Skip to Content
Find dismissed updates here
Edit My Preferences
23:29 Video

Moving AI from Pilot to Production: A Guide for Financial Institutions

Everpure's Chad Primakow joins The Money Pop at Money 20/20 to discuss how financial institutions can move AI from pilot to production by mastering data context.
Click to View Transcript
00:06
Hello, and thank you for joining us, and welcome to the MoneyPop podcast. We're live at Money 2020 USA in Las Vegas at The Venetian. I'm your host, Om Naiesadu, and on this episode we're gonna be talking about what banks and fintechs need to get right as AI moves from pilot projects into real world production. I'm joined by the fantastic Chad Primico,
00:25
AI and data leader at Everpure. For those who don't know, Everpure helps financial institutions modernize their data infrastructure with a unified platform designed for mission-critical banking workloads. From ransomware recovery and cyber resilience to AI-ready data and registry compliance, Everpure helps simplify operations,
00:44
reduce risk, and accelerate innovation while maintaining the performance and availability financial institutions demand. Chad, it's great to be in the booth with you. You spent 20 years at American Express before joining Everpure. What are you seeing now at Everpure that financial institutions
01:01
still underestimate about moving AI from experimentation into production? They underestimate simply knowing their data. They, they underestimate how expansive it is, the large volumes, the, the internal and external pressures, that they have that are affecting them. The data rot, which is redundant or duplicate data, obsolete data, data that is,
01:26
is no longer meaningful, and it is not, fully there, as well as trivial data. Trivial data, that really it doesn't have any reference, with today's data, that's there. That along with the inability to identify dark data, which dark data is collected without, knowing the value that it brings to the table itself. And that leads to having the inability to, to see the full data state.
01:55
So this is, it's something similar to when a person decides to travel. So if you can tell me, how many individuals do you feel travel, without knowing their destination? How many travel without knowing where they're gonna be staying? Without knowing where their weather, what their weather is gonna be like,
02:15
the temperature, the, what they're gonna be doing on the trip, and is it for business or is it for holiday or vacation? So packing my entire suitcase would not be very efficient, in that particular case. It would also be time-consuming and likely very expensive. Yeah. I think if you exclude backpackers,
02:34
that's basically nobody, right? Like- That's correct you, you said a really interesting term that I'm not super familiar with, dark data. Can you give me an example of that? So it's a really interesting, like, name to give to data.
02:44
So dark data, when you throw data into the ecosystem and that data itself is You don't know that it's there. You don't understand the meaning behind it. It just exists. Yeah. And until you're able to define the context, it's perfectly meaningless,
03:02
and now all of a sudden within your AI ecosystem, you're gonna get results that you didn't expect because it's data, that is leading to a conclusion that you don't understand. Oh. Have you got any examples that we can pull? So, so generally dark data itself is, for financial institutions,
03:22
you may have ledgers that are, that are brought in, and no one knows why they brought the ledgers in there. The ledgers aren't necessarily telling anyone anything about their data. They were just loaded because they just happened to be, part of someone's, data portfolio, and they loaded everything in, and they didn't even realize that they brought
03:40
it in. And because they didn't realize what they brought it in, they didn't have a business purpose for it, and that leads to the data being dark, and not understanding exactly what value that it could possibly bring to the table. Sure. Yeah, that ma- that makes sense.
03:54
And I think you mentioned, you touched on, like, I guess deprecated data. And once, this is very much a, a anecdote, but I s- I once saw, a bank in the UK that had shillings and pence in their data somewhere, and I was like, "That was pre-computers." Like "How did this ever get there?" And it was just, it was just there in the historic data record.
04:15
So when everything got digitized, it went up with it, too. Um- Exactly so, so there's, there's a lot I think. There's, there's a rich area. We've seen significant AI experimentation across financial services, and AI and data go together like ducks to water, right? From a data perspective,
04:33
what separates the organizations that are actually able to scale AI from those that remain stuck in pilot mode? Well, it, it really starts with, defining a few measurable outcomes. So not trying to bo- boil the ocean. This is then enabled with well-defined, high-quality, and data that,
04:53
that builds trust in your data ecosystem. So organizations that haven't identified and classified their data, with business context, it, they generally tend to turn, churn, without achieving any types of, meaningful results. So if, if you revisit the, the metaphor that I had earlier, regarding travel, it's,
05:15
it's really you need to understand, in order to scale the, the purpose and the activities for the trip that you're taking. So if you don't understand that, you're not gonna pack properly. You also need to know everything that you have in the closet, and that way you can, you can select the clothing that's gonna meet that purpose, that you have in place.
05:35
