Data Readiness for the AI Economy

Unlock the true potential of your organization in the AI economy, where data quality, strategic focus, and reusable solutions are the keys to transformative success. In episode, host Vinnie Schoenfelder welcomes Cameron Snapp for a discussion on why AI is only as powerful as the data behind it, and how companies can avoid common pitfalls while accelerating innovation, discussing:

• Data quality and governance are essential for effective AI.
• Start with focused, well-defined problems—don’t try to solve everything at once.
• Reusable methods and accelerators help scale AI and integrate new data sources.
• AI changes job roles; those who adapt will thrive.

Transcript

Vinnie Schoenfelder:

Hello, and welcome back to CapTech Trends. Today we're talking about data solutions for the AI economy. I'm Vinnie Schoenfelder. I lead some innovation areas within CapTech. And with me I have Cameron Snapp. He's a technical director and capabilities lead for the data analytics practice. Welcome, Cameron.

Cameron Snapp:

Thanks, Vinnie. Great to be here.

Vinnie Schoenfelder:

Yeah. You had reached out to me a little while ago about having this podcast, and we've done a lot of different AI podcasts over the year. So when you say data solutions for the AI economy, let's help set this up in terms of what you want to talk about that's a little bit different today.

Cameron Snapp:

Sure. So I think AI is very much in the future for agentic systems and it's really important that it's a data play. So companies that are really excited about adding new capabilities in the AI space have to be thinking about the maturity of their data platform. So when I think about data solutions, it's the holistic way that we're going to catalog, ensure quality, and have good governing processes about how the data gets used so that the agentic answers that are being created at light speed are accurate and transformative for the business. So data is imperative any more than ever before.

Vinnie Schoenfelder:

Right. So let's talk about the core of it. There's a lot of pilots going on, we're seeing some early wins with people who are implementing it realtime, both operationally and functionally for apps and services. What do you see as people doing well or correctly and working, versus where people are stumbling and having challenges?

Cameron Snapp:

So I think they've bought into the idea that I just pitched, which is data is essential. So I've seen a lot of maturity in obviously the migrations toward the cloud and using big data platforms, so a Databricks, a Snowflake, an Azure Synapse. The idea of getting your data in a usable way and in a centralized data lake. That's a trend that has happened for 10 years.

Vinnie Schoenfelder:

I was going to say I remember those slides from 10 years ago.

Cameron Snapp:

We've done it finally. You don't have to convince clients anymore that that's what they have to do. They understand that. But I think where there are some misses is making that data prepared for AI, and so that's the next thing we have to do on that maturity model. Both on the metadata cataloging side, but then on the transformation and performance side to make the right data that they want to make your decisions with easily available to AI scraping and agentic processes.

Vinnie Schoenfelder:

There's a lot of analogies to that. AI doesn't make your data better, your data has to be better for AI. I'm reminding of things like automation, process automation. Where if the processes were poor, you're just doing poor things more quickly and more often.

Cameron Snapp:

Yeah.

Vinnie Schoenfelder:

And I think about music. Good stereo systems, they don't make bad recordings sound good, they make them sound worse because they're accurately recreating how bad it is.

So talk to me about has that changed? I don't think the necessity has changed. But does the impact of getting it wrong, has that changed?

Cameron Snapp:

The impact, it's way worse now because when the humans were in the loop, if you will, where they were looking at the data, it was really easy for them to recognize with their business context experience that this recording sounds really bad now. Or this number that's being produced in this report's wrong, so they could go track down why that is. And they would know better then, to make a decision with that bad data, or that misguided or incomplete picture of the data.

Now with the processes running on their own, if you figure just AI's automating some of these things, it's just going to roll with the bad data. So it's going to be like, "Well, this recording sounds as bad as it does, but I'll work with it and move forward anyway." So that's a concern is that it's not going to correct your data, it's only going to amplify how bad or misguided it could be.

Vinnie Schoenfelder:

Yeah. I know we had a question laid out for this later, but I'm going to pull it forward. Does it mean you have to put your AI hopes and dreams on hold until you get all your data correct? Or is there a parallel path where, as you are focusing on high value data cleanliness and accuracy, and all the views of data, can you do these things in parallel as you're maturing your data, you're implementing AI? Or is it more in large chunks?

