Latest / The Adults In The AI Room / Earley AI Podcast – Episode 92: Supply Chain Intelligence, Knowledge Graphs, and the Limits of the Easy Button with Ilya Levtov

Transcript

Earley AI Podcast – Episode 92: Supply Chain Intelligence, Knowledge Graphs, and the Limits of the Easy Button with Ilya Levtov · 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_02: Well, welcome to today's Early AI Podcast.
• 0:32
  SPEAKER_02: I am your host, Seth Early, and in each episode, we explore how artificial intelligence and data are reshaping business strategy and operations.
• 0:41
  SPEAKER_02: Today, we are talking about one of the most consequential blind spots in enterprise operations, and that is supply chain visibility.
• 0:48
  SPEAKER_02: Most organizations know their top suppliers well.
• 0:52
  SPEAKER_02: Below that, the picture gets murky fast.
• 0:54
  SPEAKER_02: And as recent years have shown, the risks hiding in the middle and long tail of a supply network can be catastrophic when we've seen these interruptions.
• 1:03
  SPEAKER_02: Joining me today is Ilya Levtoff.
• 1:05
  SPEAKER_02: He's founder and CEO at Kraft.co.
• 1:08
  SPEAKER_02: And that they are a supplier intelligence platform that uses AI and knowledge graphs to give enterprises and government agencies visibility into their full supply networks.
• 1:17
  SPEAKER_02: Ilya founded Kraft after career spanning Goldman Sachs, Venture Capital at Venrock, and enterprise partnerships at Deutsche Telekom, and has spent nearly a decade building what he describes as the source of truth on companies, which has evolved into a platform for mission-critical supplier intelligence.
• 1:35
  SPEAKER_02: Ilya, welcome to the show.
• 1:37
  SPEAKER_01: Thank you, Seth.
• 1:38
  SPEAKER_01: It's great to be here, and I'm looking forward to our conversation.
• 1:40
  SPEAKER_02: One of the things I like to start with is our common misconceptions.
• 1:43
  SPEAKER_02: What is it when you talk to procurement leaders and executives about supply chain visibility, what do they most consistently get wrong?
• 1:52
  SPEAKER_01: I would say the biggest is really what you alluded to in your introductory comments there, which is that having good knowledge of, let's say, the top 100 or 200 suppliers that are going into the key products of this enterprise and have that information sort of through a strong relationship between a category manager, a buyer, and that supplier, that that's good enough.
• 2:14
  SPEAKER_01: That really is kind of where things are at.
• 2:16
  SPEAKER_01: And as you alluded to, there's that whole middle and the long tail.
• 2:21
  SPEAKER_01: It can be up to 20,000, 30,000 suppliers easily for a Fortune 500 style enterprise.
• 2:27
  SPEAKER_01: And then many more when you go deeper into the multi-tier, the suppliers of those suppliers.
• 2:33
  SPEAKER_01: And so one of the biggest misconceptions is that, you know, we're good.
• 2:36
  SPEAKER_01: We've got everything that we need with our visibility into the top couple of hundred suppliers.
• 2:41
  SPEAKER_01: And what one of the things that contributes to that misconception is the feeling that having great visibility, insight, knowledge, and control over the rest, the entire supply network, is is so hard, is so it's insurmountable that we we may not as well even bother trying because it's too much data and too and too much intelligence.
• 3:02
  SPEAKER_01: That's really the big one.
• 3:03
  SPEAKER_01: It's exciting to overcome.
• 3:05
  SPEAKER_02: Yeah.
• 3:06
  SPEAKER_02: So people think of it as, well, I got this because I have my top suppliers.
• 3:11
  SPEAKER_02: But what in the reality is it's such a complex ecosystem of interrelationships, and it's a constantly evolving and dynamic environment that you know people think of it as impossible to really understand it and to get that clear picture of what's going on.
• 3:27
  SPEAKER_02: It's just too much, right?
• 3:28
  SPEAKER_02: Just too complex, too many data points, too many moving parts, too many interrelationships.
• 3:32
  SPEAKER_02: There's no way a human can do this, there's no way technology can do this.
• 3:36
  SPEAKER_02: And this is where you guys have made some incredible stride.
• 3:40
  SPEAKER_02: So when we talk about you know that visibility into the top supply base, you know, where does it start breaking down and what is missing?
• 3:49
  SPEAKER_01: So again, if you start with kind of the status quo and how the visibility that is typically had develops and is maintained on those top critical suppliers, it's really done mainly by relationship.
• 4:01
  SPEAKER_01: And it's between the buyer and the relationship person over there.
• 4:04
  SPEAKER_01: And right there, when you see a blind spot emerging, because what is the information they have access to?
• 4:08
  SPEAKER_01: It's the information that the supplier is providing.
• 4:10
  SPEAKER_01: And of course, the supplier is saying everything's hunky-dory, we're great.
• 4:13
  SPEAKER_01: Yes, we've got capacity, yeah, everything's gonna be fine.
• 4:16
  SPEAKER_02: So it's my supply chain is great.
• 4:18
  SPEAKER_02: Yeah, my suppliers are awesome, they rock hard, they rock solid.
• 4:23
  SPEAKER_02: Yeah, they're yeah.
• 4:24
  SPEAKER_01: There you go.
• 4:24
  SPEAKER_01: Exactly.
• 4:25
  SPEAKER_01: Immediately you're seeing the N-tier problem because the supplier is also representing that they've also got great visibility of their supply chain, which you know they don't because you don't.
• 4:34
  SPEAKER_01: So these gaps are uh are coming in immediately.
• 4:37
  SPEAKER_01: And so one of the first ways to mitigate that is to really establish a 360-degree surround sound view, a perspective on the supplier using data that can be aggregated from very many different sources.
• 4:53
  SPEAKER_01: For example, you know, I'll take the most simple obvious, just crawling the web very vociferously for every and any public signal, any digital breadcrumb that's out there, whether it's a news article or it's uh the number and type of job openings that this supplier is publishing today, and how has that number of job openings changed in the last couple of days or weeks?