And if you do that, immediately your packing will become much easier. Henceforth, the data, that you're bringing into an ecosystem will also become much easier to understand, and you'll, you'll know the, the context, and the measurable results that come from it. Oh. So know what you have,
05:55
know where you're going, and know what you'll need when you get there. That is correct. Right? Yeah. Yes. So what are some of the warning signs that- the data foundation isn't ready to scale? Like what, at what point do you know when you're looking at a data set that, that we need to do more work here?
06:09
One of the ways that you know is you're not able to cite the data. So you don't know what, where the data was used in that particular result, so it doesn't make sense to you. You, you kind of get lost by looking at the result, and you go, "Where did that come from? I have no idea what it's, what it means and what it's about." and that usually means that,
06:29
organizations are throwing data, into AI, and they're trying to see what sticks. So back to the metaphor, it's like opening your luggage and, and you're doing that during your trip, and you're, you're not really sure if the clothing, itself that you packed is appropriate or not. Right. Agreed.
06:48
So once an organization has identified an AI use case, it wants to put it in production, right? Like, "There's an opportunity here. We wanna move forwards." That's right. That's the essence of so many businesses.
06:59
What does it actually need to understand about the underlying data before moving forward? You talked about how you need to make sure the clothes are the right fit, right, and they're, they're appropriate. But knowing the who, what, when, where, how, and why, that's very journalistic, of the data, what does that look like in practice? How do you, how do you make that real?
07:17
So you make that real by who are the data owners, and also who are the consumers of the data. What is that, that data that the context is understood, and it answers the outcome? What is the, when is the timely data, that is available when it's needed? And it's not too late.
07:38
You, you don't want it too late. Then again, if you have it too early and you don't understand the what, the, the when isn't very valuable. The where is where are you getting that data from? Is What's the location of that data?
07:52
Do you trust that data? Is the safety known, for that data itself? And then you, you jump into the how, and how do you get that data into, into AI? What are the people, processes, and tools, that are utilized in order to make sure that you have high-quality data, you have data that is highly consumable as well,
08:15
and that it's, it fits in. And finally, and it's most important, it stitches everything together, it's, it's really the why, and that's the definition of the need for the data. If you don't understand why, it doesn't matter what, where, when, or how, the data gets there. Cool.
08:32
Yeah, so having the right purpose is the first step. That's correct, yeah. So as AI is moving into production, how does governance, security, and DR management need to evolve alongside the data itself? Like, these are tough topics at the best of times. That's agreed.
08:50
So all three of them that you just mentioned, they need to move at the speed of business. Mm-hmm. So if they're not moving at the speed of business, then the value is gonna be lost, or you're not gonna go back and revisit, what was stated, by the time the, the data has already aged, as part of it. So building trust and, and safety into the processes will, will allow you to,
09:16
to take the controls, and you're, you can apply them in order to be able to, to balance protection and the usage, while meeting your specific risk tolerances. Specific risk tolerances is a very interesting phrase, right? Do you think that ties into the why with regards to knowing your data in the beginning? It absolutely does. So if you don't understand the risk that
09:39
is being presented, and you don't understand the outcomes that you potentially can get from it, you're not gonna balance that appropriately. Sure. Yeah, I think it's, that's a huge amount of work to do right at the very beginning. Do you think there's strong appetite to do that, though, in the end?
09:56
So the appetite, generally is to throw everything in- Yeah and to see what, what the results are. Mm-hmm. Well, if you're introducing risk, then it's really difficult to get rid of it. Sure. So, so going too deep into the data, into the breadth of data- Mm-hmm can create unintended outcomes for your pilots
10:16
and your programs. That's absolutely correct. Yeah. So once organizations start properly discovering and classifying the data they want to use for AI, what kind of problems or risks tend to surface? We know, we know that if you throw everything and the kitchen sink at it, you'll get loads of problems. But even the best laid plans rarely survive
10:36
deployment p- like unscathed. What is it that people are gonna see? Geron, what have you seen people see? It, it generally starts with the data sensitive- sensitivity, especially with regards to PII, so personally identifiable information, PHI,
10:51
protected health information. It, it then expands into trade secrets, and intellectual property. You need to know that you have those in place because problems will, will crop up very, very quickly. And especially if you introduce stale and dark data, you don't know whether it's, it's data that,
11:11
that has any risk or not because you don't even know what the business value is and the context of the data that you're looking at. Yeah. What's, what's the solution to that, do you think? Like, there's, there's all these risks associated with it, but how do you filter?
