Data Governance and Trust Infrastructure for Scaling AI

Amit Shivpuja: If we were to equate AI to launching rockets, data's the launch pad, right? You need a good launching infrastructure to be in place. Then you can launch as many rockets as you want. So the thing I think a quote that I really like is garbage in is garbage squared out. So just basically making sure you can minimize biases, making sure you're transparent and clear about what data you have. The other piece of it is we have to get our data thing right. Like we've kicked the forgive me for saying, but we've kicked the can down the road too many times.

Chris Hutchins: Well, I have a special treat for you all today. I am live on location at Plenty Hollywood in Las Vegas, Nevada for the Put Data First conference. And joining me today on the Signal Room is a new friend, Amit Shivpuja. He currently is a director of data and AI and enablement for Walmart. And I'm really happy to be able to get to meet you and just to meet you at such an important event where we're here talking about AI. We're talking about the currency of trust, we're talking about the importance that's going to getting this foundation right, thinking about it from an ethics and responsible perspective is really built into the fabric of how our how we're we're designing everything. So you and I've talked a little bit about this, you know, from the topic that we want to talk about. It's really the hidden infrastructure of trust. And you said you can't scale AI without scaling trust. Tell me a little bit what that means to you and what does that look like in practice?

Amit Shivpuja: Sure. And first, thank you so much for having me and welcome the opportunity. Um, so if we were to equate AI to launching rockets, data's the launch pad, right? You need a good launching infrastructure to be in place. Then you can launch as many rockets as you want.

Chris Hutchins: So it's can be heat sensitive.

Amit Shivpuja: Well, it depends how you define heat. But no, the thing is that since AI learn patterns from the data, you want to get that as right as possible in order to minimize things further on in the thing. I think a quote that I really like is garbage in is garbage squared out. So just basically making sure you can minimize biases, making sure you're transparent and clear about what data you have, making sure that there's accountability because you don't want to deploy something and then have people come back and say, what do we do now? or you know, who's gonna fix this, et cetera. So it's all of those layers coming into play in order to make sure that the output that gets generated is actually trusted by the users. The thing with especially today's generative AI and all, it's so easy to use. And people tend to put more trust in it because the friction is so low. And it's very easy to lose that friction also because you know, if you keep giving wrong answers, either due to hallucinations or the training data that was used. So, you know, building that trust becomes extremely important. So it's transparency, it's accountability, and it's also showcasing how do you get and where this came from, right? Right. So, yeah.

Chris Hutchins: So it's an interesting concept when you when you're talking about the building the trust when you already got something kind of in motion, but how do you think about it in terms of where you launch from and at what point do you start bringing in subject matter experts and people who are actually on the front line and might actually use it? And how important is that?

Amit Shivpuja: I think you have to start that very early on because a lot of the context comes from them, right? I I wish we were in a world where everything was documented and all of that, which is not the case. But so even the users, the subject matter experts, you want to involve them as early as you can. In fact, it's worthwhile actually figuring out your requirements, your exact needs, and what business impact it could potentially have, even before you go down building anything. So that way you actually have some direction or some baseline against which you want to compare the outputs that you get. Right. So them being part of the journey is an extremely critical component as early on as possible. And the other benefit is that once you launch this stuff, they might become your biggest champions in the organization, helping with adoption, right? And also figuring out the roadmap into the future.

Chris Hutchins: Yeah, I think that's a it's a really interesting point you're bringing up. Because I I think trust can be lost before you even start. And I think it's really critical to understand that if you want human beings to benefit and leverage this capability, they have to believe in it. And they they're not gonna believe in it if they don't understand the purpose behind it and what it's intended to do.

Amit Shivpuja: Um there's a there's a lot of change going on.

Chris Hutchins: So in healthcare, one of the things that's always been problematic is that we often tend to design something first. And it's almost I call I call it the the field of dreams kind of approach. If you build it, they won't come. Well, if we build something really cool and exciting, someone would obviously need it. Um, but this is really not not always the case. Um talk about the sensitivity and the fear piece of it, if you wouldn't mind. How do you get people comfortable to actually highlight a workflow maybe that would be a candidate for automation on some level, even if it's just basic automation, when they're also balancing this fear that AI could take their job.