• 5:18
  SPEAKER_01: It's actually giving you some clues as to what's going on inside that company, right?
• 5:22
  SPEAKER_01: That you probably aren't, you're not going to every supplier's website every single day and seeing what's changing.
• 5:28
  SPEAKER_01: There's that.
• 5:28
  SPEAKER_01: Then connecting with specialty data providers that evaluate and analyze companies' cyber footprint, right?
• 5:36
  SPEAKER_01: There's an example.
• 5:36
  SPEAKER_01: Really monitoring has that company had any kind of data leaks or breaches or appearance of confidential data on the dark web that would indicate there's a vulnerability that could come back and cause a problem for me.
• 5:49
  SPEAKER_01: That these are the kinds of uh data sets and data sources that are that are actually available to bring together and integrate to enrich this picture.
• 5:57
  SPEAKER_02: Yeah, and and and people can think, well, theoretically that sounds great, but how do I operationalize that, right?
• 6:03
  SPEAKER_02: How do I actually pull those signals in that make sense of them?
• 6:06
  SPEAKER_02: And obviously, you know, there's kind of endless permutations of how that data could be uh sourced and and aggregated and analyzed.
• 6:15
  SPEAKER_02: And the idea is taking separating all of that, finding the signal in all of that noise because it is very noisy, right?
• 6:23
  SPEAKER_02: There isn't stuff that you'd have to wait through.
• 6:25
  SPEAKER_02: And a human being can't do it, obviously, by themselves.
• 6:27
  SPEAKER_02: And even with computer technology and AI, it's still a very difficult problem to solve.
• 6:32
  SPEAKER_02: We've evolved this from just structured supplier data, the basic attributes, and then brought that into knowledge graphs where where we have layers of granularity and layers of details that can be applied in different ways and then can integrate different data sources.
• 6:50
  SPEAKER_02: So, do you want to talk a little bit about the impact of knowledge graphs and how that has provided value to the analysis and to the aggregation?
• 6:59
  SPEAKER_01: So you you describe the journey exactly correctly.
• 7:01
  SPEAKER_01: It starts with the collecting of data, structured data, semi-structured data, and unstructured data.
• 7:06
  SPEAKER_01: And that in all in and of itself is an enormous amount of work when, especially when you describe all of the different lenses and angles of view on your supplier.
• 7:16
  SPEAKER_01: There's financial, there's operational, geographic, cyber, I mentioned, ESG, reputational, regulatory, and compliance.
• 7:23
  SPEAKER_01: You know, really building up that picture.
• 7:25
  SPEAKER_01: And it's still really all about just collecting the data and building into what we call a data fabric.
• 7:30
  SPEAKER_01: And it's the foundation of the entire value and the solution.
• 7:34
  SPEAKER_01: It's hard in itself.
• 7:36
  SPEAKER_01: I like to emphasize that just to set people's expectations.
• 7:39
  SPEAKER_01: It's it's necessary, but not sufficient to get to the value.
• 7:43
  SPEAKER_01: And anyone who's got just that far collecting lots of data finds themselves stuck exactly in the place that you said, which is tons of, you know, data by itself just looks like noise.
• 7:53
  SPEAKER_01: How do we get it into the signal?
• 7:54
  SPEAKER_01: And so, exactly as you say, it's about then building on top, call it the you know, the next layer of the cake, which is that it's converting that data into intelligence.
• 8:03
  SPEAKER_01: And what we found over the years of doing this, this is where the concept of the knowledge graph really started to become clear.
• 8:11
  SPEAKER_01: Because at the beginning, structuring all of that data is you typically done in a relational database, right?
• 8:16
  SPEAKER_01: Columns and rows, like a spreadsheet, static attributes and data points.
• 8:21
  SPEAKER_01: And again, it does start there, but in and of itself, it's nowhere near enough.
• 8:25
  SPEAKER_01: Because the reality of this complicated world around suppliers is it really is more a network than a chain.
• 8:31
  SPEAKER_01: The whole supply chain concept is a misnomer, right?
• 8:33
  SPEAKER_01: It's a supply network, but in reality, company A is a supplier to company B.
• 8:38
  SPEAKER_01: Quite often, company B is also a supplier to company A, right?
• 8:42
  SPEAKER_01: You know, there's both, and they're different divisions, and there's different product lines and products, and then and then you've got people.
• 8:48
  SPEAKER_01: So so graph is is a very helpful construct with which to understand that you've got nodes and lines, and the nodes are typically corporate entities in this context.
• 9:00
  SPEAKER_01: It's it's a company, but it's also the parent and the subsidiaries, and it's the manufacturing locations, and then then there's people as well, a really important type of node.
• 9:11
  SPEAKER_01: And then there are all these connections among these entities and these people in in totality, which is giving you an accurate picture of what's going on.
• 9:21
  SPEAKER_02: When we talk about them, you have these entities and then you have relationships between them.
• 9:26
  SPEAKER_02: And we when we build out you know the ontology, right?
• 9:29
  SPEAKER_02: That's all of the taxonomies and all the vocabularies, and then the relationships between them.
• 9:34
  SPEAKER_02: So you have these entities and these vocabularies, but then you have all those relationships between them.
• 9:40
  SPEAKER_02: Here's a supplier and a competitor, right?
• 9:42
  SPEAKER_02: Here is a risk for a supply chain or a risk for a particular supplier, right?
• 9:49
  SPEAKER_02: All of these things.
• 9:49
  SPEAKER_02: And when I talk about this, I talk about the ontology as the knowledge scaffolding of the organization, right?
• 9:55
  SPEAKER_02: It's all those organizing principles and relationships.
• 9:58
  SPEAKER_02: And then when we overlay that on data, we we build the knowledge graph.
• 10:02
  SPEAKER_02: And so that gives you entry points into all of these different entities that allow you to navigate through the knowledge graph to find those other relationships.