11:23
'Cause having someone manually go through is difficult. It's gonna take ages. That's when it's important to have a solution that is in the mix. Yeah. And the solution itself needs to be at the speed of the business. You can't wait around in order to, to clean everything up and make sure everything
11:39
is perfect. The old saying, "Perfect is the enemy of the good enough," really comes into play here. So really that's, that comes about in making sure that you're able to identify what data is risky to you. Being able to, to understand that and make sure that you eliminate that data in your data
12:00
pipeline as you're bringing it into your AI ecosystem. Right. Yeah, I think having, a gateway of some sort to make sure that you prevent any of that data entering your AI keeps everyone or your consumer safe, keeps all your customers safe, it keeps the company safe, and also keeps the AI products safe, right?
12:20
It means that they're able to Or it is able to, like, move in-flight without tarnishing itself or training itself on the wrong issues and looking out for unrelated problems, right? Defining what risk looks like, defining that, that personal information, defining the, the PII, the PHI, the, the secrets, that you don't want
12:45
to be released from your company, that you hold very, very, carefully because it's your intellectual property. Yeah. And having that out in the open, would potentially cause more harm than help, for any, any outcomes that you would get, from that specific data. Making sure that you identify how those are defined ahead of time,
13:06
will allow you to avoid having that, that data enter into your ecosystem. Nice. Can you give us any real-world follow-up examples of what this c- process can uncover, especially in a financial services environment? I think PII obviously is a major issue, and a lot of companies do have that quite locked
13:25
down, but without, without giving away names, is there anything that you can share with us that are, like, good examples of the, the risks that get encountered? So generally, in financial services you have large amounts of account numbers, national IDs, names, addresses, phone numbers, contact information. They exist in high volumes of, of unstructured, data and files, that are,
13:53
that are present within the organization. You, you couple that in with secret information such as, pre-release earnings, fraud models, credit decision, processes. You have, pre-release marketing campaigns. That exposure itself could significantly damage, the,
14:15
the company legally, financially, and even affect the, the reputation and damage the brand. Which makes total sense. We, we often hear that h- more, the more data the better, right? I- feels like you have a very different opinion.
14:31
I think that's, that's p- safe to say. We've seen, like, a lot of the LLMs, for example, especially in, in the West, have consumed all the internet, all of Reddit, for example, and hu- just huge, huge sums of data without really accounting for anything, to be excluded. But organizations have to be a bit narrower, like you said.
14:55
Mm-hmm. And they're often dealing with duplicated, stale, irrelevant, and poorly understood data. At what point does more data, having more data become a liability, r- I mean, I think we've kind of covered that it's not necessarily an advantage, but when is it actively becoming a liability?
15:15
So it's actively becoming a liability when the data context is unknown, it's redundant, it's obsolete, it's trivial. That leads to false outcomes. It skews the results, and, it also makes you question the, the accuracy and the validity of the data itself.
15:32
When, when you intersect that with laws, regulations, directives, and requirements, that can lead to compliance that's broken. It And that will end up, as I mentioned before, in, in financial harm, legal liability, and reputational brand damage. A metaphor that I like to use in order to, to judge whether too much data is enough,
15:56
comes down to adding sweetener to your morning beverage. So once you're, once you add that sweetener to your morning beverage, it's very difficult to remove it. So now it's, it's actually, it's almost impossible to remove in, in some cases if you no longer want that sweetener that's in there.
16:14
And in addition to that, diluting doesn't eliminate the issue. So, you know, bringing in sometimes more data is not always better. Just because a little bit of data is good doesn't mean a lot of data is, is not. Make sure that you match it with your, your ecosystem. Ma- make sure you match it with your outcomes, and then you can, you can correctly size it.
16:40
You can always add more sweetener to your beverage, as you need it, instead of dumping all the sweetener in and then realizing that you have too much. Yeah. To extend your metaphor a little bit, I think if it was a cup of coffee in that situation, you'd have to dump it out into the sink and start again, right?
16:56
And- That's right that's fine when it's a coffee you made at home, but when it's your entire product, that's potentially millions that you have to start again from scratch from, right? That is correct. Which is a risk I don't think many people wanna take on, in terms of starting again.
17:14
How should organizations decide which data actually belongs in a particular AI use case? Like, we talked about the f- the who, what, where, when, why, and how. Mm-hmm. But what does that look like really? How, how I've got my data set, how am I examining it? How am I filtering it to create a specific AI use case?
17:35
