Power of Patterns: A Practical Approach to Introducing Enterprise AI

[00:00:00] Chris Falloon: Less than 10%, seven or so percent of companies have been able to get some sort of AI or gen AI based workload running in a production environment at scale. It's a very small number. We are at the peak of of peaks in terms of the hype cycle. I've not seen this level of frenzy since we were selling fiber, really. When fiber connectivity became a big deal.

I think making decisions at the top of a hype cycle is always difficult in any industry with any problem. But in this particular case, we've got a frenzy that's coming from the board level of most organizations. Well, it's not useful to build technology for a use case, or even a handful.

What's important is to understand not just the ROI and how to have a conversation around the viability or the plausibility of a particular technology in the business and what it provides to the business in terms of return. But how do I start to categorize use cases into meaningful patterns so that I can understand the implication it has to my business at all layers?

I think success at this early stage is keeping yourself from falling into some predictable mistakes.  

[00:01:09] Mike Reeves: This is Solving for Change, the podcast where you'll hear stories from business leaders and technology industry experts about how they executed bold business transformation in response to shifts in the market or advances in technology.

In every episode, we'll explore real world strategies and technologies that fuel successful evolution. I'm your host this month, Mike Reeves.

I'd like to welcome Chris Falloon from Dell Technologies to the podcast, Solving for Change, today. And I'm really excited to have you on. I appreciate you taking the time, especially now in the global role that you have to come in and be a part of MOBIA Connect here in Whistler. And for the partnership, I'd like to thank you as well and now selfishly, I'd like to thank you for being on the podcast today.

So thanks for taking the time to join us. And if you don't mind Chris, it'd be great if you could just take a couple minutes because we're going to spend a lot of time talking more about you, anecdotes in what you see as you've been traveling around the world in your role as we get into the discussion.

So maybe if you could just give people an intro on who you are, kind of what you do, a little bit about where you've been going, and so on.

[00:02:14] Chris Falloon: Happy to, happy to. And listen, thank you for for having us here and selfishly on my side being able to work within an hour in my home and in in Vancouver is an unusual pleasure.

So I really enjoyed the the drive up and real happy to see some faces here from when I worked in technology for the first half of my career in Vancouver. So thank you for having me.  

Look, I went down into the U S markets and into the global side of EMC and, subsequently, Dell Technologies in 2014, so about 10 years ago. And one of the groups inside of Dell Technologies that's poorly understood, I guess maybe even inside of our company, is this group called the transformation office, of which I am part. And the transformation office was really pulled together about eight or nine years ago to help solve some of the platform complexities that our customers, the frustrations that our customers were seeing, trying to bolt together all the things they knew needed to bolt together to deliver technology services.

And so, as an infrastructure vendor our predominant role in this is world class technology at the bottom of the diagram. But our role within the transformation office is to help customers align their technology strategy with what the business expects from them in terms of services and really navigate the selection of best of breed technologies. Navigate the complexities of engineering and platform engineering so that they can figure out how to successfully deploy and achieve the benefits that some of these software packages say they have on the box. Not, not always that easy to do.

My personal role, I run a group within the transformation office called the practices organization. I've got four practices, globally. We've got a data and AI practice, a telecom transformation practice, a cloud and multi-cloud practice, and a security practice. And all four of those really with the mandate I just described is just to help customers navigate those platform outcomes.

So, I've got people in 11 countries, I get a good chance to be with Dell's largest partners, and our largest customers globally. So, it affords me the privilege of being able to take in a fair bit of customer-specific information and I'm really glad to be able to share some of that with you.

[00:04:27] Mike Reeves: Yeah, and that's what I want to spend some time talking about today. Because you're here at the conference, you had a great session this morning and there's... We start talking about digital transformation, now you layer in AI and everybody's under a lot of pressure at the executive level in the boardroom and then throughout the organization of any company now. And it's around what are we, or how are we going to approach AI and it's really interesting to talk about that because if... You know, we were talking a bit about social media this morning. If you go and you look at some of the anecdotes online, you think, "Wow, a lot of people got this figured out a lot of companies, like they're really, they've got a lot of richness and they've gone quite a ways down the road and had great success deploying AI or finding use cases and come with a great business outcome around AI."  