• 10:11
  SPEAKER_02: And what this does is it grounds our large language models and any AI in the source of truth, it's a grounding in the entities and the actual data that we are concerned about, right?
• 10:23
  SPEAKER_02: Because you could say, well, tell me about this, your favorite LLM, but it doesn't have that contextualized information and it has not structured it in the way that that is meaningful to an organization.
• 10:35
  SPEAKER_01: You're absolutely right.
• 10:36
  SPEAKER_01: And just to add very briefly, succinctly to that, that knowledge graph is n-dimensional and it's highly dynamic.
• 10:43
  SPEAKER_02: Yep, yep.
• 10:44
  SPEAKER_01: Constantly changing at every single moment.
• 10:47
  SPEAKER_01: And just as you mentioned LLMs, you're absolutely right that the work with LLMs is a massive unlock and accelerant to finding a signal in there.
• 10:55
  SPEAKER_01: But you're absolutely right as to why a generic LLM doesn't work.
• 10:59
  SPEAKER_01: I had a very interesting stat recently, which is that only 1% of enterprise knowledge approximately is it exists inside the LLMs today.
• 11:07
  SPEAKER_01: One percent.
• 11:09
  SPEAKER_01: And of course that makes sense because companies don't want to give all of their data to the LLMs.
• 11:14
  SPEAKER_01: Other data providers who have got, you know, such as that cyber data or that ESG data, they don't want to just give it for free to the LLMs.
• 11:20
  SPEAKER_01: They lose their business.
• 11:21
  SPEAKER_01: And so the LLMs actually don't have this data fabric that we describe, and that's why you need a specialized approach and an application, and then leverage the LLM, the models and the power of those models on your data set.
• 11:34
  SPEAKER_02: And uh organizations compete on their knowledge, right?
• 11:37
  SPEAKER_02: They differentiate on their knowledge, knowledge of customers, knowledge of solutions, knowledge of supply chains, knowledge of routes to market, all of those things are competitive advantage, and they don't want those in LLMs, right?
• 11:47
  SPEAKER_02: Because you don't want to lose that.
• 11:49
  SPEAKER_02: So that's why it's so important to be able to do this in a way that's internal.
• 11:52
  SPEAKER_02: Now, about knowledge graphs.
• 11:53
  SPEAKER_02: Now, you when when we look at this, obviously knowledge graphs are going to be contextualized differently depending upon industry and depending upon the organization.
• 12:02
  SPEAKER_02: So, how do you go into an organization and either derive or mine or try to adapt to their particular entities and their ecosystem?
• 12:12
  SPEAKER_01: Yeah, it's it's a great question, critical to this.
• 12:14
  SPEAKER_01: And the answer is a lot, a lot of bespoke work.
• 12:17
  SPEAKER_01: You know, our approach, and we found this is the only way to do it, is to really operate in a spirit of true partnership with that enterprise.
• 12:25
  SPEAKER_01: You know, and we've got we've got you know numbers of customers in the dozens, not the hundreds, and not certainly not the thousands, precisely for this reason that it's not solvable as an off-the-shelf cloud-based SaaS login and just run and make it work.
• 12:42
  SPEAKER_01: No, you're absolutely the reality of these departments, these enterprises that might have five major divisions and 20,000 suppliers at tier one could very well still be running three different ERPs from different eras and different orchestration layers, you know, leading to all manner of very company-specific data issues and challenges.
• 13:02
  SPEAKER_01: And the only way to really solve it and turn it into this all-seeing supply supply chain solution is to be very, very specific to that company and also understand that it's a multi-year project of illumination and constant iteration and progress towards the goal.
• 13:21
  SPEAKER_01: It's very exciting for that reason, but it's also not solvable overnight with some static or out-of-the-box uh approach.
• 13:28
  SPEAKER_02: I'm really happy to hear that because you know, so many of these organizations that are doing these things say it's auto magic, our AI does everything.
• 13:35
  SPEAKER_02: Now, AI is an accelerator, right?
• 13:37
  SPEAKER_02: Yeah, we did a project for a large global consultancy that's you know multi-billion dollar firm.
• 13:43
  SPEAKER_02: It's you know does a lot of AI work, and we built the ontology for their business units.
• 13:47
  SPEAKER_02: And we we'd used AI accelerators.
• 13:50
  SPEAKER_02: So we ended up building the ontology and then the knowledge graph, and we did it in such a way that work, according to my knowledge architect.
• 13:57
  SPEAKER_02: Normally the work that we did would have taken 12 to 18 months, and we did it in eight weeks.
• 14:02
  SPEAKER_02: And fantastic.
• 14:03
  SPEAKER_02: Yeah, and that was and that's the acceleration that we get by integrating methodologies with the accelerators, with the technology, right?
• 14:11
  SPEAKER_02: It is not just about pointing your AI to data, it is about doing it in such a way that you have those best practices.
• 14:17
  SPEAKER_02: We call it uh virtual information architects.
• 14:20
  SPEAKER_02: So I imagine you have those kinds of accelerators that you're using when you go into an organization.
• 14:24
  SPEAKER_01: That's exactly right.
• 14:25
  SPEAKER_01: And let me just give one top-of-mind example.
• 14:27
  SPEAKER_01: With one of our customers, that is a federal agency in the defense community.
• 14:33
  SPEAKER_01: There was an average time taken to conduct a due diligence analysis on a tier one supplier that was roughly in the in the area of eight hours.
• 14:44
  SPEAKER_01: The analyst would spend on average eight hours pouring through different data sets, pulling it together, and producing the report.
• 14:50
  SPEAKER_01: Using our platform, we've now accelerated that the first version of that report comes out in about 45 seconds.
• 14:57
  SPEAKER_01: Yep.
• 14:58
  SPEAKER_01: The analyst then spending on average 15 minutes reviewing, going back, making some configuration, making some human edits, and then and then hitting rerun and ending up with a with a finished product in somewhere between 15 and 30 minutes on on a supplier.
