Asking the Right Questions
with Frank Farrall, Strategy & Analytics Ecosystems and Alliances Leader, Deloitte
Strategy & Analytics Ecosystems and Alliances Leader, Deloitte
Frank Farrall is a leader in the growth of AI/ML, hyperscaler, cloud data platform, digital content, and e-commerce alliances at Deloitte, where he helped found the Deloitte Digital practice and is the company's lead principal for data and AI alliances. He has also shared his thought leadership in media outlets including CNBC and The Wall Street Journal.
Co-founder & CEO of Alation
As the Co-founder and CEO of Alation, Satyen lives his passion of empowering a curious and rational world by fundamentally improving the way data consumers, creators, and stewards find, understand, and trust data. Industry insiders call him a visionary entrepreneur. Those who meet him call him warm and down-to-earth. His kids call him “Dad.”
Producer 1: (00:01)
Hello and welcome to Data Radicals. In today's episode, Satyen sits down with Frank Farrall. As Deloitte's AI and data alliances leader, Frank is responsible for developing smart business strategies and implementing AI solutions to help drive client growth and efficiency. A widely respected industry leader, Frank has built a reputation for delivering data and digital transformation results for healthcare, financial, and telecommunications organizations. In this episode, Satyen and Frank discuss prompt engineering, data modernization, and investing ahead of the AI curve.
Producer 2: (00:36)
This podcast is brought to you by Alation. Does data governance get a bad rap at your business? Today, Alation customers wield governance to drive business value, increase efficiencies, and reduce risk. Learn how to reposition data governance as a business enabler, not a roadblock. Download the white paper, Data Governance is Valuable: Moving to an Offensive Strategy at alation.com/offense.
Satyen Sangani: (01:08)
Today on Data Radicals, we have Frank Farrall. Frank is one of the founding partners of the Deloitte Digital Practice. A global business builder with 20 years of experience in AI and digital transformation, he's helped clients in all industries and scaled startups become billion-dollar businesses. In his current role as artificial intelligence ecosystem leader, his specialty is in the application of innovative cloud, AI, and analytics technologies in a business context to support strategic change initiatives. He has contributed to CNBC, Wall Street Journal, CIO, Mashable, and more. Frank, welcome to Data Radicals.
Frank Farrall: (01:39)
Satyen Sangani: (01:41)
So you, Frank, have been at Deloitte for 15 years, and we've, of course, all heard of Deloitte. Tell us something that we wouldn't know about the firm.
Frank Farrall: (01:49)
When I talk to our clients and our technology partners, most of them don't realize that our firm is about 177 years old. So we have a very significant history. We started in London, in the UK, and we have built up a significant legacy of relationships with clients. And so what's really great about that is, in a lot of the clients we've been in, we're there and are able to help them evolve over quite a significant timeframe. So we've been with them on the journey, and so as we're making recommendations to them and we're helping them go through significant change, because we have such history and such a significant track record, we're able to relate more with what their challenges are and and what they're trying to do. I don't think most people realize how long we actually have been in operation and existence as a firm.
Satyen Sangani: (02:36)
Yeah, I mean, 177 years feels like a pretty long time. Well before we're even having the AI conversation.
Frank Farrall: (02:42)
Defining the AI ecosystem
Satyen Sangani: (02:42)
It wasn't probably a glimmer in folks' eyes. And so your job is to be the AI ecosystems leader. What exactly does that mean? AI is its own thing. There's ecosystems, which is another thing. Those are both often fraught and overused words. Tell us about the combination of the two at Deloitte.
Frank Farrall: (02:57)
We are seeing a significant explosion in the number of companies that are coming to market that focus on AI topics. And particularly with generative AI, we're seeing significant venture capital money coming into that space. We're seeing many of our most strategic, longest-existing technology partner relationships invest heavily in artificial intelligence. And so, someone has to make sense of what is happening here. What are these technologies? How do they apply to our clients' business challenges and opportunities? How do they all work together? Particularly in a cloud-oriented world with hyperscalers as a foundational technology, in a lot of cases, how do you bring together an ecosystem we'll refer to it as a constellation in a lot of cases to make sure that the right technology is being applied to the right business topics and opportunities and challenges?