And I think the realities are, that is a very small segment of what is happening within a lot of organizations globally. And that's kind of why I want, where I wanted to start today is maybe it's to massage some of those concerns of a lot of people in companies and senior executive roles of thinking like, I'm not being successful, I'm not getting anywhere with AI yet. And there's a lot of fear around that. And so people are just kind of rushing to the table to try and find ways to drive it into an organization. I think what I'd like to start with, with you, just based on your experience, like you travel all around the world and you talk to a lot of customers, a lot of different profiles of industries, et cetera.

And, you know, listening to you this morning, I thought it'd be very helpful for a lot of folks that listen to our podcast just to get your perspective on what you're seeing.  

There's lots of ways we can go with the discussion, but really there's kind of maybe three or four major anecdotes or themes that you're seeing as you've been traveling or traversing the world, talking to different customers or you can kind of share some insights there.

[00:06:22] Chris Falloon: Sure, yeah. Look, I think first of all, nobody should feel bad about being, feeling like it's moving slowly. I think one of the data points I shared this morning was that less than 10%, 7 or so percent of companies have been able to get some sort of AI or gen AI based workload running in a production environment at scale.

It's a very small number. And even among those, those use cases are not terribly exciting. They're usually a rather mundane sort of data retrieval type workload. And so despite all of the, you talked about social media, I mean, we are at the peak of peaks in terms of the hype cycle. I've not seen this level of frenzy since we were selling fiber, really. You know, when fiber connectivity became a big deal.  

I think making decisions at the top of a hype cycle is always difficult in any industry with any problem, but in this particular case, we've got a frenzy that's coming from the board level of most organizations. There's mandates to try and solve for not just how we use these technologies, but how we implement them into the business. One of the things I touched on this morning is that, not only is it difficult to make the technical choice, it's also difficult to make the use case choices, and it's difficult to write business cases for some of these use cases.

So, we've got board level members struggling, we've got line of business leadership struggling with these answers, we've got data delivery and data management folks and application developer teams, and the underlying infrastructure teams. Everyone's struggling at exactly the same time, which I think in this particular case is, it's unique in the tech industry to have everybody struggling with the same problem at the same time.

Now that's positive in some ways, which at least we all have to talk and, and because we're all facing the problem at the same time, talking makes sense. But now we need a framework or a common language because these teams tend not to talk to one another. So I spend a lot of time, we as an organization spend a lot of time trying to create some patterns or some way to talk between the business, the tech deployment platform folks, and the underlying infrastructure folks.

[00:08:26] Mike Reeves: And you've got, maybe to unpack that a little further. So, this morning you were talking, you were really kind of had tried to categorize things or put some structure around a lot of this looseness that you're describing that we're talking about. That is, you've got an underlying infrastructure that needs to be there, you've got a whole bunch of things around governance and risk, and then you've got data. Kind of blending across all those things. And then you take the other complexities around hybrid IT, hybrid cloud, multi cloud, et cetera. And it's interesting, you talked through this morning, you've got some different models and approaches in terms of how you traditionally have a discussion with a customer or somebody that's trying to figure some of this stuff out. Maybe if you don't mind take a minute and walk us through that. It was very helpful.  

[00:09:11] Chris Falloon: Look, I think pattern recognition is something that technology delivery organizations have. It's been important for them to be good at that since there was a technology delivery organization. Understanding how to bucketize the request from the business into something that's usable to a technology engineer or a platform engineer. And this is really no different. I think what's missing here is the maturity, the language, and the structure to have these conversations with the business leaders.

One of the things we walked through this morning is a patterns discussion where we start up at the use case layer. And I think Dell's internal use cases now, the requests of IT are in the hundreds: a few hundred use case requests coming to the technology teams. Some with good business cases, some without.