• 15:12
  SPEAKER_01: So, you know, we're talking 80, 80 to 90 percent acceleration right there.
• 15:16
  SPEAKER_01: Um, and we're still only just getting started.
• 15:18
  SPEAKER_02: Yeah, and I've we found the same thing.
• 15:20
  SPEAKER_02: It's it's a 10x improvement in in what you can do.
• 15:23
  SPEAKER_02: So there's a good bit of work that you know organizations need to do, and obviously the more mature they are and the more work that they have done, the better prepared they will be to do the work that you need to do with them.
• 15:35
  SPEAKER_02: So we talked about this already, that when we supply chain, the term itself implies a linear structure, and we know that that's a misconception.
• 15:43
  SPEAKER_02: So when you start treating this as a complex network, what kinds of things happen?
• 15:48
  SPEAKER_02: What kinds of things change?
• 15:49
  SPEAKER_02: What kinds of visibility do you have?
• 15:51
  SPEAKER_02: And what insights do you have that you don't have when you in a linear fashion?
• 15:56
  SPEAKER_01: Yeah, the the the first example is is how insight often comes from cross-correlating different data streams.
• 16:04
  SPEAKER_01: I'll I'll give I'll give one example.
• 16:06
  SPEAKER_01: We found that sniffing out the digital footprint of companies and looking at various social media platforms, we were able to start building up a picture of the culture inside the company, including the quality and level of employee engagement in that company.
• 16:20
  SPEAKER_01: You could almost sort of, you're getting a window into the morale of a company whose products you actually depend on.
• 16:26
  SPEAKER_01: And then what we were surprised to find was a very high correlation between that level of morale and the ultimate cybersecurity posture of that company.
• 16:36
  SPEAKER_01: And that very clearly, the companies with lower employee engagement, lower morale, a sense of a slightly disaffected workforce that's a bit rudderless were much more prone to spearfishing and leaks and vulnerability to the kind exactly the kind of cyber hack that has then brought down entire companies because of a supply chain hack, right?
• 16:58
  SPEAKER_01: Just one example of taking multiple data streams, integrating them with strong entity resolution, and starting to look and hop across them.
• 17:06
  SPEAKER_01: Here's another example.
• 17:07
  SPEAKER_01: Everybody's familiar with financial risk, you know, in a supplier.
• 17:11
  SPEAKER_01: It's arguably the most traditional, basic one that everybody thinks about.
• 17:15
  SPEAKER_01: Will that supplier be financially solvent and still around six months from now?
• 17:19
  SPEAKER_01: What people weren't really thinking about is that when the answer to that started to become a little shaky and no, a company's financially vulnerable, the company then becomes actually vulnerable to adversarial capital and the influx of investment and connectivity in from entities and actors that might have, let's say, a hidden agenda.
• 17:42
  SPEAKER_01: And remember, with us, the world we're in is mission critical supplier intelligence for national infrastructure, for aerospace and defense, the kinds of places where, given the geopolitical dynamics we're in today, uh the the these are real concerns.
• 17:55
  SPEAKER_01: But I would say that they are concerns for any enterprise because we're all because there's always a concern about IP theft or exfiltration.
• 18:04
  SPEAKER_01: Okay.
• 18:04
  SPEAKER_01: And so what I'm getting at is that in the supply chain, a financially vulnerable supplier becomes a target for outside capital because they're looking to survive, right?
• 18:15
  SPEAKER_01: And so if they've got two or three choices, or even only one, to survive, and it's take some money from this entity that's coming in from an unfriendly nation, this is happening all across industry in very concerning ways.
• 18:27
  SPEAKER_01: And again, you're getting this picture of connecting two different data sets, connecting entities, such as an investment entity over here and a small supplier over here, and getting to a very significant risk insight that you need to go and act on before it becomes a problem for your enterprise.
• 18:45
  SPEAKER_02: And that kind of visibility is just not there.
• 18:48
  SPEAKER_02: And who would have thought that that hidden morale issue would risk the organization, put the organization at risk?
• 18:55
  SPEAKER_02: I mean, that to me is so tremendously exciting that you can find these hidden signals, these latent attributes and these latent signals that are immensely impactful, immensely impactful, but you just don't even know they exist.
• 19:08
  SPEAKER_02: You're blind to it.
• 19:09
  SPEAKER_02: You have no idea.
• 19:10
  SPEAKER_02: And yet it's like it's a systemic risk.
• 19:12
  SPEAKER_01: It's going back to that misconception that, you know, I couldn't imagine solving it, so I don't need to.
• 19:17
  SPEAKER_01: And because I don't need to, I effectively have, and I don't need to think about it.
• 19:21
  SPEAKER_01: That is literally where quite a lot of the global enterprise market is on this topic.
• 19:26
  SPEAKER_01: And of course, it's such a massive opportunity because and you mentioned it earlier.
• 19:30
  SPEAKER_01: There really is genuine competitive advantage for enterprises that take that that go after this as an opportunity and see I really can know more.
• 19:41
  SPEAKER_01: Yes, it takes effort and it takes time, um, but it you know it's worth it.
• 19:46
  SPEAKER_01: And these companies bring themselves to a point of of using their supply network as a true competitive advantage in what is ultimately only a more dynamic and competitive market by the day.
• 19:57
  SPEAKER_02: I'm this is the type of innovation and the type of application of AI that I think is is truly revolutionary, right?
• 20:05
  SPEAKER_02: Because these are things that that defied a solution.
• 20:08
  SPEAKER_02: And the solution to this is not only that data aggregation and integration and uh and correlation, but also the overlay of all those insights and those connections and those patterns that start to surface when you have this.
• 20:22
  SPEAKER_02: And then again, asking the right questions and understanding how this is impacting the organization.
• 20:28
  SPEAKER_02: This is tremendous.