Frank Farrall: (03:53)
And then there is quite a lot of management that comes with bringing that together, both inside our firm — so, understanding what these technology companies offer — and then how they interface and then how that solves the client problem. So my role is spent heavily oriented around that education piece and then the understanding of the client's needs and delivering that, and then the relationship building with the companies that we're working with. So that forms an ecosystem... There's actually multiple ecosystems given specific circumstances, but then there's a broader set of companies that we see working together in an ecosystem. So my job is to make sense of that and apply that to maximum effect to our clients.
Satyen Sangani: (04:31)
That sounds like a fairly difficult job because there's so many different AI technologies that already exist that are themselves changing. So like even thinking about a company like Alation, like our AI offering is changing quite significantly and quite consistently. And then you've got all the new entrants that are getting funded. And today, it seems like every day there's another AI or generative AI startup that I'm hearing about that helps somebody or something do something slightly better. How do you, one, keep up with all of this change? Like what are the hacks that you use in order to figure out, our customers have to do the exact same thing? So what are the hacks to keep up? And then how do you decide, for Deloitte, when to double down on a partner or when to keep them at bay?
Frank Farrall: (05:14)
Those are good questions. So, basically, it's constant information consumption and more so in say the last... Well, ChatGPT was launched in November 30 last year. Ever since then, the amount of time I'm having to spend reading and picking up information on all the social channels, the newsfeeds that I anchor on, the volume of the content and changes has gone up significantly. It's been exciting and really interesting but it's been a real challenge too in that timeframe. And so, basically, where we start is with our clients. So we're always looking at, what are our clients' needs, what are our clients talking about. In a lot of cases, with the generative AI piece recently, our clients are calling us. They're saying, "Hey, do you have a point of view on this? We need to get informed we need to figure this out." And so, sometimes it's around a specific technology partner or alliance, sometimes it's about a particular problem or opportunity and we'll look at that, but it's constant information consumption.
Frank Farrall: (06:11)
Now, also, I used the term constellation. What we find is that a number of our particularly larger more modern and successful technology partners will have gravity around them. And so they will have particular ecosystems where in a lot of cases they have a venture arm. And so that venture arm has made investments in these somewhat early-stage or fairly early-stage companies. And that is a really strong signal to us that if one of the companies that's already a strategic partner of ours is making a formal investment — in some cases a pretty significant formal investment — that gets our attention. And then we start to think about, "Okay, how do we work together again in an ecosystem play?" So how do the three of us then go and work at a client? What does that look like in terms of overall market need and architecture and things? So that is a really, really strong signal to us.
Frank Farrall: (07:02)
And then, as we are out engaging with our technology partners, whether it's, again, in client efforts or the yearly annual conference programs and things like that, we listen to who we hear about and who wins the awards and things. And you start to see certain names get repeated over and over and over again, and that signals us very strongly. So again, it's really with us starting with our most strategic technology alliances and partnerships and emanating out from there.
Satyen Sangani: (07:29)
That sounds pretty familiar to me. I mean, we, obviously, like all companies, had a lot in our 10-year history of exposure to various SI practices of the global SIs at various levels of scale. But I don't think we found or had found success at scale that was repeatable until pretty recently, like over the last year or two, and I think a lot of that has to do with the fact that, for a company like Deloitte, you just can't invest in every technology.
Frank Farrall: (07:55)
How to identify a worthy investment
Satyen Sangani: (07:55)
Enabling a massive practice is not something that you'll do lightly, and without that level of scale you're just not gonna have enough of a message to get to all of these different customers and obviously to all of your consultants who have to be enabled and then trained en masse. Are there cases that you decide to invest ahead of the curve where there's this constellation of both customer demand, existing strategic-partner-sort-of-gravity, if you will, and marketing or so also media proof. That's a constellation of three different things. Is there ever a time where you'll invest ahead of those three signals, and if so, when does that happen and how does that happen?
Frank Farrall: (08:29)
We absolutely will. We take it very seriously and it may not be a fast decision, but the best situation for that is when there is a really important client involved. So there may be an emerging, fairly small technology partner, or they're not a partner yet, a company that has managed to... They've got a relationship inside a client and they've landed a project and they really need help with services. And in most cases, we're already there. We're already working with that client, and this company, there's kind of like a, in a lot of cases, it's a client introduction. "Hey, we're really excited about this new company. We'd like you to look at them," and yeah, we have a conversation. There's a program of work. There's something specific. So it's not a theoretical discussion. There is a specific program of work that needs to be addressed for an important client and this new company seems to be important for that.