Cameron Snapp:

Yeah, the days of large chunks are over, so I would never recommend that. I do think there's opportunity for parallel. I think the most important thing for companies looking to figure out where they can accelerate some things or automate some things in the AI space would be find a very well-defined problem and define what good looks like, and then measure your proof of concept towards that. So make sure that the data that's relevant for that specific question is in a good spot. So spend the time now, get that data proper formatted. Don't worry about your whole system, don't boil the ocean at once. Worry about that one little slice ecosystem and then work forward. And in parallel, figure out what you're going to learn, and figure out what your good patterns and methodologies are going to be.

So the operationalization of AI systems is all about re-usability. So if you spend a lot of time perfecting one AI solution with one set of really good, clean data, you then can have a bunch of multithreaded approaches to fixing data for other spaces. So start stacking and doing that flywheel approach.

Vinnie Schoenfelder:

I'm going to pick on one point you made, which is find a really good place to start. And I would challenge that in that I think this is so transformative. That people are bringing up historical problems that they're trying to solve and they're not thinking in a new way, because I think this unlocks things we've never even thought of or dreamt of. So we can't go back to the same problem domains with the same people managing the same problem domains to get that true spark of an innovation and creativity.

So I'm going to turn that a bit to say we've all looked at that MIT report saying that 90-some percent of AI projects failed to prove an ROI.

Cameron Snapp:

Right.

Vinnie Schoenfelder:

And I think that goes to this find a good problem. So help me understand, if this is so transformative and we're trying to find the right problems, why so many people aren't?

Cameron Snapp:

So the whole idea of ROI for AI systems, I feel like a lot has been just what's possible without understanding what they wanted to achieve. So missed ROI for companies is about, "We spent a lot of money, we added a lot of systems, we bought licenses to LLMs, and we've told people to go get comfortable without a strategic direction for what we want to achieve once they get that comfort."

So one of two things might be happening. This might be just our comfort adoption stage where everybody is understanding this is here, they've bought into the idea, but humans are a little slow to change quickly. So it's going to be another 18 months of a lot of churn and bad ROI, then everybody's going to start seeing some efficiencies.

The other side of the coin is without those reusable patterns in place, I'm really concerned that a bunch of different what were supposed to only be proofs of concepts or pilots that people are using to get comfortable are going to make their way into production, and then it's going to become an unsustainable support model and you'll have a lot of potentially automated things that aren't cohesive and aren't aligned.

So a lot of companies today have started to go with a more federated data model in terms of how they put their business operation. So they'll do little portfolios based on the entity space or the domain of expertise. So customer and marketing, locations and assets, or even product and supply. So if you let each of those divisions do their own AI methodologies and outcomes and they put stuff into production, I do think you would have a really big miss on ROI because you're going to spend a lot of money to make each of those individually good when they're not brought together with one overarching strategy.

So this isn't a situation where I'm proposing bring in IT and have them make some yearlong effort to do one thing really well. But what I'm saying is let's do something in a few weeks to a few months, track the ROI, find the patterns, and do a lot of teaching, and then do your scale-out. So then you can distribute those reusable patterns and approaches to the individual portfolios and say, "Best practices, corporate guidance, this is how we're going to do this, everybody has to get aligned to it. And then this is the type of measures that we want to be achieving." So you can have your own business specific use cases in those domains. I'm not here to say that everybody's outcomes or problems they're trying to solve should be the same. Everybody's business is unique, but it's important that you solve them similar to how the rest of your company has found the best success.

Vinnie Schoenfelder:

Good answer, and it always gets me thinking about the modernization of the data. You don't finish that task.

Cameron Snapp:

Never.

Vinnie Schoenfelder:

Right. So as you're saying all this I'm thinking, "Gosh, is it now more difficult to bring in new technologies?" Is mergers and acquisitions more difficult? Because now you have a much mature framework and the organizations you're going to be bringing in or partnering with don't. So yes, you can unlock some great stuff with this, but are you also making it more difficult to grow in a more traditional business sense? Probably not, because whenever you do things right, unexpected dividends get paid, I believe that. But how do you manage that? Is it a governance model? Is there a group that's responsible for that? How do you make sure that as you are maturing, you're not blocking off other areas of business integration?

