like these are the things changed, but these are the things remain unchanged, even even in an AI first world.
• 35:41
  SPEAKER_03: So let's get our basics in place.
• 35:43
  SPEAKER_03: So I genuinely meant that that awareness is also coming.
• 35:47
  SPEAKER_03: Like I'm talking to a lot of like business leaders, CXOs, CMOs, which uh which have grew very technical from an understanding.
• 35:55
  SPEAKER_03: They're not supposed to code, but the new world we live in.
• 35:59
  SPEAKER_03: I mean, look at marketing.
• 36:00
  SPEAKER_03: I'm sorry, I'm going sidetrack on that point.
• 36:03
  SPEAKER_03: If you look at marketing today, it's full of algorithms.
• 36:06
  SPEAKER_03: I met a CMO the other day.
• 36:08
  SPEAKER_01: Full of what?
• 36:08
  SPEAKER_01: Full of what?
• 36:09
  SPEAKER_03: Algorithms and data.
• 36:11
  SPEAKER_03: Algorithms are driving the budget.
• 36:13
  SPEAKER_03: If you look at a CMO who's got a hundred million dollar budget to spend the CPG brand, look at their distribution for this year.
• 36:22
  SPEAKER_03: They had to spend$100 million.
• 36:24
  SPEAKER_03: There are certain set of data and algorithms are deciding where to spend the hundred million dollars.
• 36:30
  SPEAKER_03: I mean, this is the same situation how like hedge fund manager and more and like all the financial guys had, but they had a long team, you know, a lot of data, a lot of like maturity of this analytic workbench or data scientist workbench they had.
• 36:47
  SPEAKER_03: But look at the marketers have they have a bunch of agencies who who get incentivized only if they deploy more, and then they have creative people, which is the DNA of marketing.
• 36:58
  SPEAKER_03: But look at what how marketing has now turned into a completely left brain function, right?
• 37:03
  SPEAKER_03: Which is all about data algorithm, you know, one wrong move, you end up blowing up so much budget, you know, completely in wrong audience, wrong campaign, wrong messaging, who knows?
• 37:14
  SPEAKER_03: All of that.
• 37:15
  SPEAKER_02: Yeah, yeah.
• 37:16
  SPEAKER_03: So I think it is getting increasingly important for them to bridge the gap and not believe in just like my IT guys will figure it out, my data people will figure it out.
• 37:29
  SPEAKER_03: So I'm seeing increasingly they're kind of leaning in and they're saying, like, okay, we we don't have a common knowledge layer which exists between our data and our use cases.
• 37:40
  SPEAKER_03: So let's let's focus and invest there, right?
• 37:44
  SPEAKER_01: Right, right.
• 37:44
  SPEAKER_01: So that's great.
• 37:45
  SPEAKER_01: And uh, you know, we talked a little bit about the idea of productizing some of the services into something you would call a new age knowledge cloud.
• 37:54
  SPEAKER_01: What would it take to bring something like to mark that to market?
• 37:56
  SPEAKER_01: And how do you see collaboration playing a role there?
• 38:00
  SPEAKER_01: What is what does that really look like?
• 38:01
  SPEAKER_01: What would might that be?
• 38:04
  SPEAKER_03: I I think it is spot on what we were just talking in the earlier question that it is and when we started CDP as a category, the whole reason of CDP category was to bridge the gap between the IT folks and the data folks and the business marketing use case.
• 38:21
  SPEAKER_03: So the ethos remain the same.
• 38:24
  SPEAKER_01: This is the analogy to the analogy to that.
• 38:26
  SPEAKER_01: Go ahead.
• 38:27
  SPEAKER_03: Right, right.
• 38:28
  SPEAKER_03: So this is the same analogy.
• 38:29
  SPEAKER_03: Where now what we need is a little bit more context, and we also need context about our audience beyond our first party data set, what they're doing in real life.
• 38:40
  SPEAKER_03: Like what do they, what are their interests, what are their trades, and using LLMs and some of the kind of more compression and kind of you know, data leverage technology nowadays, a lot easier to consume a ton of data, both internal and external.
• 38:57
  SPEAKER_03: In marketing now, all the plays are happening, and this is what we do enable in iCustomer as well.
• 39:02
  SPEAKER_03: Like a lot of GTM plays or marketing plays are very signal-based.
• 39:07
  SPEAKER_03: Why are you targeting these kind of people or these kind of companies if they're not belonging to your ICP or they're not ready?
• 39:14
  SPEAKER_03: Or in an e-commerce business, like you know, we have like a tons of like third-party data attributes, but instead of you know, like, can we use all those attributes and unify with your first party audience data and then see when is the right time for what campaign based on like the publisher feed you're getting?
• 39:36
  SPEAKER_03: So a lot of those things which were existing earlier was only available to data scientists for manual work, manual cleansing, manual employment.
• 39:46
  SPEAKER_03: But now suddenly agent is kind of taking over the heavy lift, all that data, and that whole data pipeline work and whole ML pipeline work can be automated for like business people.
• 39:58
  SPEAKER_01: That's the stuff we talked about with our platform via, right?
• 40:02
  SPEAKER_01: The virtual information architect, where you can start to ingest that data and start to operate on it.