Amit Shivpuja: Yeah. Um, I mean, what impact it'll have from a job perspective is currently up for debate, right? There's camps on either side. But I think to alleviate the fear is, like I said, bringing them early on in the conversation, but it's also showing them value. Right. Let's say, you know, they're spending 10 hours doing something manually, and it brings them down to one hour, right? So that kind of manual effort, which can introduce errors and all of that, gets reduced. So it helps them be more efficient. But also, I think it's good to have a conversation on what the rest of the save time can be used for, right? I'm sure they have things that they want to improve, they have things that they want to do, or other challenges that their teams need to address, which can now be focused on because you're gaining those efficiencies and the time. Uh now that also means that people have to become some level of data literate and AI literate, right? Because that's the landscape, right? So the question then becomes how do you support them, how do you help them transition and pick that up? Now, everybody doesn't need to know all the technical details, but they need to know at least a baseline of what an LLM is, what it can do, what's an agent and stuff like that. So I think some of that time can be invested in upskilling and learning as well, so that you know they're ready for the next thing because you may not want to lose that subject matter expertise as well. So yeah, I think that combination is what will work.

Chris Hutchins: I think you you're touching on something that I think people really need to be much more cognizant of. Um you can stay in the place where you're fearful, of course, and that's the natural tendency. But the reality is you have the ability, you have the time now to take advantage of learning to use the capabilities for your own benefit. Uh-huh. And I think AI will replace some jobs probably at some point. But what AI is not going to do, it's not going to replace people who are have embraced AI and figuring out how to use it to their advantage and to be able to contribute at a higher level. So you mentioned something about saving time, which I think is something I want to pause on for a second and have you talk a little bit about it. As a leader, um, oftentimes when some kind of a process improvement is implemented, there is an or some sort of an order that comes down from leadership around how that time is to be used. What's your advice to leaders as they're leading through this kind of transition around balancing hearing and listening to the workforce that are closest to the action versus maybe making some uninformed dictator dictation of instruction and how to do that?

Amit Shivpuja: So I think it's twofold. One, I think as people like us who are at the bridge between leadership and the people actually doing the stuff, it's good to bring them together to kind of showcase what, if possible, options are. But I think what's worked for us is prioritizing based on business value or business impact. If what comes from leadership, we can showcase that business impact and the need, then that takes priority. But if we can make a case using the feedback and you know the examples that we have bottom up, that I think should take business priority for the same, right? And so it comes down to business value, right? So, and that there's only four of them. You can save time, you can save money, you can make money, or you can mitigate risk. So it's just a question of bucketing it into that and having that messaging, right? So then it's neutral, it's transparent, it, you know, you can have a conversation, you can tie it to goals that the organization has, right, etc. Right. So that happens. So but what I found is a lot of the top-down stuff is more in terms of hey, we need to transition towards this or adopt this. It's not so much do this, right?

Chris Hutchins: Type an actionist. I want to talk about the trust thing just a little bit more. So trust is something that typically takes time to build. Um as you're working through something is is um I'll say disruptive because it it tends to feel that way. Yeah, um how do what are the early warning signs that we should look for that the where maybe trust is starting to erode a little bit? So I want to I'll tell you a little bit more about what's behind that, but I'd love to hear your thoughts.

Amit Shivpuja: I think some of it is just reactions that you're getting, right? If you if you're keeping stakeholders in the loop frequently, you know, you build a prototype, you get feedback and all that. Uh you just the signals you get in terms of usage, in terms of feedback, the quality of the feedback. Even if you if you release the first version and you put it out there and see how it's adopted, I think those are all key signals. But I think as leaders, if we have the right relationships, we'll we'll get those signals as well, right? If they trust you, they'll be like, hey, I don't want to do this, kind of a thing, right? So that perspective is also really valuable. So yeah, I think it's being constantly in contact with stakeholders and the enablers to so you you're you're suggesting it's not a flyby activity.