Cameron Snapp:

Yeah. The mergers and acquisition thing, we're seeing that with a lot of our clients that are in that space, and the first one always goes twice as long and twice as hard as they expect. But the really good companies that are doing that successfully spent a little bit of time doing the retrospective and say, "What did we learn? How are we going to improve this next time?" And we have to improve and iterate over time with those, not to beat the topic too much, but the reusable pattern and methodology to say as we do this, we have to bring them on board in a way that has worked for the internal workings of our company. We want to find a way to map their existing data assets into our core data model so that we can have a central version of the truth.

I've seen companies really not do well with merger and acquisition, where they just forklift in the entire data ecosystem and then they end up with two versions of, quote-unquote, "truth." So having a single view of a customer is nearly impossible because they didn't take the time to map the acquired company's data model or the acquired company's source application, if you will, into that lake house environment. So we at CapTech have built some reusable IP in this space specifically to make those type of acquisitions easier.

Vinnie Schoenfelder:

Using AI, I assume?

Cameron Snapp:

Using AI to build the code, but using AI to analyze the data itself, so there's two aspects there. So we can go down that path if you want to.

Vinnie Schoenfelder:

Well, yeah, let's jump into that. So we didn't do the technology for technology's sake, we saw a repeated business problem.

Cameron Snapp:

Right.

Vinnie Schoenfelder:

And we came up with IP that addresses that problem. So step into some of the problem domains that you and your team saw and how you created an accelerator for that?

Cameron Snapp:

So the one that's specific to mergers and acquisitions, it's really interesting, or even a lot of companies that are having similar data, but with different structures to it, is a solution we call Green Arrow. And that has an AI component to it, where it'll actually analyze the source, think of it as just a file or a table coming from a database from a core application, and it'll profile what those data elements are and if you know what your target model looks like.

So in this case, we have files coming in from claims or insurance policies on the left side of your architecture diagram, coming into a single source of truth model, that data model that says, "This is how we structure our policies." So what it can do is profile the source against the target and say, "This is the right way to do the mapping, here are the data quality cleansing exercises we would suggest." And then this gets away from a manual human person having to look at hundreds of rows of data, hundreds of columns of data and do that mapping manually. So we can use AI to accelerate that process, add it, and then allow us to onboard more assets quickly, but also do it a lot more accurately.

So that takes away some of that pain that we said we introduced with the mergers and acquisitions idea. If you want to bring more sources of information into your lake house environment, or bring more companies that you weren't familiar with before, or make it easier on your clients, your customers to send you data that you need to get into your analytic environment. So that's just a good example where AI can really move the needle of a task-oriented thing that was very burdensome, that was very manual, you can accelerate that and you can help to grow your business into new areas.

Vinnie Schoenfelder:

So you have much more depth in this area than I do. I remember these tools though, the data loading and transformation tools that vendors would push that would automate a lot of that work. Won't this just be a feature and there's larger vendors? What did you do that was different?

Cameron Snapp:

It really goes into looking at the actual elements of the data and helping give guidance. So the other thing it can help you do is, say a state that was Pennsylvania can now be PA, but it allows you to say once it gets into the target model, it gives you a lot more flexibility. You don't have to ... If you use these vendor tools, you have to use a lot more of their rigor and their output capabilities, whereas this, it allows for complete customization on the left side and the right side. And it also allows you to write different means of moving the data, you don't have to use a vendor's, however they're going to do ELT. You can use the Python code, or the SQL code, or the Databricks or the Snowflake code that you already have in place. So it makes you a little more agnostic because the process itself is just a Python library, as opposed to an ongoing expense that forces you into a technology that maybe isn't the right one for you.

Vinnie Schoenfelder:

So I can see this being useful for companies like CapTech, that we're doing consulting services and we're solving this problem over and over again in multiple clients. Is this identity of reusable tasks and creating accelerators relevant for more traditional companies?

Cameron Snapp:

Right. It's definitely a play for the consulting firms to be able to have what we call door-opening conversations. Of, "This is what we're seeing clients struggle with. Does it resonate for you? Here's how we've solved it." We're not trying to be a product company where we say, "This is a black box item that will just solve your problem and move forward." This is something that will help us accelerate you on the journey. So companies should be building their own accelerators, too. So it's not exclusively a consulting firm thing.