Frank Farrall: (09:20)
So we will put the effort in to maybe do some training, learn their technology, build some relationships, and the like. And then if that project is successful, then you have a case study, and then that's a key thing for... Basically, inside Deloitte, you have to activate our leaders — who go to market and serve clients — around believing this is a valuable partnership, this is a valuable capability for our clients. So if you have a case study around, there is a successful project, it's a modern topic, it's an important client, this has been delivered, then you get the increased confidence that you could do this again, you could replicate this again. And then the wins lead to confidence, and then that confidence leads to a greater willingness to invest ahead of the curve. And we have had a number of our technology relationships that have scaled that way.
Frank Farrall: (10:10)
We tend to prefer that there's a level of scale because we're 400,000 people globally. And so, while it's difficult for us to manage the kind of relationships with a smaller company, we can be absolutely crushing onto that company in terms of the reach-outs and if we think that there's some excitement around that. So having some level of scale helps that company almost cope and and deal with us if there is something exciting that does move ahead.
Satyen Sangani: (10:35)
That case study element I think is the sort of Occam's razor, as it were, for all successful tech and IEC partnerships, I think, or even SI partnerships. I mean, it's funny because when we were growing up earlier in Alation, we talked to a company, I think it was like, I remember talking to Cloudera and I'm like, "We could really help you and we could really help what you were doing." And this is very early, like back in like 2013, 2014. And they were like, "Yeah, you know, that's not really all that exciting. Can you show us a case study? Are there any customers that we've got?" And I was like, "But like we could really help you," and I didn't, at the time, quite understand that that was actually the right answer. What they were doing was the right thing. And this is exactly what I do with our partners or prospective or would-be partners, is just to say, "Look, this is so cool, so interesting. What a great demo. But you’ve got to get three customers that we have in common, otherwise I'm not gonna spend the time of day because we've just got too many people already trying to pursue our reps." It's a very thoughtful thing.
Why AI — and why now?
Satyen Sangani: (11:26)
What made you decide to do this job now? I mean, was the interest in AI, was the interest in tracking the ecosystem? What made you decide that this is the role that you wanted to take on?
Frank Farrall: (11:35)
It's a great question. It started with digital. I ran a digital practice in Asia Pacific for a number of years. And in doing that, I started working with partnerships, particularly Salesforce and Adobe, and that muscle memory built up, and I really enjoy the aspects of having to understand a technology and a set of scope, and then building the relationships with the companies and then seeing them scale. It's very exciting. You wanna be in a high-growth, energetic type of a mindset in professional services. So when your technology partners are growing that fast and having success, it's really good. I anchored myself in digital and data, and then technology partnerships. And then it became very clear, we made a significant investment as a firm about three years ago around artificial intelligence offerings. We were very convinced that cloud was becoming an anchoring set of offerings and capabilities. Artificial intelligence was gonna take everything we were doing to the next level.
Frank Farrall: (12:31)
All of our clients were gonna need to adopt it in one way or another, but it's a specific space. There's a specific set of topics, there's a specific set of personalities, there's a new set of companies, and now with generative AI fueling it, there's even more so. There's almost like an initial level, four or five years ago, of growth. There is now this VC money coming in and we're seeing an explosion, and in fact even ventures funds and hyperscaler investment going into companies that have been relatively small, but we expect them to scale really rapidly. It's an exciting space and it's something that our clients need to focus on. I think it's a really important key to our future going forward.
AI’s “iPhone moment”
Satyen Sangani: (13:11)
Yeah. So I'll ask the somewhat obvious and maybe dumb question, why is it so exciting right now? What is it about right now that's changed? When did it change? What's transformed it and what's making the conversation happen for you with your customers?
Frank Farrall: (13:25)
We saw the same thing with mobile phones when the iPhone came out. It was that moment where the usability was such that a mid-level executive at one of our clients would call us up and say, "Hey, I think there's something here that our customers could use to transact with us, to interact." An airline executive: "I think my customers would check in on their flight for their flight on this." Banking executive: "I think that our customers would bank on this. Can you come talk to us around what this might do?" And so we've been talking about artificial intelligence for a few years. We have seen adoption of artificial intelligence in our clients. However, when ChatGPT launched late last year, that was the iPhone moment for AI, where your typical executive that does not have a deep technology background, that had not been doing proof of concepts around AI for multiple years, could interact with the technology and say, "Okay, I could visualize how this would transform a number of our functions, how this might disrupt us, how it could give us competitive advantage. We now need to act."
