Sandeep Anand on the Biggest AI Deployment Challenges Facing IT Leaders

Guys, thanks for the throw. We're here with Sandeep Anand, who's vice president of Machine Learning Solutions for Infor, and we're talking about, well, the five things that are keeping IT leaders up at night about ai. Sandeep, welcome to share.

Thank Mike. Nice. Happy to be here.

All right. So I imagine there's probably a lot more than five, but at least there's five that comes to mind. But, um, lead us off a little bit.

You know, when you think about it, what should it folks be? Well, maybe actually worried about versus maybe worried about too much. But, you know, start us off.

Uh, yes. And, and, and thank you for the opportunity to talk about, uh, artificial intelligence. Um, so, you know, my role in Infor is about leveraging AI ML for, from a enterprise perspective, right?

And so if you think of what that means from an info perspective, we focus on business applications, ERP software, things that help, you know, businesses, help businesses be productive. Um, the number one thing is, um, you know, driving useful, measurable business gain out of it, right? And so it's, it's from a, from a benefit perspective, the, the number one thing to worry about is how is this helping a business improve productivity, right?

So if you think, what is the value? This is driving my organization in leveraging this because we know that technology works, we know that's used across the board. So the question just becomes, what am I going to get out of it?

That that'll be my number one consideration, And how do I evaluate that successfully? 'cause I think a lot of it, people are trying to figure out what's the math here on the ROI that makes sense? Absolutely.

Which is the, which is, you know, in some ways, um, the, the, the secondary point to it, right? Is, you know, always you are going to talk about, uh, productivity measurements. Um, you know, it's very easy to say, well, this is going to reduce the, the amount of time I need to spend in trying to figure this out, um, manually or in my current business process.

The other thing you can look at is, uh, this is going to help me be more, um, accurate. You know, when you talk about inventory optimization or anything to do with, uh, scrap management, right? So we are looking at, uh, you know, we are looking at challenges in your business that are impacting the revenue side or the cost side, or the quality side, right?

So those are good, easy metrics to, uh, anchor against because you are probably tracking those, uh, in some analytics, uh, part of your business. So an improvement of those. Um, you know, the other metric would, could be considered around timeliness, right?

So we talk about productivity, savings, business benefit, um, being able to do it faster, right? Will be another thing to consider. Uh, and it's not a, uh, or it's an and, right?

Is the, is three prisms of it. The fourth thing, uh, is also how is it helping my workforce, uh, be productive? You know, allowing them to, uh, be more satisfied because in some parts of our, our ecosystems and manufacturing and distribution, uh, the retention of these workforce, so the ability to leverage, uh, you know, do more with less of sorts, how can you get them, uh, excited to leverage these technologies to drive that better business outcome is also very important.

I think part of the issue too these days is that there's no shortage of these projects, but there's only so many resources and so many people with the skills. So, um, is there's some way to think about prioritizing these things. 'cause I don't think we can do everything we wanna do all at once.

Yes. And, and, um, um, it's a very important question, and I think it talks to the, the company culture. Uh, you know, if you think of, uh, what do you want your business to be in the next quarter, next half a year, next year, where is your business priority, right?

So from the prism of what are you looking at? Are you looking at it from a revenue perspective, uh, business transformation to, you know, do more with less? Are you looking at, um, a challenge to the way your business is run because you have supply chain issues because of the current macro conditions, and it's really important for you to stay ahead of your competitors, right?

So that's where, from the culture perspective, what is driving your businesses push and how are you enabling your, your, your team and driving and pushing that would ultimately be the best way to push this forward. So it's not a side project, it's critical for your business to be successful. And, and, and if so, how are you then getting your team on, on board, right?

And be the, the right prism of how to start, which use case within doubt. I always say pick something that impacts revenue on cost, you can go wrong, right? But also from a financial forecasting perspective, it's also good to understand how your business is moving forward.

So generally, that'll be another consideration. But, um, I, I am always partial to anything related to supply chain. Uh, when you wanna get started, It also seems like there's more of a separation of concerns these days, and the data science teams are maybe focused a little bit more on training and maybe the creating of the initial AI model, but the IT teams are taking more responsibility for the inference engines and the deployment thereof.

So is that part of it more where the IT leaders are more concerned about things versus, say the actual training of the model itself? It's, it's, um, it's a very important question, and you talk about the blurring of the lines that have traditionally governed, uh, these types of projects. And you know, of course when you even talk about generative ai, right?

And agents, and how that is also causing further blurring whether business are now able to do things that the IT or the science team were able to do, right? So your example, you're talking about the science team builds the models IT team manages, maintains, executes on the, the models in the terms of the value. And then with gen ai, you have the business also coming in, Mike, and also coming in and saying, well, I can automate these things, uh, or I can leverage, uh, a gen AI type of assistant to drive better analytic knowledge.

So I don't need as much investment in analytic dashboards, right? And so it's important question of, uh, as your organization, uh, evolves and depending on how they're set up, how those roles can play nice together, right? So everyone is now on everyone's turf, which is good and bad depending on how you manage it.