So I say the reusable methodology, the way you do your data ingestion, your ELT patterns to say, "This is how we're going to connect to our disparate sources. This is how we're going to acquire data on a high frequency basis. And this is how we're going to store, and have proper lineage and governance in our lake house." So the medallion architecture, the DataZone concept of your bronze, silver, gold. Doing that one way across your firm is really important. So CapTech has an accelerator that helps us do that called Adept, but specifically any company should be building their own single methodology so that it's easier to move fast, bring more data sets in, and have that comfort and confidence that the data's going to be of good quality.

So back to the original question that kicked off the podcast about the things that are going poorly, I think building data systems where it was just get all the data in, quality be darned, I think that has put people behind the eight-ball when it comes to that challenge where they need to have a single way to manage it so that they can apply what they need to from a transformational lens.

Vinnie Schoenfelder:

So the next topic I want to jump into, and this we got to keep at a high level because this could be an hour discussion alone. You're laughing I'm sure because you know where I'm going with this. The impact to the human-

Cameron Snapp:

Right.

Vinnie Schoenfelder:

... the human talent. There's a lot of sky is falling reporting in terms of all jobs are going to go away. Heck, even Bill Gates is saying how devastating this is going to be. We talked about this a bit. But what's your take on how AI is going to disrupt, well, our industry specifically, but also more generally, all other types of jobs as well?

Cameron Snapp:

Yeah. Being in tech, we have been talking about this a lot lately so I think a lot of people in all sorts of spaces are concerned about the same thing. So I think it's easy to have a completely dystopian view of everybody's not going to have a job, to a utopian view of the robots will handle everything for us. It's neither of those, and I think the key component for any person in an industry or in a job that's worried about losing their job to AI, I think you're going to lose your job to a person that knows how to use AI and that's the right mindset to have. So I think this is all about adding more capabilities, experience, comfort to your toolkit. I talked to a lot of people at CapTech around building a toolkit of things you know how to do well and I think this is certainly the one that everyone should be focused on right now.

So how can we leverage AI to make sure that our job stays? I think AI is very much going to take away a lot of tasks, not necessarily take away a lot of jobs. So if you have a job today that is exceptionally task-oriented, you want to find a way to be a part of when that thing gets automated, how can you offer insights and value to the firm making those types of things.

So we're seeing a lot of job cuts right now and they're using the excuse of AI, and I think that's very much just an opportunity to add cost reduction to their balance sheet. So the Fortune 1000 companies are very excited about saying they're adding AI so that their stock price can go up because that usually means there's going to be cost reduction in operation spend. I don't know that they've actually unlocked the new capabilities or opportunities with AI yet, they're just using it as a way to automate some processes.

I'm excited about the new jobs and opportunities AI's going to create. So a good analogy I've heard used around this is when ATMs were invented and started to be deployed, everyone thought, "Well, we'll just automate away the bank teller." But it allowed companies to open more branches and reimagine what bank tellers could do from a customer service lens. So I think it's going to reimagine what the workforce spends their day-to-day doing.

Vinnie Schoenfelder:

So it may get messy in the interim, but you believe we're headed in a good spot?

Cameron Snapp:

I think it's going to be very choppy here in the next couple years, and hopefully what we'll see is both the idea of better government regulation around what AI should be allowed to do. There's a whole data privacy, ethics conversation that we can try to have around this, but more specifically-

Vinnie Schoenfelder:

Well, it's a space race issue, too. If we have regulations on what we can't do and China doesn't-

Cameron Snapp:

Sure.

Vinnie Schoenfelder:

... where they're going to be creating things faster and more innovative than we are.

Cameron Snapp:

Yeah. Europe is already the leader on a lot of this. The reason that we have all of the cookie policies on every website you go to right now is because Europe passed legislation so every company had to adopt to that. And I think there's a lot of room to make sure that AI doesn't go completely unregulated or just unfettered, if you will. So there's a lot to worry about there.

But back to the practicalities of what people should be doing is thinking of ways that the company that you're in or the industry that you're in can take away some of the painful parts of your job. So there's a lot of repetition in reading documents and doing data entry. AI is already doing that. So you fill out your patient care form at your doctor's office, that can be scanned in and automatically, AI can not only do the data entry, but it can do the data quality assessment to make sure that the right thing was entered on the right fields and that the spelling mistakes were fixed. So anybody who does that job needs to be thinking about how to bring those types of tools or capabilities in-house so that they can be thinking more about what is that data used for? What can we do from a marketing opportunity? What can we do from a patient care opportunity? What can we do from an efficiency standpoint that will make customer experience better and make our business have more efficient operations?