• 40:08
  SPEAKER_01: You know, AI will not fix your data, but AI can help you fix your data, right?
• 40:11
  SPEAKER_01: If you have the right reference architecture and the right methodologies and so on.
• 40:15
  SPEAKER_01: So that's that's great.
• 40:16
  SPEAKER_01: So, yeah, that's really exciting when you think about where that's going and how organizations are going to need help with managing that over time.
• 40:23
  SPEAKER_01: We're actually doing that for a couple of our customers at this stage, kind of outsourcing that capability to help maintain that.
• 40:30
  SPEAKER_01: So you've you've scaled companies, you've built and exited um from organizations where you built products and exited, you you're shaping a new category.
• 40:38
  SPEAKER_01: What advice would you give founders and technologists or enterprise leaders who are trying to navigate the complexity of AI, of data, of product strategy in 2025?
• 40:50
  SPEAKER_01: What kinds of things would you have them keep first and foremost in mind?
• 40:56
  SPEAKER_03: Yeah, no, it's such an important question, and I cannot tell you.
• 41:00
  SPEAKER_03: Um, as a first-time founder back in the days, I fall in the same trap, and this is what I keep reminding folks who are doing first-time entrepreneurship founding companies and things like that.
• 41:13
  SPEAKER_03: Don't follow the hype.
• 41:15
  SPEAKER_03: Follow the real problem.
• 41:18
  SPEAKER_03: There are still so many real problems.
• 41:21
  SPEAKER_03: The more specific that problem is, and you know, I used to do this mistake like let's build platform, let's do categories making, like all the things.
• 41:31
  SPEAKER_03: But I realized those are like you know, that come much later automatically.
• 41:37
  SPEAKER_03: So you don't start from there and come, even though you can plan, you can have a vision towards it.
• 41:43
  SPEAKER_03: Right.
• 41:44
  SPEAKER_03: But you gotta start with like tool.
• 41:46
  SPEAKER_03: I I remember in my it's funny in my first few years of like first-time entrepreneur, you know, I used to like people used to call our toolkit as like, okay, this is a this is a tool.
• 42:00
  SPEAKER_03: And I used to like, no, this is a platform, you know, you can do this, you there's data, you can build your app, you can do this, blah, blah, blah.
• 42:09
  SPEAKER_03: And it it never serves the purpose.
• 42:13
  SPEAKER_03: So you you need a very small widget, a small to stay laser focused, build an audience, and then solve bigger problems, you know, use the audience to get continuous feedback.
• 42:28
  SPEAKER_03: And in 2025, this is the way to go, I guess.
• 42:32
  SPEAKER_03: Yeah.
• 42:33
  SPEAKER_01: You're right.
• 42:33
  SPEAKER_01: And I think that because there's an overwhelming number of hyped choices and emerging ideas and emerging tools, it is very overwhelming for the typical CIO, CDO, CMO.
• 42:47
  SPEAKER_01: It's overwhelming for the experts in the space, right?
• 42:50
  SPEAKER_01: But if we take a step back and we focus on the very specific problems that need to be solved, as you say, there's lots of current state problems that need to be solved.
• 42:58
  SPEAKER_01: You know, organizations are not necessarily thinking, you know, five steps down the road, or maybe they are thinking that way, but they say, well, we have a lot of stuff that we need to do right now.
• 43:08
  SPEAKER_01: And that's the thing to focus on, focus on those very specific, measurable problems that can be solved today.
• 43:14
  SPEAKER_01: Well, listen, thank you so much, Abby, for joining us today.
• 43:18
  SPEAKER_01: I really appreciate you sharing your insights.
• 43:20
  SPEAKER_01: It really is clear that the next wave of enterprise intelligence is going to be, you know, it's going to be powered by these thoughtful designs that will help people activate on the data that they have, right?
• 43:31
  SPEAKER_01: And be able to collect that from lots of sources.
• 43:34
  SPEAKER_01: And we do need all of those uh pieces.
• 43:36
  SPEAKER_01: And to our listeners, thanks for tuning in.
• 43:39
  SPEAKER_01: Stay us for more conversations on how AI is transforming business across industries.
• 43:44
  SPEAKER_01: And again, thank you so much, Abby.
• 43:45
  SPEAKER_01: Appreciate your time.
• 43:47
  SPEAKER_03: No, thank you for having me.
• 43:48
  SPEAKER_03: This was a fun conversation.
• 43:50
  SPEAKER_01: Absolutely.
• 43:51
  SPEAKER_01: Okay, everyone, we'll see you next time.
• 43:53
  SPEAKER_01: And thank you, Carolyn, for your work behind the scenes in producing the Early AI podcast.
• 43:58
  SPEAKER_01: Thanks, everyone.
• 43:59
  SPEAKER_00: Bye, thank you for joining us on another deep dive into AI innovation.
• 44:06
  SPEAKER_00: Tune in next time when we introduce another industry expert and discuss how to maximize AI in your world.
• 44:12
  SPEAKER_00: The Early AI Podcast is sponsored by Early Information Science and Dr.
• 44:18
  SPEAKER_00: To learn more, visit early.com.
• 44:20
  SPEAKER_00: That's E A R L E Y dot com.
• 44:23
  SPEAKER_00: Thank you for tuning in.