Mm-hmm. Also, I think that there's some concern about, well, what should I actually go build myself versus quote unquote buy? Because there are software vendors that I currently rely on who are building AI and ML into their various offerings.

So I'm trying to figure out like, uh, I don't think I want to be in the position where I just spent a year building out an AI model to wake up one morning and figure out that my software vendor's given it to me for, you know, a nominal extra cost within the application itself, Right? Um, the age old, uh, challenge with the build versus buy, right? If you think of it, and with, uh, with, uh, if you think of cloud providers, it's no longer, um, uh, just simply build versus buy, but which part of the things are you building versus which parts of the things are you buying, right?

It's a, it's, it's an unenviable position of how to make those choices. Uh, I was recently, um, we were recently talking about, um, you know, how do you from a prism of an end user navigate these challenges that are, uh, or opportunities, depending on how you look at AI and the excitement, uh, and potential of it is, do you look at a vendor as a monolith saying, give me, give me the outcome and you just take care of everything? Or are you in, um, in the kind of a maturity phase where you're saying, I, you know, um, if you think of a shirt, right?

I I don't want the shirt, I just want you to take care of the sleeve. I'll take care of the buttons and you help me get the fabric. Right?

And so I think that is kind of the, a case by case decision. I think, uh, if you're starting off, uh, again, if I follow up my, my very simple example, just get the shirt, and then ultimately when you get comfortable wearing your shirt and you want different shirts, you can start trying different things. Uh, but once you, you know, as you get more and more into it, you are going to want to have more control on those decisions yourself, right?

We, we haven't talked about IP and data rights, but ultimately over time, uh, there's a gravitation towards I want to do something special for myself, but I do want things that are very easy and common just to be handled by my partner. Mm-hmm. The other thing, it's still unclear to me, but who's responsible for securing all this stuff?

Because to your point, um, not only are bad guys trying to poison models, but in some cases they're trying to just steal the model entirely because, well, it's valuable ip. So, um, do we need to rethink security in the age of ai or will we just kinda have the same old it does the deployment and security people secure it? Um, I, I, I definitely think that we have to keep up with the times.

Um, right? If you think of, um, if you think of all the challenges we have, uh, with some of the items, the examples you just gave in the news, where I think the, the agents, the LMS can now catch security intrusions as they're being tried, injected into your organization. So there is a, a new found respect for the power of, uh, using, um, advanced technologies to protect you beyond just the data and, uh, uh, kind of people protections of, you know, keeping things behind firewalls, right?

So those will continue being a necessity looking at, you know, LMS that allow you to be more secure, making sure your partners with the, those models aren't leveraging your data to learn from, right. And causing your IP being pushed is important. So I think, uh, it's evolving, right?

We'll continue adapting with all the, the advancement of technology, but I don't think you can, uh, have a stance that I want to firewall myself. I think the trade off of protecting overly protective, uh, uh, setups will ultimately limit you from being able to power off all the advancements in technology. Mm-hmm.

Is there something that we're overlooking and maybe not paying enough attention to as IT leaders? Because maybe we're spending too much time obsessing about one thing and not paying enough attention to something else. I think that's why we have such white hair, right?

Because I don't think that's ever going to change. There's always something, Mike, that we will always not be taking care of, right? I think when, when we got the, um, when we got Gen ai, we were like, oh, this is great.

And then we were like, oh my God, but this is just taking all this information and pushing it out. And then we were like, oh my God, this is telling me the wrong information. What we should have some guardrails.

And it was like, well, it's just telling me this information I can't do much without. So I think that's a, I think a two year on me. That's a IT problem.

We all cherish and love to have, not business, don't you think? Right. So we always be something.

What impact will all this have on the role of the senior IT leaders out there? Do you think it's gonna elevate their positions a little bit within their organizations and, um, how should they kind of view themselves within the context of an organization? Because you could argue conversely that AI is really a line of business issue because you've got people who are experts in a process and they need to understand how the models work and validate them, but at the same time, you're probably not gonna get very far without it.

So how's the relationship gonna change? Uh, it's always, it's always symbiotic, and I use that word intentionally, right? I think from a, um, investment perspective, there is always a need to continue investing in it.

Uh, not only from a security and governance perspective, but project management of these transformations. You're always going to look at using different tools to drive your business. You're never gonna have a monolithic single, uh, software policy.

And this is where it is specially suited when it comes to technology and technology change. We are also looking at, you know, how do we be, um, how do we get the right evangelist from a business perspective to help drive those productivity benefits, those revenue benefits and things like that. So I think that, I think there was always a need for it.

I think where with gen AI and agents agent tech ai, I think we're starting to blend some of the business and IT responsibilities, especially when it comes to analytics. But now you're also getting a new level of understanding around, to your point, the security, the governance, the concept of how is my data interpreted in my organization or outside as we start looking at, uh, you know, some of the new technologies around the agent to agent, where the hope is two businesses can talk to each other without requiring anyone, right? And so I think, um, older paradigms get replaced by newer paradigms where it has to change and upscale, but also provide more and more help, and even if not oversight, but some way to control the, the, the use of this AI in and outside our organizations.