• 20:29
  SPEAKER_02: So now one of the things that we talked about, and we alluded to this uh earlier was you know, people kind of try to say, Well, what is this, what does this boil down to, right?
• 20:38
  SPEAKER_02: What's the level of risk?
• 20:40
  SPEAKER_02: Is it red?
• 20:40
  SPEAKER_02: Is it yellow?
• 20:41
  SPEAKER_02: Is it green?
• 20:42
  SPEAKER_02: Is it is it uh 75 points?
• 20:44
  SPEAKER_02: Is it 90 points?
• 20:45
  SPEAKER_02: Is it 50 points?
• 20:47
  SPEAKER_02: Talk about why that aggregate, while it may have some meaning, talk about the little the elements that make it dynamic and the nuances and the context that make that so difficult.
• 20:57
  SPEAKER_01: Everyone wants an easy button.
• 20:59
  SPEAKER_01: Single score, single score, wouldn't that be wonderful?
• 21:02
  SPEAKER_01: And we're always pushing towards it.
• 21:04
  SPEAKER_01: I mean, absolutely it is about distillation, but in practical reality, the reason that we found early efforts to distill supplier risk to a simple score or high, medium, low really were it was it was convenient, but it wasn't accurate.
• 21:20
  SPEAKER_02: And my top example, it's not meaningful and it's not actionable.
• 21:25
  SPEAKER_01: It might lead you down a course of action that you could very well be wrong, so which is counterproductive.
• 21:30
  SPEAKER_01: So so my top of mind example on this is the fact if you look at the deep interconnectedness of of the global supply network uh with upstream materials from China, okay, in the industrial supply chain.
• 21:43
  SPEAKER_01: It's very obvious and easy for for an aerospace defense or frankly any company here to say, Well, I don't want any China in my supply chain if I'm if I'm in the United States.
• 21:51
  SPEAKER_01: It's just not possible.
• 21:52
  SPEAKER_01: Right?
• 21:54
  SPEAKER_01: Is cornered on certain materials, it's decades of evolution of this supplier network.
• 22:00
  SPEAKER_01: That means some some material is coming from China.
• 22:03
  SPEAKER_01: It may be going through other countries, and we just don't have the manufacturing capacity here yet.
• 22:08
  SPEAKER_01: That's being addressed.
• 22:09
  SPEAKER_01: All kinds of upstream commodities, right?
• 22:11
  SPEAKER_01: And so a risk analyst looking at this might want to say, okay, well, that's high risk because it's coming from China, and therefore I want to take it out.
• 22:18
  SPEAKER_01: It's not possible.
• 22:20
  SPEAKER_01: The whole product line is going to go down.
• 22:21
  SPEAKER_01: And so what that analyst really has to be doing is saying, is this level of risk acceptable?
• 22:26
  SPEAKER_01: Can I continue to produce with it or do I really have to stop?
• 22:29
  SPEAKER_01: Because if stopping is going to cost$20 million to my business to go get supply from a more expensive source or stand up my own production of that particular component or some combination of these alternatives, if there's a$20 million cost to that, what I really need to know is that greater than or less than essentially the value of this risk.
• 22:48
  SPEAKER_01: And this risk has nuance to it where, you know, for some reason you might determine that it's absolutely acceptable.
• 22:57
  SPEAKER_01: And we're just going to keep an eye on it.
• 22:59
  SPEAKER_01: We're going to look to mitigate.
• 23:00
  SPEAKER_01: We're going to prepare with alternative supply in any way we can in case that pipeline gets shut off.
• 23:07
  SPEAKER_01: But we are determining that for the time being, we have to continue as is, because that is significantly better than taking a$20 million hit.
• 23:15
  SPEAKER_01: There's so much nuance in that that doesn't that cannot be expressed by a single score that, you know, supplier risk equals high because domicidal equals China.
• 23:24
  SPEAKER_01: Right.
• 23:24
  SPEAKER_02: Right.
• 23:25
  SPEAKER_02: Right.
• 23:25
  SPEAKER_02: And when you start breaking it down into different components, so you can have some kind of a score, but but it's really about, well, what are the elements and what are the factors that lead to that score?
• 23:36
  SPEAKER_02: So we have AI maturity readiness scores, right?
• 23:40
  SPEAKER_02: And you can get an aggregate, that's fine.
• 23:42
  SPEAKER_02: But what is what is what are the components of that aggregate, right?
• 23:45
  SPEAKER_02: Because it's lumpy, right?
• 23:46
  SPEAKER_02: It's not evenly distributed.
• 23:47
  SPEAKER_02: There's some areas that are very mature, some areas that are not mature, right?
• 23:51
  SPEAKER_02: And it's the the divergence of those scores that really have an impact, right?
• 23:55
  SPEAKER_02: So it's saying in this area we have a very high maturity, in this area we have very low maturity.
• 23:59
  SPEAKER_02: Well, that's going to be a problem because the lowest area is really the bottleneck.
• 24:03
  SPEAKER_02: And so you're doing the same thing.
• 24:04
  SPEAKER_02: You're saying, well, where are the bottlenecks?
• 24:06
  SPEAKER_02: And where are the risks?
• 24:08
  SPEAKER_02: The supplier risk is pretty low, right?
• 24:10
  SPEAKER_02: It looks pretty good.
• 24:11
  SPEAKER_02: But there are some elements of that that have very high risk, and that that's that can skew things, or or you know, that has to be weighed, that has to be deconstructed, that has to be understood at that granular level.
• 24:23
  SPEAKER_02: And then when it depends on the nuance and depends on the context in which you're evaluating that.
• 24:28
  SPEAKER_01: Well, completely.
• 24:29
  SPEAKER_01: Absolutely.
• 24:30
  SPEAKER_01: To what end, to what purpose, in what very specific context.
• 24:32
  SPEAKER_01: All of that is the nuance that leads to the the really dynamic nature of the model.
• 24:36
  SPEAKER_01: I think what's fantastic here is that the the people in procurement and looking at supplier risk should feel very secure in their jobs because this is not getting solved by an LLM.