I was in Australia last week with another client. They've got 1600 potential use cases identified by the business. So what that leaves you thinking is, "Well, it's not useful to build technology for a use case or even a handful. What's important is to understand not just the ROI and how to have a conversation around the viability or the plausibility of a particular technology in the business and what it provides to the business in terms of return, but how do I start to categorize use cases into meaningful patterns so that I can understand the implication it has to, to my business at all layers?"  

What are the implications to the data teams? What are the implications to the infrastructure, developer, and application teams? If we start to bucketize them, then we can we can build technology for large chunks of use cases, which means that if I have a strong ROI for two or three, and I've got 30 that happen to be supported by the underlying architecture, now I can build a platform that delivers 30 use cases. Even though the TCO or the ROI is supported by one or two key.

So that type of patterning, is something that we spend a lot of time talking about. I think we're seeing some very obvious technology patterns. Things like RAG and graph-based databases, the retrieval augmented, even small language models is becoming a more common technology pattern.

There's some of these things that are a little mundane but there are many, many, many use cases that fall into these really simple technical patterns.  

[00:11:30] Mike Reeves: The other thing that we talked about a lot this morning and you've touched on it a bit, but we get down to the people layer. I'd be curious just to get your insights in terms of what you're seeing again as you've traveled around and talked to different customers around how they're dealing with the people side.  

[00:11:48] Chris Falloon: Well, I mean, it should not surprise you maybe that the most difficult part of IT transformation for the last decade has been people and process. The technology almost is never the problem. Not to say it's not a problem, just it's the easiest one to solve.

If we have weaknesses in our people and process in IT or in our tech delivery environments, it becomes exacerbated by the conversations around AI and generative AI workloads because they're just that much more demanding and making mistakes is that much more expensive. And so poorly deployed assets or poorly managed people, resources of any kind really have drastic effects on overall costs. And so, it's critical to get your operational house in order, to deploy these. And it doesn't matter whether it's on-prem or off, those mistakes can be made in any premise.

Lots of idle capacity sitting out there, costing people money, not generating real value. And that's a real concern.  

[00:12:48] Mike Reeves: Just going to move on a little bit. That's great. I'm trying to knit this together in a fashion that gets us from problem statement to what you're seeing to what you've seen in terms of success. And we've talked a little bit about that now, and you've talked about it a bit already. Now we start considering use cases because everybody always goes to: give me the big bang. And you talked a bit about that this morning. As you mentioned, every company has some sort of backlog of initiatives that they feel would be good candidates. And, I know everybody's trying to figure out, "What is the best approach? How do we start? What is that launch point in terms of a use case?"  

So again, I'd just like to lean into you maybe a little bit more. And I know you said there is no answer, there's no one answer for this, but everyone's kind of looking for an approach or approaches to get to a point where they can get to a use case.

[00:13:50] Chris Falloon: Yeah. Look, it's easy to say things like, "There's no killer app." And I think there really isn't a single use case that's obvious. However, when you start to look at industry specific use cases, there are definitely patterns that are emerging where there's some no-brainer initiatives that take such big cuts at the cost of operations of an organization out. Or create such significant competitive advantages. The boards understand those. By now, the boards will understand those so implicitly that there are plans in place to go ahead and launch those already.  

It's really the first wave of ones we're unsure of that customers are really struggling with. And I think the technical plausibility and the return are both difficult to articulate, right? People are immature in the technology. So, I think they think about getting it stood up in a non-production environment is much different than running something at scale. And the maturity around costs, we talked about cost this morning, but the maturity around cost and cost to serve and TCO is very immature in these AI use cases. And so it becomes difficult for a marketing lead or a finance lead or someone to articulate an AI use case, a business case rather, internally that's coherent enough to say yes to because there just isn't the maturity to make those assumptions.  

The last thing I would say, I did say this morning, which is: as most customers are confronted by hundreds of use cases, the ones that are the most technically intriguing or intellectually stimulating may not be the ones that actually provide the obvious payback. Just because the ones that are technically curious to us, tend to be the ones that are the most difficult.  