Mm-hmm. Am I gonna see organizations build these kind of massive AI models that everybody's worried about what the ROI is gonna be? Or is it more likely that for the purposes of a business, I'm gonna rely more on smaller models that are distilled maybe from those bigger models, but ultimately they don't need to be nearly as big and as difficult to manage?

And should I be, you know, focusing my efforts on, uh, uh, extent of AI models that are maybe better trained with a particular set of vetted content? I, I think that, I think that's unlikely, um, that we would have companies invest in their own large language modules. To your point, uh, soft, uh, smaller language models, a very small language models or domain specific language models will probably be, uh, for certain companies a better preferential treatment.

I mean, if you think of our phone, it's this example of a, a small language model. If you think of the, the fact that when you type a email, you get the autocorrecting the information, right? That is not what a SLM is.

But at a basics level, we already at some point, uh, uh, able to take advantage of very specific use cases, uh, around large lang around, uh, language modules. But, um, you know, I just don't see that big need for your own LLM. I do think there will be more and more adoption of, uh, help my business make better decisions.

Uh, but it, I don't think it needs to be a small language model. I think you could do something even more pragmatic very quickly, uh, with the technologies in place. And I think that's where a lot of, where from an info perspective, we are also seeing the initial stepping into the, the waters of sort, is that we, you know, we want to help us with our procurement process.

We wanna help with our customer service, uh, and being able to look at documents and help with knowledge bases to help kind of troubleshoot information. Those don't require such a large investment. In fact, those things could be almost out of the box in this day and age, the way technology is moving, right?

And we see a lot of that initially. Um, I think some will go into the s SLMs, but, you know, I don't think that's fully needed to get the advantage of, um, the benefits of ai. And in my conversations with people that it seems like they're starting to realize that there is this need for more context when you're working around those LLMs and you hear the phrase context engineering is kind of the next super set of prompt engineering.

And that's right. But then it seems to me, as I look at it more and more, it's as much art and skill as it is science. So how do I make sure I'm putting the right data at the right place at the right time to ensure, uh, the context that I need to get some sort of output that's more reliable is current.

It, it, it is a million dollar question. I think that is, uh, one of the questions that is keeping, uh, uh, organizations up at night, right? They, this is, uh, this is part of the reason it jobs are always going to be, uh, required, right?

Because we started off with, you build a software and then you're, you're, you're good, right? You just have to worry about the software being used by the business, and then you realize, oh my God, this is just connecting to other software. And then you're like, oh, but this is putting in data that needs analytics dashboards.

And you're like, oh, these dashboards are just telling me what's in the backend. So we need ai. Now we have ai.

You're like, oh, I don't know if this AI is the right AI for me because it's just telling me things that probably not true and need to be curated or need to be better managed. And then when they are managed, you're gonna say, well, I want to automate this so that I don't actually have to worry about it. I don't need to have analytics.

I don't need to have software. Like you see the thing, right? We build a software and you build analytics, and you build the ai, then you build the, the, the ages that run the AI on the software so you don't have to use the software and then what's next?

What's next? What's next? Right?

And I think, I think that's the, that's why we love this job, right? We are moving so fast and moving so fast in some places that I didn't think we would be able to do in five years ago, right? And so I'm very happy with that.

So in some ways, half glass full, right? That's the kind of thing and the way I look at this, yes. So what's your best advice to folks?

Because the opposite of the glass that's half full is the glass is half empty, and a lot of folks are intimidated by all the moving parts here, and almost to the point where they basically freeze and do nothing and are just kind of watching and waiting for things to kind of maybe evolve. But, um, what should folks be doing to that? I, I think we'll go back to the, the premise of the conversation.

Don't worry about, don't worry about all the hype around the technology and the promise of it. I think distill it to, um, what's important for my business to do now, next, later, uh, how am I able to quickly do the now, next, later, uh, with the people, process and technology at my fingertip? How do I measure?

Uh, and, and those are, you know, simple principles not related to ai. That's just anything. Anytime you're investing in any, um, initiative, you, you look at, you know, what do I want out of it?

How am I gonna measure it? How quickly can I get it pragmatically without having a big, large investment? And is this gonna be a market differentiator?

And then you just take that action. And, and the reason that's so important is because it's tried and trusted. It is not about ai.

It's ultimately about how you want to succeed. And therefore, you're less likely to worry too much about, am I doing the wrong thing? Because it's the thing you every business should always be doing.

And if this helps that, then you're easy more likely to do it versus trying something new and unexpected. Because at the end of the day, we're still talking about quick ROI, measurable, RROI, bringing people along, measuring business outcome, and then, you know, rinse, repeat, right? And so, as long as you keep it in the prism of what everyone does, you're less likely to just, you know, be paralyzed with this indecision, right?

And I think that's how I think everyone should think of it. All right, folks. You heard it here.

Despite all the hype and concerns about the fear of missing out, turns out slow and steady still wins the race. Hey, Sandeep, thanks for being on the show. Thank you so much.

All right. And back to you guys in the studio.