It, it comes down to simply answering if the context and the risk appetite have been defined in order to be able to match those overall business outcomes. And if no reasons are, are able to be stated, for introducing that data, and there's It's not known, if it adds risk, then it likely shouldn't be utilized for that AI use case, at least yet.
17:58
Yeah. So once you understand what the use is behind it, then you introduce it. Yeah. You, you do it quickly, and you do it carefully. Yeah. So it's the quest- the, the response to people saying- Let's put this data in.
18:13
It shouldn't be yes, but why, right? It should be- That's right it should be let's start from a position of slight, slight skepticism, let's interrogate this, let's figure out the purpose it's gonna serve, and then say yes. You can still say yes. Okay. So you start by saying yes, and you say yes when, when areas
18:34
are understood. Okay. Right? Yes- Yeah if you've identified the context, and you've also calculated the risk of introducing it. Yeah. I s- I like that it's like, introducing an F function for your data. That's right. Exactly.
18:48
So what changes once AI is no longer just producing outputs for people, but is being used by agents to carry out tasks or, and actions? 'Cause agents, they're gonna be changing everything I hear, so they're gonna be the most important users of this data potentially. Absolutely. So it, it becomes more important to,
19:10
to trust the quality of the data and ensure its safety, when AI agents are introduced. Remember that an agent itself won't, won't question its objective by default. That's not something that agents generally do unless it's, it's told that that's what it should do. So ensuring the pro- processes themselves are, are well-defined and they're mapped
19:34
to the overall data context will ensure that it's carrying out the necessary actions without introducing unnecessary risk. Amazing. So let's say I'm a business, and I've done all, all of this. I've listened to Chad, I know my data, I understand why for everything, I understand the when.
19:53
My product is in production, it's in-flight. What do I need to be looking out for? What do I need to be monitoring? So you need to be ensuring that it stays within the guardrails. If something goes outside the guardrails, and you have that set up properly, you'll, you'll be able to take action,
20:09
before it goes out of control. So you need to make sure that everything is well-defined, the output itself remains within quality and risk parameters. If you have those defined, those guardrails themselves, you'll be able to monitor the, the data, and instead you'll be able to look at the business outcomes,
20:27
that it's generating. Those results will become extremely valuable, and you won't worry about something that is doing it at, on its own. That also comes into play with, with having to start with human in the loop until you build that trust and you understand the safety of the data, and then the human is gradually pulled out of the loop, and then you're able to,
20:49
to just simply watch it, and, and see exactly what it does over time. I l- I love the idea of pulling the human out of the loop, because at what point do we develop trust, right? And I think, like you said, it's repeated positive, unchanged reviewed results, right? That's right. I think that's, that's the way forward.
21:09
For a data or technology leader who wants to look at their AI program today and really understand what they should be focusing on, what, what do they need to be focusing on first if they want to move from experimentation and pilots to something that can scale and really build the future of their business? So you start by identifying the highly relevant data.
21:30
So make sure that it's defined, it's, it's understood, it's risk-balanced. Know the measurable outcomes. And make sure that, that you've defined all of those areas, because there's nothing wrong with growing the data as it's validated instead of just throwing it in and, and trying to see what happens.
21:49
You can avoid very, very costly mistakes. You can avoid, the rush to, to be able to put areas that are out there that you can't take back. So be careful, and also be fast. So we have a little bit of time, but fairly quickly, what's the one thing We have a quite lovely audience here.
22:10
What's the one thing that leaders here should be doing now that they may not be paying enough attention to? What should everyone listening think about on the way home that they should be like, "This is what I need to do when I get back"? They should be looking at their data estate and understanding it fully.
22:28
So back to the metaphor of, of the clothing, for a trip. Understand what's in your closet. Don't lose track of that. Once you've identified everything that's in your closet, don't, don't just add more clothes in, and, and just throw them on top and, and forget about everything that's, that's been there.
22:47
Make sure that you're able to, to identify what you're consuming, when you're consuming it, and avoid all of the pitfalls of, of having too much data that's coming into AI that's really not gonna bring you any value. Amazing. Chad, thank you for your time for joining us on The Money Pot. It's been brilliant.
23:05
You've been fantastic. Thank you for being here. That wraps up our podcast. Thank you for listening. Make sure you subscribe to The Money Pot wherever you get your podcasts, or find us at money2020.com. We'll, where we'll continue to bring you the latest in the trends shaping finance
23:19
across the world. We'll see you next time. Thank you.
  • Video
08/2026
Everpure FlashArray//X: Mission-critical Performance
Pack more IOPS, ultra-consistent latency, and greater scale into a smaller footprint for your mission-critical workloads with Everpure™ FlashArray//X™.
Data Sheet
4 pages
Continue Watching