What we've seen in terms of production, AI use cases that are being rolled out at scale, they're the most mundane and uninteresting things in most organizations. Risk and compliance work, sort of the back office type work that you might assume someone has already maybe outsourced to a body shop or an offshore partner. Many of those now can be looked at and reassessed as something where a combination of humans and machines can probably take a ton more cost out of those environments.

[00:16:02] Mike Reeves: And if you look at the whole discussion around, and continuing on the theme of use case, and I've seen this anecdotally, not just within our company, because we're trying to do a lot with with AI, less so gen AI. If you start to look at some of the customers that we have, some services and consulting that we do to help them move down that path... To your point, operational items seem to be the path of least resistance where people are able to build a business case and create some sort of financial construct that says, "This is how this will pay for itself in reduction of time, cost, people effort, et cetera."

I find that really interesting that there's an evolution, I hope, that's going to start to move from here is, when an organization gets a couple of use cases under their belt, they start to pick up a little bit of moxie around what they're doing and how they're going to start to take it further inside the organization and develop their own framework or approach to how they want to build out on their strategy around AI to start to execute at some sort of scale or velocity.

I don't know if you have any further comments around that, but just wanted to echo and share some of the themes that you just hit on there.  

[00:17:18] Chris Falloon: I think just, we talked a bit about patterns and putting these use cases into sort of a structured format that makes it easy to talk about them. I think one of the things that I've seen that really helps think this through is this paradigm, or this continuum rather, between answers to actions. And if a use case is providing answers, meaning you're querying your information or you're looking for a coherent response from corporate data maybe, or a corporate library, et cetera, or public based information, or you want to talk to your database in a way that allows you to use natural language. Those use cases are are data retrieval oriented and they're much, much simpler than an actions-based use case. As soon as we introduce data science, as soon as we introduce decision-making, then the risk profile changes massively and the talent required to build the decision models, and tweak them, and constantly be tweaking them. It's very difficult for most organizations, not just from a talent perspective, but from a strategy perspective to get to that end of the spectrum.

So, I think it's likely that the use cases that are most commonly deployed in production for the first wave, maybe even the second wave, are on the answers side of that answers to actions continuum.  

[00:18:36] Mike Reeves: You hit on one other thing that I wanted to chat about, and this is one I think it's almost palpable in every discussion when you start going into a discussion with the company around AI, and that is the data piece. Security, governance, compliance. Because I see so many companies get stuck there. And it really gets in the way of trying to move further with anything else around trying to plan for, develop a strategy, look at use cases, et cetera.  

You touched on it a bit there, I don't know if you have any other comments or observations, because I find that a lot of companies, I can say honestly that we've been guilty of this earlier on, and we would just stop and say, "Okay, I guess we can't do this." And I see companies like Dell Technologies trying to help customers get their arms around that and solve for that.

It's a fundamental problem. I almost call it more of an infrastructure related problem because the data is a core element. It's a foundational piece to what you want to build from. So, I don't know if you have any comments around that.  

[00:19:46] Chris Falloon: I mean, we just talked about the predominant use cases of answers, that category of answer based use cases. The main ingredient in an answer in most cases is your data and data readiness, data platform readiness, not just the availability of it itself but how it's managed, how it's how it's generated, where it's stored, how available is it, how controlled is it. Data platform management is probably the least... In terms of technology service readiness in most organizations, the data platform is by far the biggest gap, I think, in most organizations for the idea of deploying AI use or ML use cases at scale.

The business case for data platform transformation has been very difficult in most organizations. Data is always, almost always an afterthought, data platforms are not inexpensive, the software is not cheap, and the talent to operate them and write coherent data platform strategies is not always inside of every organization. So, we end up with massive data sprawl. We end up with massive data platform incoherency, copies upon copies upon copies of corporate data. With all of that comes risk and security.  

We talked a bit this morning about this idea of minimum viable platform or starting small and allowing your platform decisions to evolve with your business. That's a core principle of every platform, whether it's a cloud platform, an AI platform, but it's definitely the right strategy for data platform transformation is to pick a desired end state, pick a North star of where you want to be from a data coherency, data management standpoint, and then allow yourself to only make decisions that get you closer and closer to that end state. It's so expensive, as I say, to transform data, it's not going to happen overnight.  