Vinnie Schoenfelder:

I'm reminded of a conversation I had probably a year or so ago with a journalist and she was interviewing me. And on the side she was like, "AI kind of scares me and that's why I'm not using it. Is it going to replace my job?" And I said, "No, but a journalist who isn't afraid to use it will."

Cameron Snapp:

Exactly, same thing.

Vinnie Schoenfelder:

Right. So I think that's very similar. Although I do think people are going to be losing their jobs because task-oriented workers, if there's 4000 of them, all can't at the same time find out how to use AI. That's the choppy waters part I think we're going to have to work through.

Cameron Snapp:

One of the really interesting ones about the job loss, there's a lot of headlines around Amazon and Microsoft, but the small to medium businesses that have started to use AI to do their website build outs, to do their social media and their marketing campaigns, those ... If you're in a small business that you did that role for a bunch of other small businesses who might have bakeries or landscaping companies, that's where I'm actually thinking the job loss is more detrimental because that's going to be the loss of entire small businesses, rather than just a few employment numbers.

Vinnie Schoenfelder:

The other thing that concerns me with the heavy use of AI in our field in particular is that right now, we have great architects and engineers that provide a human in the loop oversight, governance, that type of thing. But if you are now coming out of school and you've used AI in high school, in college to write a lot of your code, to create your data structures, to do all those things, are we developing that same level of expertise that knows the nuts and bolts and knows how it really does work so that there will still be humans in the loop later that have enough experience to know what AI is doing well and not well?

Cameron Snapp:

Yeah. This is also something I think is a fascinating conversation over a coffee or a beverage. That's young folks that have had technology as their first and foremost comfort zone are ready for this. They're already expecting that the companies they work for are using AI because they used ChatGPT in high school or college and they have had technology as a part of their life. So they're expecting a company to have personalization when it comes to customer sentiment. They're expecting the company to have Microsoft Copilot available to them. They want to use AI to be productive and innovative. And those of us that have been in the workforce longer, I won't be aging myself here, but it's we know how to do things. We've seen what good looks like. And I think you need both.

So the critical thinking, the prompt generation, how to interact with these systems is absolutely what everybody needs to be focused on. Fact-checking AI's results. We talked a lot about data quality and that's where we want to look at what AI generates and makes sure it fits our standards, best practices, but also accuracy. You have to run some experiments to say if AI gets the right answer to a problem you know the answer to. So I think it doesn't matter what your experience level is, that's a human role that we have to have in the companies that are on this journey. So rather than just automating the tasks, it's let's make sure the people who did the tasks are checking the systems for accuracy and compliance.

Communicating value, information storytelling, being able to tell a compelling narrative is absolutely the thing. So I think people that are going to get left behind by this are the ones who didn't take that as a new opportunity for their own growth.

Vinnie Schoenfelder:

So from a business perspective, how does that shift the whole business intelligence side of things? Going from more dashboarding to more predictive to more proactive. Walk me through that.

Cameron Snapp:

Yeah. The days of Spreadmart historical reporting, "This is what our sales were last quarter," that's over. We really need to be thinking about the agentic systems are totally capable now of saying, "This is what happened last quarter, this is what is going to happen next quarter, and here's what you, Vinnie, should do with your business because of it." So that's, again, we're on that information maturity curve where we're up many levels than we were before.

Vinnie Schoenfelder:

This is what will happen if you do this.

Cameron Snapp:

Right.

Vinnie Schoenfelder:

So you can ask it, "What if I address it this way? What if I address it that way?" And have a what if realtime analysis with it.

Cameron Snapp:

The what if analysis is something companies were really chasing maybe in the 2010s. There was a lot of conversations I was having in the how do you build a mature data platform space. It was, "We can do analytics so that you can do predictive." To ask a question to say, "If I add this menu item to my restaurant menu, what will the cannibalization be?" So there's a lot of what if analysis that was possible. Now it's I don't even have to ask the question to the AI, it's just being able to look at my data and make a recommendation to me because it understands what those scenario simulations look like, and I think that's more, "Here's what you should do next," is really where we're headed now. So that's where analytics are going.