• 24:46
  SPEAKER_01: Again, the LLM will very confidently give you an answer, and it could very well be extremely wrong.
• 24:51
  SPEAKER_01: The scale of the data and intelligence problem is such that it was not solvable before AI.
• 24:56
  SPEAKER_01: It can be vastly accelerated by AI, but it still requires a great deal of human judgment working with AI to get to the right best answers for the for the for the enterprise.
• 25:07
  SPEAKER_01: I just wanted to make one more point as you were describing.
• 25:10
  SPEAKER_01: In practical reality, we we're finding material risk findings in approximately 20% of supplier portfolios, and then serious situations where you've got to go take almost immediate action in approximately two percent.
• 25:23
  SPEAKER_01: So it's about 120 and 2%.
• 25:26
  SPEAKER_01: Previously, you had no idea who was the 20%, and you had no idea who was the two.
• 25:31
  SPEAKER_01: All you were looking at was the entire pool and thinking, well, I don't know.
• 25:36
  SPEAKER_01: And so it's that triage factor that says, look, your human force, workforce, can now be focused through this triage to point out that those two percent of the suppliers are the ones that need the attention.
• 25:49
  SPEAKER_01: And it's not the same, it's not just the top hundred suppliers which you've been paying attention to.
• 25:53
  SPEAKER_01: It could very well be one in the middle or some tiny supplier that happens to have a connection to your point of sale system that is the source of a very significant cyber vulnerability.
• 26:03
  SPEAKER_02: Right.
• 26:03
  SPEAKER_02: And and and even though that may be an insignificant supplier on the scale of things, they are on a critical path.
• 26:10
  SPEAKER_01: That's completely right.
• 26:11
  SPEAKER_01: If you just think back to the target, the the hack in which the retailer target lost approximately 60 million consumer credit card file.
• 26:21
  SPEAKER_01: That hack came in through the HVAC vendor in one of their stores.
• 26:26
  SPEAKER_02: Wow.
• 26:26
  SPEAKER_01: Okay.
• 26:27
  SPEAKER_01: The HVAC vendor.
• 26:28
  SPEAKER_02: Unbelievable.
• 26:29
  SPEAKER_02: Unbelievable.
• 26:30
  SPEAKER_02: So here's the other thing that I wanted to point out about this is that you know there's a lot, there's tremendous fear about job dislocation and about elimination of jobs and elimination of humans.
• 26:41
  SPEAKER_02: What this tells me is that you're actually going to increase the number of humans who will be able to act on this and analyze this because now you have data points that were not visible, you have risks that were not visible, you have nuances and details that were not visible, that you will need more human analysts to understand.
• 27:01
  SPEAKER_01: Is this General's paradox?
• 27:03
  SPEAKER_01: But that's exactly what we're seeing.
• 27:04
  SPEAKER_01: We are empowering and we're enabling.
• 27:06
  SPEAKER_01: We are opening eyes and ears, but but that that human brain still has a very, very long useful life in the context of what we are talking about.
• 27:17
  SPEAKER_01: And I say that just seeing how wrong uh the generic LLM can very easily be and and you know, with great confidence, uh produce an answer that doesn't stand scrutiny.
• 27:30
  SPEAKER_02: And and and and what we're saying here is yes, there will be fewer administrative or clerical roles to gather this data, right?
• 27:37
  SPEAKER_02: That's yes.
• 27:39
  SPEAKER_02: But now we do need that higher order of analysis and thinking and and problem solving because we because we didn't have that data in the first place.
• 27:47
  SPEAKER_02: So that's a perfect example of that shift in that in that demand for human knowledge and judgment.
• 27:54
  SPEAKER_02: You know, organizations sometimes will either abandon or maybe they they put less trust in supplier intelligence programs when they have bad data, right?
• 28:03
  SPEAKER_02: They suddenly don't uh trust the information, they don't trust the insights of bad data.
• 28:09
  SPEAKER_02: So there's this idea that you know you need to build confidence in the data, but perfect data will never come, right?
• 28:15
  SPEAKER_02: So how do you weigh that?
• 28:16
  SPEAKER_02: How do you balance balance that and how do you manage that expectation, data quality issues, and then the confidence in the results?
• 28:24
  SPEAKER_01: Absolutely.
• 28:25
  SPEAKER_01: So so as you described, it's almost the fundamental challenge in this entire domain, uh, the idea of getting 100% perfect data on 100% of your supplier network.
• 28:35
  SPEAKER_01: I mean, I don't want to call it impossible because why not ambitiously aspire?
• 28:40
  SPEAKER_01: We're always trying to get closer to that mark, but just being aware that it's a journey from, you know, no data and no visibility in many contexts through to getting some and then more.
• 28:51
  SPEAKER_01: And you're constantly increasing both the quantity and the quality of relevant data that's going into the overall model, right?
• 28:58
  SPEAKER_01: On a journey from essentially zero to essentially 100.
• 29:02
  SPEAKER_01: And so then to talk a little about what are some of the things that can be done to really ratchet up the quality and the confidence in the quality.
• 29:11
  SPEAKER_01: One thing is to multi-source data, even on a one particular data type.
• 29:17
  SPEAKER_01: So in our platform, for example, when I talk about the cyber security insight, the cyber resilience of the supply of the supplier networks, we began that journey with our first specialty data partner.
• 29:29
  SPEAKER_01: Before long, we um we integrated a second, and not long after that, we've integrated a third.
• 29:35
  SPEAKER_01: Today, our platform delivers insight from three completely independent providers of insight onto that same question, which is how cyber secure is that supplier of mine?
• 29:49
  SPEAKER_01: And um, I think quite obvious what start what happens is that simply when all three of those independent adjudicators agree, that in and of itself is a signal.
• 29:59
  SPEAKER_01: I still won't say that it's 100% true, but it's it's it it builds confidence.