Maybe the last thing said there is that, we've never had the business case to get our arms around the data problem in an organization holistically because the ROI has never been there. It might be in some organizations that these AI and gen AI use cases are finally the business case that drives us to make coherent investments in data management because there's a clear ROI in doing it correctly. And there's a clear, clear danger in not following... The governance and risk pieces alone are, as you say, they're a showstopper.  

[00:22:09] Mike Reeves: Just a couple other things to cover off here. It's been great discussion, thank you. So this would be more your opinion. So Chris's opinion, what does success look like for an organization that's trying to go down the path of, whether it's a use case, setting up the environment and the platform. What would success look like either as a starting point just to kind of get going or maybe it's maturity. I'll let you decide which way you want to go with that. It's going to be more opinion.  

[00:22:48] Chris Falloon: Yeah, that's a really interesting question.

And of course it would be opinion, I think. Look, I think there's... I think success at this early stage is keeping yourself from falling into some predictable, mistakes, some early mistakes that will, I think, really cause some regret midway through next year or the year after as you start to realize that you've either built technology silos because you haven't thought about this horizontally, or you've selected tools that are only available in one premise. Maybe you've doubled down on a tool set that's available in Google or Amazon or on-prem and they're not available anywhere else. Therefore, you've made some early bets that didn't allow you to be flexible. But avoiding the key pitfalls might be the number one pointer of success, but that's not really success.

I would say in terms of delivering outcomes under severe pressure from the board, I think we've got a lot of technology organizations that are desperate to stand something up to show that they A) know what they're doing, and B) that they can keep their jobs, because there's a lot of folks that are demanding that they make this pivot.

And what we do when we have unrequited demand from senior leadership, we often build something because we can, not following a North star or not following some sort of plan.  

And so those three things are related. I think if we can: identify the potential pitfalls, build ourselves a high level North Star for our data platform strategy, our cloud platform strategy, and our AI.

And those North Stars can be as simple as, we're not going to choose tools that lock us into a premise, we're not going to have data silos that create a bigger problem than we already have, and we're not going to launch things that we don't know how to control, right? We're not going to put ourselves in a place of increased risk.

If you just followed North stars that had that type of objectivity and granularity, then you can start to make principled decisions. You can build something that not only shows that your tech teams can deliver on an initial use case, or use cases, but you can also articulate that you have a vision. As simple as it might be, what you're building today is extensible and follows the business. And I think that if you can build something now and show that it's future-proof, that's as good as you're going to get in the short term.

[00:25:08] Mike Reeves: You actually took on my last question by answering the previous one. A lot of good ideas in there and it's great.  

To me, one of the things I hope that our teams are trying to do and that is that customers need help and they need knowledge and experience. And as you say, there's no recipe or, prescription that's available yet in terms of here's the approach you want to use. There may never be, but some references, or reference material, or architectures and things to lean on. So, it's great and I really appreciate you taking the time to share your experience from all your travels and your customer meetings.

I just find that there's so much wealth of knowledge you can share, with all the interactions that you have and...  

[00:26:02] Chris Falloon: I'm happy to have my contact information dropped into the show notes here, that's that's probably the easiest way, and happy to take an email. The partnership, with you guys is one of many, but I'm really pleased to see the outturn of folks here, many of whom were in my rolodex from early in my career, including a couple of folks inside your organization.

Very happy to work with you guys and I'm happy to have anyone reach out. I'll direct you to someone in our business that can help.  

[00:26:29] Mike Reeves: Great. Wonderful. Well, I appreciate your time today. And again, thanks for the partnership.  

[00:26:33] Chris Falloon: Thanks very much. Take care.  

[00:26:34] Mike Reeves: Thanks, Chris.  

Thank you for listening to solving for change. If you enjoyed this episode, leave us a rating and review on your favourite podcast service. Join us for our next episode. Thanks very much.

‍