Frank Farrall: (14:26)
And so, those executives saw that, the VCs saw that, a bunch of individuals that were thinking about starting up companies saw that. And so we're having this almost Cambrian explosion of generative AI–oriented companies. And so the growth is exciting. The fact that you have to keep up with the news almost every day — definitely every week — is exciting for people that are intellectually curious and technology-oriented. So it's, I think, a very special moment in time. I think we will look back on this in 10, 20, however many years, this last 12 months will have been a really important period in the history of technology.
Satyen Sangani: (15:02)
Yeah, and it's coming, interestingly enough, at a time where the rest of software would've otherwise been in a secular recession, right?
Frank Farrall: (15:09)
The intersection of AI and data
Satyen Sangani: (15:10)
Like massive runup in investment, particularly in data technologies. Everybody saw Snowflake and everybody thought every data company was gonna become Snowflake. That didn't seem to happen at scale. And now, all of a sudden, here's yet another trend to save the day. And you also lived in the data ecosystem where we obviously have grown up and we've been at the intersection of these two things. How do you see the intersection of data and AI happening, and is that a conversation that's happening with your clients?
Frank Farrall: (15:37)
Satyen Sangani: (15:38)
Yeah. So tell us more about that.
Frank Farrall: (15:39)
In order to have an effective AI application or capability, you have to have good data. You have to have lots of data, and it has to be good data. If you've got bad data, and you've got really good AI and an interface, then you get a bad result faster in a more confident or quirky way. And we have been pushing particularly cloud-oriented modernization very heavily for the last five or six years or so. And so you mentioned Snowflake and Databricks, the hyperscalers, they're all investing in this, having a lot of success. The market is really changing, and I think that is fertilizing the ground for adoption of a lot of the AI technologies. And our clients that were already doing the right thing — so, getting themselves organized around modernizing, moving to cloud, getting data cleaned and out of silos and the like — that was important work, and work that needs to continue anyway, and it's gonna add to their capability as generative AI becomes something that they adopt. And so I'm actually really excited to see the intersection of the cloud data modernization we've been doing for a number of years and the new capability around generative AI. And I think the organizations that get that right are gonna have competitive advantage.
Initial AI investments
Satyen Sangani: (16:46)
What are the projects that people are investing in right now around generative AI? What are the first-order, first-generation projects that are happening now with ChatGPT getting releases? What trends do you see, if any?
Frank Farrall: (16:56)
Yeah, there's a few flavors of these. One is, my executive boss has been playing with ChatGPT and thinks it's really important. I saw an interview with the CEO on the news last night, and we need to figure out what are the use cases. And there's a real spectrum, too. Generative AI is still new and immature. So there are certain things that it's quite good at. There are certain things it's quite bad at. Even individuals that lead the companies that produce this technology have said it's not good at certain things and it'll probably get there, but you've got to be where it's at right now. And so a lot of the work with clients right now is, what are the use cases and how do we basically programmatize the adoption of these?
Frank Farrall: (17:37)
So if you're a life sciences company, doing the summarization of publicly available FDA information is an easy use case. It's something you can do really quickly. Drug discovery and doing really complicated things with molecular biology, generative AI is not gonna be perfect for that right now. It's gonna take some time for that to evolve. And then there's everything in the middle around AI-assisted and autonomous coding and the adoption of that, and IT transformation, the adoption of text and imagery capability, in a marketing function around marketing transformation. So basically every function in an organization will be touched by generative AI, and I think can be improved by generative AI. But you’ve got to figure out, some of these are low risks, some of these are medium risks, some of these are higher and require additional advances in the technology. So the use case piece is something that's really important that we're working with clients.
Frank Farrall: (18:27)
The other is, I am doing a cloud modernization, a big cloud transformation program. And some of our clients have a primary hyperscaler, and some of them are multi-cloud and they've got their pick of all three. The conversations that we're having around generative AI at a technology level is what do I need to do around security and architecture — aligned to what I'm already doing in my cloud transformation with my hyperscaler — to be able to adopt this for a set of certain use cases? And then as the technology advances, how do I build AI and autonomous coding into my DevOps process? How do I make sure that the testing is effective? Can I 10x my developers? Do I turn some of them into prompt engineers? And that's a new term that we're working through. The actual work that's done changes. So there could be quite a technical angle to this, most likely already aligned to a program that's ongoing, and then at some point, these two intersect.