Chris Hutchins: No. Okay.

Amit Shivpuja: No, because somebody has to support this going forward.

Chris Hutchins: That's right.

Amit Shivpuja: Um, especially something like an agent, it will drift, it will do something it's not supposed to do. What do you do in that situation? It's not fire and forget kind of a thing, right? So yeah.

Chris Hutchins: Yeah. So you well, let me tell you a little bit where that's coming from. So I actually was in a had a conversation with with a clinical psychologist very recently. And we're talking about trust and the erosion of it. And I think intuitively we kind of know that things have shifted in society here in the U.S. for in particular. Um, but the numbers are staggering in terms of how few people actually trust some of the basic institutions that they 20 years ago, if you'd have said, you'd trust your government, people would have said yes, eight the at least eight out of ten. Today that's different. Um, you can say that and apply it to almost any kind of societal um role that someone plays, whether it's clergy, law enforcement, or whatever, it's kind of across the board. But people not trusting their leadership in the organization that they work for, or not even trusting their colleagues is something we have to be cognizant of. And so I really love the fact that you're you're bringing everything back to this human relationship and the need for that constant monitoring and the back and forth opening, open dialogue so that we're we're really not missing those. Um kind of wrapping up our time. If you're gonna advise an organization, well, we you can pick an if it could be healthcare, it could be retail, any organization. We're at this really amazing pivot point where we're talking about some disruption that's probably, I don't know, 10x, 100x, what the internet was in the 90s. Um, what are some, if you tell two or three things that you you would encourage leaders to be really mindful of and cognizant of as they're trying to navigate and lead their organizations forward through this transition?

Amit Shivpuja: Uh one that comes to mind is just building off the trust conversation we have, which is the trust but verify piece of it, right? Uh trust the technology, trust the teams to deliver, but verify whether it really does what it's supposed to do, right? It's really key to do that, especially given the amount of automated capabilities that these tools have. Because unlike the internet, AI is building the tool in some cases. It's not AI doing something quicker alone, right? So it's got its own capabilities. The other piece of it is we have to get our data thing right. Right? We've kicked the, forgive me for saying, but we've kicked the can down the road too many times. And that foundation, even with getting, even in an idealistic scenario where we get the data right, we're still gonna have biases. We're still gonna have, you know, patterns that occur because of the way the business is set up or the data that's collected and the like. So there's still a lot of work that needs to be done. But if you don't do this, then you can't do that. The third, I'd say, is human in the loop is non-negotiable.

Chris Hutchins: That's important. I love that.

Amit Shivpuja: Uh just the fact that whether it's verifying things, whether it's having them involved in the thing, et cetera. Uh I mean, at least for some time we're building these things for a human being. Right. Right? So somebody has to represent the human beings in the So the third thing I'd say is that, yeah.

Chris Hutchins: Oh, that's outstanding. Well, if if someone wants to get a hold of you, they've heard you're talking today and they they want to have a engaging conversation around what do they how do they need to proceed with this AI journey? How do we get in touch with you?

Amit Shivpuja: Sure. The best is to get me on LinkedIn. Uh so you'll find me on LinkedIn. But uh, if people are interested in the my thought and a couple of other frameworks that I use personally, I put out a book called The Data and AI Compass. It's available on Amazon. Uh and it's available both in soft and in Kindle versions. And uh that's another thing. And I also blog quite often on my Substack, which is datacompass.substack.com. So that's the other way that TFM.

Chris Hutchins: Fantastic. Well, I I will make sure for my listeners that all of these things are are available in in they'll be on our site and in the the show notes. Um it's been a pleasure speaking with you today. I am excited about this event. I think we're gonna have some more some more fun. It's already been a good start. So yes, yeah. Thanks again for being on the Signal Room. Appreciate you. That's it for this episode of the Signal Room. If today's conversation sparks something in you, an idea, a challenge, or a perspective worth amplifying, I'd love to hear from you. Message me on LinkedIn or visit SignalRoomPodcast.com to explore being a guest on an upcoming episode. Until next time, stay tuned, stay curious, and stay human.