Vinnie Schoenfelder:

And that's where you can refine it because you know it doesn't know everything so you can say, "Well, that makes sense, but what if we're relocating the business?"

Cameron Snapp:

Right, and then you can vary it.

Vinnie Schoenfelder:

Right, right, and then it goes down.

Cameron Snapp:

And that's the back-and-forth nature. This is not just some system that answers one question or knows how to do one thing well. It's an interactive capability.

Vinnie Schoenfelder:

It's really quite interesting to think about pivot tables in Excel from 15 years ago, amazing, until now you're actually typing out full English sentences, having a conversation with your data.

Cameron Snapp:

Right, and that's part of my information storytelling communicating human aspect in prompt engineering. You have to be a robust communicator, you have to understand specificity, and you have to use the language that works with your business' documentation. If you can do an automated document scan of your entire knowledge base, you have to use the words and ask questions that it's going to be able to find. If you don't understand what your business does or what your data is about, you're going to have a real challenge interacting with any sort of LLM or agentic system that is put in place by your company.

So that knowledge of what you do and how your industry works, and making sure that your data reflects those types of parts of your solutions, that's what this whole podcast is about. The data solution at your organization, it needs to be really specific and well-documented for these new AI systems to take advantage and actually open the door, and not just be a huge spend that doesn't lead to that ROI that we're all chasing.

Vinnie Schoenfelder:

Gotcha. At the end, I'm going to ask you for steps people can do and take and move forward tomorrow, but one topic before that. It's a question I like to ask all data people when I have this AI conversation, and I actually moderated a panel a couple weeks ago and asked this question. And I'll give you their answer and see how you feel.

Cameron Snapp:

Okay.

Vinnie Schoenfelder:

The question is how urgent is this? There's a sense in some organizations that we got to be doing something because everybody else is, and perhaps that excitement negates finding the correct problem that you spoke about earlier. So the answer to that I got from the panelists were, "Don't panic, take a deep breath. This is still emerging. You still have time to do it right." As opposed to Lord of the Doom, "If you've missed it, you're too late." So where do you come down on that?

Cameron Snapp:

I'll be in the middle. It's not too late, but you have to get started. I think any day that you're not thinking about how AI's going to transform your business is a day lost on that progress. It's not too late, but the companies that you're competing with, if you wait, that's just time they're getting to outrun you. So I think that it's certainly a competitive advantage to getting on this train now. Companies will have a two to three-year advantage on you because that's how long this has already been out there. So the capabilities compound over time too, so if you're not spending time getting your data better, you're just acquiring more data and it could potentially hold you back in a few months or in a few years.

So I think getting started on that idea now, putting in place some of those consistencies on your governance patterns to say, "We know AI is coming. We might not be ready to do the full blown agentic automated decisioning right now, but we sure as heck can get our data in a better place." Or, "We can be thinking a lot more forward about centralizing our data into a place or getting some of our unstructured documents into a vector search database capability." So there are a lot of mini efforts organization can consider now, knowing that this is not going away and this train is moving quickly. So getting behind-

Vinnie Schoenfelder:

You introduced a new concept that I hadn't thought of, which is not just being on the sideline when something's progressing. But if your processes are not creating clean data, you're actually creating a larger-

Cameron Snapp:

A larger hurdle over time.

Vinnie Schoenfelder:

Right.

Cameron Snapp:

You're going to make it worse.

Vinnie Schoenfelder:

I hadn't thought about that angle.

Cameron Snapp:

So I'm not saying right now everyone needs to deploy an agent in their environment, no. But what you do have to do is think about those foundation fundamentals and what you can do to better prepare your house structure, if you will, for the hurricane that might come soon.

Vinnie Schoenfelder:

Right. So let's wrap up with you listened to this podcast, you want to do something meaningful in your organization. We'll make the assumption that there's been some interest and some steps to move forward. But what would you advise a data-minded person, a business-minded person to do starting tomorrow? What sites can they go to? What should they be learning? What should they be moving into their normal work patterns and habits? What have you changed over the last year or two in terms of your work patterns and habits that you would recommend?