• 30:04
  SPEAKER_01: Whereas in a situation where they diverge, in and of itself, that's an interesting flag that makes us see, again, it's that triage point.
• 30:12
  SPEAKER_01: It makes us see, well, here's a supplier that these three are disagreeing on.
• 30:16
  SPEAKER_01: Let's go take a look at that one and come to our own judgment about which of those views is is right.
• 30:22
  SPEAKER_01: So so this overlay of multi-sources, and we can do the same with financial data, geopolitical data, you know, reputational insight, and you're stacking these different independent sources on top of each other to build confidence.
• 30:35
  SPEAKER_01: That's just one of the approaches.
• 30:36
  SPEAKER_02: So we talked a little bit in our our uh prep call about how you know organizations believe they have reasonable visibility into their supply chain and their supply base or their supply network.
• 30:48
  SPEAKER_02: The strategic opportunity, when you have visible greater visibility, that creates a strategic opportunity, right?
• 30:55
  SPEAKER_02: To have that that greater level of fidelity.
• 30:59
  SPEAKER_02: So what is what can organizations see about their competitive position based on this increase in fidelity that they could not see before?
• 31:10
  SPEAKER_01: Absolutely.
• 31:10
  SPEAKER_01: Well, the the very interesting point is that uh many of our procurement teams do use the intelligence and the insight that they get from us on their supplier network also to have an understanding of how they are positioned against their competitors for particular elements of supply.
• 31:26
  SPEAKER_01: And so this is what comes, what happens when you illuminate the N tier and you go down to tier two, tier three, tier four of the suppliers by looking at large data sets uh that give visibility into who is supplying what for whom across global trade routes.
• 31:41
  SPEAKER_01: And you can start to z hone in on um certain choke points in the supply network.
• 31:47
  SPEAKER_01: Rare earths, but rare earth minerals you mentioned earlier, is a great example of this.
• 31:52
  SPEAKER_01: And then you can actually see the volume of shipments that are coming from certain upstream manufacturers to more downstream processors, let's say, including yourself, but also your peer group and your competitive set.
• 32:05
  SPEAKER_01: And so from that, you're able to get visibility into how well you are positioned competitively.
• 32:11
  SPEAKER_01: And that's tremendously important in areas where supply is constrained.
• 32:15
  SPEAKER_02: So, in in fact, having these metrics and this visibility and this nuance and this fidelity will add a competitive uh differentiation and competitive advantage, especially when responding to large-scale RFP.
• 32:28
  SPEAKER_02: You know, how solid are you?
• 32:30
  SPEAKER_02: Because I imagine that's got to be one of the scores, or will increasingly be one of the scores to qualify an organization to be in a critical procurement process, right?
• 32:41
  SPEAKER_02: Such as defense and industrial.
• 32:43
  SPEAKER_02: Yeah.
• 32:44
  SPEAKER_01: It comes back to capacity, uh, the capacity of that supplier to deliver, which is actually absolutely crucial to look at and and know.
• 32:52
  SPEAKER_01: And so to be able to understand, well, who else are they supplying to with what sort of volumes, while you're also monitoring and tracking what's their overall capacity?
• 33:02
  SPEAKER_01: That's how you actually get at that visibility and know how secure your supply is, from which you can plan your own production.
• 33:10
  SPEAKER_02: Right, right.
• 33:10
  SPEAKER_02: That's wonderful.
• 33:11
  SPEAKER_02: So AI is generating more supplier insights and faster.
• 33:15
  SPEAKER_02: So human judgment remains essential, right?
• 33:18
  SPEAKER_02: To to be able to process, interpret, act on that.
• 33:21
  SPEAKER_02: What is the right division?
• 33:23
  SPEAKER_02: Because again, this line is moving right between AI, between what the the work that the AI does and the work that the analyst does.
• 33:31
  SPEAKER_02: How is that shifting?
• 33:33
  SPEAKER_02: How is that evolving?
• 33:34
  SPEAKER_02: And what is the dividing line there?
• 33:36
  SPEAKER_02: Because again, we're getting rid of that analyst having to spend you know an hour on aggregating data, and they can do that in 45 seconds, and then they have opportunity.
• 33:46
  SPEAKER_02: So, how is that when you look at a procurement organization?
• 33:49
  SPEAKER_02: How is that being done?
• 33:50
  SPEAKER_02: How do they need to rethink their workforce and their the makeup of their department and the workflows?
• 33:55
  SPEAKER_01: It's very positive, and it's really all about the the this massive opportunity that's being unlocked with AI.
• 34:01
  SPEAKER_01: We operate in a domain where we consider taking the right action is very important, avoiding the wrong action is very important.
• 34:09
  SPEAKER_01: So call it high cost to error in in our entire context.
• 34:14
  SPEAKER_01: And so, with that said, that's what really guarantees and secures the role uh of the human judge in these critical supplier-related decisions.
• 34:24
  SPEAKER_01: Uh, do I buy more from that supplier or less?
• 34:26
  SPEAKER_01: Or cut them out entirely?
• 34:28
  SPEAKER_01: In what way can I renegotiate my renewal with that supplier based on all of the intelligence that's now at my fingertips?
• 34:36
  SPEAKER_01: All of those decisions we see still very much and for the foreseeable future, living with the human, who, however, has just been vastly accelerated and empowered, almost given superpowers, from the data fabric, the knowledge graph, and the LLM synthesis of that in order to make great decisions and cover much more surface area in what they can do.
• 34:59
  SPEAKER_02: The network effect here is very obviously at work, right?
• 35:02
  SPEAKER_02: The more suppliers you have information about, the more you understand these subnetworks and these supplier networks, you're getting you're building a more of a global picture of what all of these ecosystems look like.
• 35:14
  SPEAKER_02: So, two two questions.
• 35:15
  SPEAKER_02: One is how do you have that uh that firewall?
• 35:20
  SPEAKER_02: The network, the supply chain network is part of a conventive advantage, right?