Satyen Sangani: (19:25)
Yeah. That's consistent with what we've seen, too. I think the internal productivity use cases — where you either have to generate English or some other language and respond to English or some other language — seem to be the ones where there's a first-order benefit. Like our customer success or support reps are now able to respond to tickets faster. In theory, our programmers, with Copilot, are able to program and code faster. Our SDRs are able to write emails faster. So those are I think some fairly interesting use cases where it's just making people more efficient right this very second.
Frank Farrall: (20:00)
Use cases and challenges
Satyen Sangani: (20:00)
Which I think is super interesting. Now, let's get back to that intersection between data and AI. So, obviously, you've got to feed data to these generative AI applications or something to these generative AI applications, and in most cases, you're feeding English. But in some cases, you'd love to feed data, and that interface to me still has a lot to go get figured out. Any use cases there that you've seen? And I guess what are the problems that you've seen in that work?
Frank Farrall: (20:24)
I guess the biggest challenge is, how do you get data out of silos? And in a lot of cases, you've got a mix of on-prem versus cloud. You don't have the explainability. I actually think the reliability and the explainability is gonna be the next frontier that has to be tackled for generative AI. I think in order to be confident in it, you're gonna need to ask it to cite sources, and so in citing sources, it's gonna have to have a lineage and there's gonna have to be governance around it. Another thing I'm seeing too is that, with the generative AI piece, because of the level of risk, whether it's hallucination or bias or things, we're seeing a much more centralized approach to governance right now around this.
Frank Farrall: (21:04)
A number of our clients are, and including Deloitte, has been highly federated in the past. With this topic, in order to get it right, you have got to have all of your right legal and compliance people aligned on the topic, and then there's got to be really strong communications, and then the pathways around the data pipelining. And then, coming back to the hyperscaler discussion, you got to figure out which of these tools are you using, which of the tools are more advanced and more mature and which are really new? And then what's the right timing around that? Then, again, you got your security and those kind of considerations. So it's linking this all together, starting from your programs, your transformation programs that you're already running, and then extending them out with a new set of topics to consider.
Satyen Sangani: (21:48)
Yeah, and it sort of ups the ante on making sure that your data state is in order.
Frank Farrall: (21:53)
And I think it makes it real for non-technical or even technical people that aren't deeply into data governance and MDM and some of the topics that we might take for granted. And so if it's like, okay, well, if there's this really exciting employee Copilot that we could deploy across our entire organization, we’ve got to make sure that this tool, this capability has got the right kind of things, the right kind of information, and so here's the governance and the MDM that it needs to align to. Oh okay, that's important now, and so whereas I think it used to be taken for granted or not understood, when you have a conversation like that about something — an application that important people are really excited about… I mean, data quality is another good example. If you can explain to an executive why data quality is gonna impact their potential Copilot across the organization, that gets the support then to address things that maybe for the last decade you haven't been able to deal with.
Satyen Sangani: (22:47)
Yeah, it's super interesting. I mean, the way I describe the category that we're a part of, whatever, cataloging governance, is this broader category that IDC calls data intelligence. And what I say is, “You've all heard of business intelligence, because that's about giving people reports and helping people make decisions, and you've all, of course, also heard about artificial intelligence, which is about helping machines make decisions.” And I think people then sort of say, "Oh well, I don't necessarily have to organize my data." But then when they realize that helping machines make decisions actually is totally a function of great data, then I think that's gonna even accelerate those investments even more.
How sexy is “prompt engineering”?
Satyen Sangani: (23:23)
At a time where I think, during a recession, what people are looking for is quick ROI, and what seems to be the case is now instead of doing the BI stuff where the humans are getting more efficient that then they're like, "Ah well, why don't we make the humans more efficient via this AI thing?" So it's a funny, interesting moment and I think it's in some sense put a second set of turbo-boosters in the data ecosystem, which obviously, from where we're sited, I'm pretty excited about. You mentioned, I wanna go back to a term though, because you mentioned this idea of prompt engineering.