Cameron Snapp:

Sure. So I think using AI in your day-to-day absolutely has to be the first thing, so that's something I've done. We have VS Code GitHub Copilot. I've been writing a lot of Terraform code and I've learned how to do better prompting. I can say from my experience the results are very mixed. So you will get frustrated and immediately you'll be like, "I could do this better than AI." But you have to learn how to work through those challenges and understand that your LLMs are extremely positive. They are always going to be like, "Great idea, Vinnie. You're right, I'll go with that." It's rarely going to say, "I think this prompt doesn't make any sense and I want to interpret what I think you meant." They're going to go down the rabbit hole you send them, so you need the practice. So individuals right now, start working with these types of automated tools.

The second one from a business leadership perspective is to take the future of this organizational strategy seriously and really set out to say, "These are the policies that we're going to have. We're going to have good transparency, we're going to have good monitoring, we're going to have good data stewardship." There are entire roles at companies, chief data officer among them, but the data engineers need to be working towards a single cohesive data management strategy. More so than ever. Every time I have this conversation with our clients, I get more and more convinced that this is the right mindset. So we talked about reusable patterns and we talked about data governance, it's just more impactful now to have those. The cause, the just case for having them is just so obvious to me from that.

Trust building needs to also be a huge part of this. Where the more and more you automate and the more you're taking humans away, people are skeptical of machines. So I think having a trust first mindset for anything you do as you build out AI is also critical.

The last one that I think a lot about is operational analytics. So we need to monitor the productivity of these tools and these processes. It was really ... You asked the BI question, and when you look at a lot of what's our business doing, where should we go, what should we do, what if analysis, more importantly too it's how are the systems behaving? Are they be retrained? Are they operating under 2023 mindsets? We have a client that worked on anti-money laundering trends and there was a really robust ML process that was able to identify, was this transaction money laundering or not, yes or no? Great. What about new ways that people could be doing fraudulent activities? The agentic systems can go do that and think about what's current in 2025 and 2026. So it's really important as you're monitoring and governing these processes to keep up with the times on what they're doing and keep an eye on them for where they might need to be adjusted.

Vinnie Schoenfelder:

That's a whole separate podcast we can jump into because, yes, AI is helping with all the anomaly detection, yet AI is also making the deep fakes and scams much more effective.

Cameron Snapp:

Right.

Vinnie Schoenfelder:

So it's a race.

Cameron Snapp:

Can it decide its own self whether that thing I thought was anomalous was actually credible and can I correct myself the next time?

Vinnie Schoenfelder:

You can have AI policing AI essentially.

Cameron Snapp:

Right. Sorry, one more topic on what data-centric people should do is to really understand what these new architectures look like. So that's the thing for me is I know how to build data pipelines and data processes and have conversations with clients. The what an agentic stack looks like is new and challenging, and there's a lot of people that are right out of college that already understand that stuff, too. So knowing how LangChain, Kubernetes, Docker all play into this. Knowing how to get your own LLM stood up inside your organization so that you're not putting your public information out onto Claude or ChatGPT, you have to be able to do that. So that's another thing people can start doing now is saying, "We don't need to have stuff in production, but we have to start learning these tools and these necessary components that are a part of these architectures so that we're ready when the use cases and the company policies come around."

Vinnie Schoenfelder:

Great. Well, Cameron, thank you for joining me today, I really appreciate it. Good conversation. A thank you out to our listeners as well, we really appreciate you tuning in.

Cameron Snapp:

Thanks so much.

Disclaimer:

The entire contents and design of this podcast are the property of CapTech or used by CapTech with permission, and are protected under US and international copyright and trademark laws. Users of this podcast may save and use information contained in it only for personal or other non-commercial educational purposes. No other use of this podcast may be made without CapTech's prior written permission. CapTech makes no warranty, guarantee, or representation as to the accuracy or sufficiency of the information featured in this podcast. The information, opinions, and recommendations presented in it are for general information only and any reliance on the information provided in it is done at your own risk. CapTech makes no warranty that this podcast or the server that makes it available is free of viruses, worms, or other elements or codes that manifest contaminating or destructive properties. CapTech expressly disclaims any and all liability or responsibility for any direct, indirect, incidental, or any other damages arising out of any use of or reference to, reliance on or inability to use this podcast or the information presented in it.