* indicates a required field.

We hope you found this preview valuable. To continue watching this video please provide your information below.

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

Your Browser Is No Longer Supported!

Older browsers often represent security risks. In order to deliver the best possible experience when using our site, please update to any of these latest browsers.

Personalize for Me
Steps Complete!
1
2
3
Continue where you left off
Personalize your Everpure experience
Select a challenge, or skip and build your own use case.
Future-proof virtualisation strategies

Storage options for all your needs

Enable AI projects at any scale

High-performance storage for data pipelines, training, and inferencing

Protect against data loss

Cyber resilience solutions that defend your data

Reduce cost of cloud operations

Cost-efficient storage for Azure, AWS, and private clouds

Accelerate applications and database performance

Low-latency storage for application performance

Reduce data centre power and space usage

Resource-efficient storage to improve data centre utilization

Confirm your outcome priorities
Your scenario prioritizes the selected outcomes. You can modify or choose next to confirm.
Primary
Reduce My Storage Costs
Lower hardware and operational spend.
Primary
Strengthen Cyber Resilience
Detect, protect against, and recover from ransomware.
Primary
Simplify Governance and Compliance
Easy-to-use policy rules, settings, and templates.
Primary
Deliver Workflow Automation
Eliminate error-prone manual tasks.
Primary
Use Less Power and Space
Smaller footprint, lower power consumption.
Primary
Boost Performance and Scale
Predictability and low latency at any size.
What’s your role and industry?
We've inferred your role based on your scenario. Modify or confirm and select your industry.
Select your industry
Financial services
Government
Healthcare
Education
Telecommunications
Automotive
Hyperscaler
Electronic design automation
Retail
Service provider
Transportation
Which team are you on?
Technical leadership team
Defines the strategy and the decision making process
Infrastructure and Ops team
Manages IT infrastructure operations and the technical evaluations
Business leadership team
Responsible for achieving business outcomes
Security team
Owns the policies for security, incident management, and recovery
Application team
Owns the business applications and application SLAs
Describe your ideal environment
Tell us about your infrastructure and workload needs. We chose a few based on your scenario.
Select your preferred deployment
Hosted
Dedicated off-prem
On-prem
Your data centre + edge
Public cloud
Public cloud only
Hybrid
Mix of on-prem and cloud
Select the workloads you need
Databases
Oracle, SQL Server, SAP HANA, open-source

Key benefits:

  • Instant, space-efficient snapshots

  • Near-zero-RPO protection and rapid restore

  • Consistent, low-latency performance

 

AI/ML and analytics
Training, inference, data lakes, HPC

Key benefits:

  • Predictable throughput for faster training and ingest

  • One data layer for pipelines from ingest to serve

  • Optimised GPU utilization and scale
Data protection and recovery
Backups, disaster recovery, and ransomware-safe restore

Key benefits:

  • Immutable snapshots and isolated recovery points

  • Clean, rapid restore with SafeMode™

  • Detection and policy-driven response

 

Containers and Kubernetes
Kubernetes, containers, microservices

Key benefits:

  • Reliable, persistent volumes for stateful apps

  • Fast, space-efficient clones for CI/CD

  • Multi-cloud portability and consistent ops
Cloud
AWS, Azure

Key benefits:

  • Consistent data services across clouds

  • Simple mobility for apps and datasets

  • Flexible, pay-as-you-use economics

 

Virtualisation
VMs, vSphere, VCF, vSAN replacement

Key benefits:

  • Higher VM density with predictable latency

  • Non-disruptive, always-on upgrades

  • Fast ransomware recovery with SafeMode™

 

Data storage
Block, file, and object

Key benefits:

  • Consolidate workloads on one platform

  • Unified services, policy, and governance

  • Eliminate silos and redundant copies

 

What other vendors are you considering or using?
Thinking...
Your personalized, guided path
Get started with resources based on your selections.
My Updates
No updates at this time.