Frank Farrall: (23:50)
Satyen Sangani: (23:50)
Which now, everybody who's like, the first thing you say is gen AI, and the next thing you say is prompt engineering. Is prompt engineering the sexiest job of the 21st century, displacing data science? And, I guess, tell us what it is first.
Frank Farrall: (24:02)
I don't know if it's the sexiest, but I was actually having lunch with a set of colleagues several weeks ago whose children were just about to graduate from college. And I was saying, make sure they learn prompt engineering because they're gonna need that in their future. And basically, the prompt engineer... Prompting is, it's using text mostly but also you can use voice and other modalities, to tell the AI what you want. So if you're trying to make an image, you're describing the image that you want. If you're trying to get information, it's the question that you ask. It's similar to search like when you go in and put in something around search. And what we're finding is that the different models, again, whether they're text or imagery or audio, have gotten nuances in them. And so a very good prompt engineer, and some of the imagery models, I've seen five-paragraph prompts to get to very sophisticated imagery that actually wins art awards.
Frank Farrall: (24:57)
Now, there's also some controversy around, should AI be winning art awards and arts for humans and things like that? So that's a societal debate that we'll stay away from. But how you get there to that outcome comes down to a skill around telling the AI what you want: what kind of lighting, what kind of expression what kind of style, what kind of angle, even, and it gets quite complicated. And I've heard senior executives from different AI companies saying the most sophisticated prompt engineers are almost like prompt whisperers. They'll spend like 12 hours a day interacting with these models and you get to the point where you can see the code, the matrix, and you can get that standout award winning image or the perfect set of text or answer to your question if you can understand the nuance of the model and get to that.
Frank Farrall: (25:43)
So I think, in a lot of cases, prompt engineering will at least become a skill that knowledge worker, creative workers use to get an outcome from the technology. And I think some people will be highly, highly specialized, and I actually think your kind of prompt libraries, so we're talking about data governance and MDM and things like that, I actually think prompts are gonna have value in organizations, and I think prompt libraries and how you manage prompts will be become a set of IP and something that's highly valuable inside organizations. So I think prompt engineering has a very significant future ahead of it and I think all of us are gonna have to learn some level of prompt engineering to be effective in the future.
Efficient engagement with AI
Satyen Sangani: (26:25)
Why do you need it in a world where, for example, Google just took searches that people did and used it to improve search? So the way Google had worked was they just said, "Look, we're gonna build this thing called page rank, which basically looks at actual behavior, what people actually do, what people actually link to most critically, and the most important webpages are the ones that are most linked to. And then we're gonna look at the searches that people do and their confirmation or denial of any given search response — we'll further edit that claim or that correspondence." Why do you even need prompt engineering? Why can't we just watch what people do with these models and just do it that way?
Frank Farrall: (27:03)
The prompt engineering is how you engage with the model to get something that you want. I'm probably more familiar on this topic with large language models, and what's interesting is the way that you've got this self-attention transformer architecture, big thing that came out in the Google paper that was released in 2017 around how words interact in language and parallelize that massively. You can obtain a really deep contextual understanding of how words relate to each other and things. And so, what's really interesting is that you think about what language is, it's an abstraction of the world. We use words to describe objects and concepts and ideas and things that we're trying to make real and communicate. So if you have a very detailed, contextualized understanding of language, then you actually have a really strong abstraction of the world.
Frank Farrall: (27:56)
Now, it's gonna be tainted by human perspectives and perceptions and things like that because they put the information in, but as a starting point, you've got some really interesting ability to then look at this abstraction and determine reasoning. And then I think as the models get more and more sophisticated, then they're gonna be able to pick out kind of like how AI now beats us at Go and in chess in ways that like using moves that we would never have used as humans and never anticipated. I actually think as these models get more and more sophisticated, there will be contextualized awareness of the world that we wouldn't have thought of, necessarily. And so there's gonna be some interesting learnings. And then I think effective prompt engineers, coming back to that topic, who can unpick that and unlock that, I think are gonna be almost like magicians in terms of the skills that they have to interface with the AI and deliver a result that we just wouldn't have been able to do just with people alone.