• 35:24
  SPEAKER_02: That's part of the knowledge of how they do business.
• 35:26
  SPEAKER_02: So separate that for for competitors.
• 35:29
  SPEAKER_02: So on the one hand, a competitor might say, Well, geez, you know, I don't want to, I don't want to have you guys have this visibility, it might bleed over into that other area.
• 35:37
  SPEAKER_02: But at the same time, there is an advantage to having that increased fidelity across multiple supply chain networks.
• 35:44
  SPEAKER_02: So talk about that balance and how you mitigate that.
• 35:48
  SPEAKER_01: You're absolutely right.
• 35:49
  SPEAKER_01: There's both sides of that that coin.
• 35:51
  SPEAKER_01: And the first, the most fundamental, is absolutely the sanctity of the confidentiality and the privacy of data that's coming in from any one enterprise, you know, customer to us, uh, including who are their suppliers, let alone more details, but beyond those suppliers.
• 36:08
  SPEAKER_01: And so as a multi-tenant architecture, highly secure, our our data for coming from customers is extremely tightly walled off in one tenancy from another with absolutely zero leakage.
• 36:23
  SPEAKER_01: That's the that's the fundamental and the and the zero term with which everything starts, every single engagement with the company is absolutely critical.
• 36:30
  SPEAKER_01: Think of how Salesforce.com operates, right?
• 36:33
  SPEAKER_01: And protects all of that critical customer data of an enterprise.
• 36:36
  SPEAKER_01: We're exactly the same.
• 36:37
  SPEAKER_01: Now, with that said, your second point about real business value being possible to create, generate, and derive to the benefit of all by observing certain relationships in across the knowledge graph, there's also very real value that customers want.
• 36:56
  SPEAKER_01: And so the way that we've been able to square the circle there is to say that it's a first of all, it's an opt-in model, and then it's a give-to-get model.
• 37:05
  SPEAKER_01: And those are two dynamics by which an enterprise can, with full agency and control, say that under these particular circumstances, I am willing for some of the insight that really belongs to me to be shared.
• 37:17
  SPEAKER_01: At first, I'm okay with it.
• 37:18
  SPEAKER_01: So I'm giving explicit opt-in permission for that.
• 37:21
  SPEAKER_01: Very important.
• 37:22
  SPEAKER_01: And then I'm I'm doing it because I'm generating the benefit from the fact that someone else is doing that as well in our network give to get.
• 37:31
  SPEAKER_01: And so let me give you a very specific example of that.
• 37:33
  SPEAKER_01: In the aerospace and defense industry, there's approximately 70% overlap of suppliers, tier one through through tier N.
• 37:41
  SPEAKER_01: Very, very high overlap of suppliers.
• 37:43
  SPEAKER_01: This might be unsurprising as a specialized world with not huge amounts of suppliers upstream doing all of the different, you know, titanium uh bending uh activities that go on.
• 37:55
  SPEAKER_01: So when you've got a very large entity, it's not easy to bend titanium.
• 38:00
  SPEAKER_01: There's not that many people that know how to do it.
• 38:02
  SPEAKER_01: There's there's dependence, there's concentration.
• 38:05
  SPEAKER_01: Well, when one large enterprise of ours starts working with us, and then we start working with one of their suppliers, what you see is that to that first entity, we're actually taking on knowledge of their tier two, because we're analyzing the tier one to their supplier.
• 38:20
  SPEAKER_01: And what we've had in many cases, because these are two entities that that are very interdependent and work together, it we've we've been able to create a situation where one is saying, hey, if I've developed and essentially spent the time and the money uh to develop a key insight, a piece of intelligence about that particular supplier that both I depend on and you depend on.
• 38:43
  SPEAKER_01: I'm actually happy for you to have access to that because it makes me more secure as well.
• 38:48
  SPEAKER_01: And I'm essentially sharing um explicitly and deliberately some of the intelligence that I've created with you because I'm seeing that as a net positive in building efficacy, security, and visibility in our shared supplier network.
• 39:02
  SPEAKER_01: And we call that supplier passporting.
• 39:05
  SPEAKER_01: It means taking a report once it's derived and certain intelligence and insight, sharing it with others such that the entire network becomes more secure.
• 39:14
  SPEAKER_02: That's wonderful.
• 39:15
  SPEAKER_02: I really want to thank you for joining me and helping our audience really understand why supply chain visibility at this level of fidelity is one of the most important and underestimated applications of AI in the enterprise, right?
• 39:30
  SPEAKER_02: Because it's going to really make material difference in risk and profitability and ability to deliver all of those things, reliability, quality, all of those elements.
• 39:41
  SPEAKER_02: Thank you for your time today.
• 39:42
  SPEAKER_02: Really appreciate it.
• 39:43
  SPEAKER_01: Seth, I really enjoyed the conversation.
• 39:45
  SPEAKER_01: Thank you very much for having me.
• 39:46
  SPEAKER_02: And and to our listeners, thank you for tuning in to the AI, uh, early AI podcast.
• 39:51
  SPEAKER_02: Be sure to subscribe for more conversations about how AI is shaping the future.
• 39:56
  SPEAKER_02: We were just named as one of the top 80 podcasts, uh global podcasts on AI.
• 40:01
  SPEAKER_02: So that was a great honor.
• 40:03
  SPEAKER_02: And so please share with your colleagues and and uh subscribe and like and you know, do all those wonderful things.
• 40:11
  SPEAKER_00: Thank you for joining us on another deep dive into AI innovation.
• 40:15
  SPEAKER_00: Tune in next time when we introduce another industry expert and discuss how to maximize AI in your world.
• 40:21
  SPEAKER_00: The Early AI Podcast is sponsored by Early Information Science and Vector.
• 40:27
  SPEAKER_00: To learn more, visit early.com.
• 40:29
  SPEAKER_00: That's E-A-R-L-E-I.com.
• 40:32
  SPEAKER_00: Thank you for tuning in.