Satyen Sangani: (28:48)
Yeah, for sure. And, in some sense, I think what it gets back to is this idea of AI — that, like a baby — learns, and early learnings are fundamentally more impactful than later ones. And so if you can train the model in a certain direction, then you're going to be releasing and yielding a much more thoughtful model, because the early training data was really good. And it's interesting because the same thing is true for many other cases. If you look at online communities, the early 15, 20, 30, 50 users of a community set the characteristic and the behavioral patterns for the community. And so this idea of learning is the key concept. And the idea that we can all be prompt engineers is a little interesting because obviously then it sort of allows us to all develop some level of expertise, which can be pretty fun and exciting and democratic in a world where this technology, in theory, could be also pretty scary.
Frank Farrall: (29:37)
Is the modern data stack yesterday’s news?
Satyen Sangani: (29:37)
And hard to understand. So, I wanna get back to the data topic obviously because that's how we roll here at Data Radicals. There's been this trend in the modern data stack, which I think prior to the advent of all these AI technologies that are obviously getting generated through this generative AI movement, ChatGPT and the like, you've got, before that, this idea of the modern data stack, which really was sort of generated through this move to cloud computing from Redshift, modular Databricks, modular Snowflake. Tell us a little bit about the modern data stack. In your view, is that trend still as interesting as it was yesterday? Are you optimistic, bullish? I was recently at a dinner conversation where half the CDOs in the room were like, "Ah, that's a not-interesting name anymore for a trend." Would love your perspective.
Frank Farrall: (30:21)
Well, it's nice when you can take something for granted. It's obvious that you can't run a Fortune 500 organization or a really fast-growing early-stage company that has expectations to scale with 18- to 20-year-old data technology. It doesn't scale, particularly when you're looking at applications that require heavy compute like artificial intelligence and what we're talking about. I think that maybe the sizzle, in terms of like marketing and some new words around cloud platforms and things like that, have come out, but it's foundational. I think what you're saying there is like, the CDOs are taking for granted. They need to move into the cloud and they need to have cloud architectures. And if I look at the virtualization and the scale capability, I think that's hugely important. And I know, too, with the economy being a little bit wobbly right now and organizations worrying about costs — and the cloud data providers hate this, but this is the reality of the world — these organizations have the ability to scale back a little bit when they need to.
Frank Farrall: (31:20)
I think the flexibility that it gives them is tremendous. So in the good times, you can add and add and add and grow and drive that, and then sometimes, when the organizations need to, they can scale it back a little bit. I think that gives... I know a number of the clients that we work with, the CDOs and CIOs and things, it gives them the confidence that “I can adopt this technology. It's the right type of thing that I need in terms of my virtualization capacity. I can sunset this stuff that's about to fall over anyway because it's old and it's obsolete, and it gives me the ability then to plug in the new AI players, the whole ecosystem concept.” I've got companies that have to have this integrate and work well, whether it's a marketplace or an ISV program or whatever the case may be, in order for their offerings and products to be successful. And then that gives me access to things to be able to keep growing and evolving. It's just really painful to do that, that migration and that movement.
Frank Farrall: (32:11)
Once you get it done, it totally unlocks capability, and we've seen a lot of our clients really benefit from it. You go through the pain, you get a business case that might be a cost takeout or something like that, but then what you can do, whether it's data sharing or AI capability or things like that, it unlocks a whole new set of capabilities. And so, it may not be brand new, but it's foundational now, I think, in our clients.
The modern modernization projects
Satyen Sangani: (32:35)
Which I guess leads us to this idea of sort of data modernization and the stack modernization. What kinds of projects are you seeing get the most traction and modernization today? Where are companies investing? Because in this world where ROI is more proximate, how does that impact what people are actually doing?
Frank Farrall: (32:56)
So we're seeing this massive trend — it's probably gonna run another 10 years or so — around our clients moving to the cloud. And there was a debate not that long ago around would a bank or a pharma, because of PHI or regulatory concerns, ever move to the cloud. That debate is over. Everybody is moving to the cloud. It's a question of timing, it's a question of... There'll be some hybrid architectures and things like that, but that is settled. So there is this massive movement to the cloud, and in a lot of cases with our clients is, "Oh yeah, we've gotta move the data." So the initial business case, maybe, we wanna shut down our data centers and get rid of those, or apps modernization drives, a lot of it. There's a business case around “We've got to be able to provide modern tools to our customers and things like that.” Oh yeah, and then the data piece, and then in a lot of cases too, like maybe Hadoop never really worked for them completely, so it's a bit of a reset.
Frank Farrall: (33:42)
We are still seeing a lot of data modernization. Some things are happening a little slower than they were, say, 12 months and definitely 24 months ago. And there is more of a push around, “Let's either get things decommissioned really quickly and get the cost savings, or let's get a new capability turned on and get that benefit as soon as possible.” And then again, like with the cloud piece, you can kind of scale it up and down. So they feel confident like, okay, we can make this investment. We had it planned anyway, we're gonna do it, if we need to scale it down for a period of time. And this is why I said the cloud companies should be happy about this. Even though it might mean a little bit of short-term pain, it provides the confidence for those CDOs and the CIOs. Let's continue doing it, let's go ahead and do it because we can manage it as we take that forward.
Frank Farrall: (34:28)
So we are still seeing a lot of move to the cloud, and then a lot of the conversations around the artificial... Like now that I'm in the cloud, what can I do? What kind of new data-oriented business models? What kind of sharing with other partners can I do? What kind of AI capability can I turn on? So I think there's kind of a ticket to play there that is still important, and we're still seeing that progress, but it's the things that add on top that make the executives excited.
Measurement for investment
Satyen Sangani: (34:56)
How do they justify this ROI? Because data projects, even in the best of times, are hard to measure, or hard to assess. What are you seeing as the things that get people over the line convinced that they should make investment right this very second in this technology for a new capability?
Frank Farrall: (35:15)
It's the same thing that I used to see with customer experience investments. Everybody knows you need to do this and this is gonna be important. It's gonna make the customer happier, it's gonna lead to better results. You're gonna feel prouder about your product. You'll protect your competitive positioning. But it's hard to make the business case work, our customer experience. Well, it's the same thing around data. What happens in a lot of cases is the cost takeout one is the easiest one to make. And in uncertain economic times, that's the one that is most likely to get signed off on. So hey, we got a data center that we're shutting down. We got 18-year-old analytics appliances that need to be shut off, they're end-of-life anyway. We think we can save $25 million a year if we move to a cloud platform. So, check, you get that signed off and you're able to get that approved. Now, knowing that that's going to progress, then you get to do the AI projects, the good stuff on top of that. But in a lot of cases, the argument with the COO and the CFO is very much around cost takeout, operational efficiency. That's the winner every time.
Where is the greatest (and least) AI investment?
Satyen Sangani: (36:14)
Yeah, it makes complete sense, and I think obviously for those people who are then trying to justify that business case, now they have a language and a set of comparables in order to be able to reference in making that investment internally within their companies. Maybe switching gears a little bit or continuing on a pull on that thread, which industries are investing faster right now and which ones are slowing down more?
Satyen Sangani: (36:37)
It's a really a good question. So the ones that are very data-rich, so your financial services, your insurance companies, your banks, your life sciences, healthcare, pharma, they've got a lot of data, they've got a lot of regulatory demands, too. So, coming back to your previous question, you have to satisfy the regulators. And so where there's a regulation around data, you must do that. It's kind of like remodeling your bathroom because you have got a plumbing update that you need to make. So okay, we have to do this because of the regulators, but while we're here, let's go ahead and make this great. And so we're seeing that a lot in highly regulated industries. And then also like in your telco and media, particularly like with streaming where an area has become highly, highly competitive, and data and AI enabling that is crucial to success. It's absolutely competitive differentiation. We're seeing continued investments in that. So in some of your more like heavy industry type organizations, it's definitely happening but not happening at the pace as to what we're seeing in like financial services, life sciences and media.
Satyen Sangani: (37:38)
And which ones are suffering, if any?
Frank Farrall: (37:40)
That's a really good question, actually, and it depends on how you define “suffering” because I would say every one of our sectors, there's uncertainty in the economy, whether it's interest rates or a recession or things like that. So everybody's tapping the brake to be prudent. Clearly, we've seen more layoffs and disruption in the tech sector and that gets a lot of the media attention, and that then, if you're a Deloitte, you've got the tech sector as clients and then you get the tech sector as really strong partnerships, and so that actually impacts on a couple of angles. So that's probably the No. 1 most effective that I've seen. Everybody is hesitant and cautious, but nobody has gone in to batten down the hatches mode. And then what's really interesting is, even where our clients have done layoffs or there's been a sector with a lot of layoffs, there are still investments happening in the things that they need to be competitive.
Frank Farrall: (38:31)
So it's not a sky is falling, batten down, cut all expenses, and just try to survive. It's much more around, let's be efficient. And to be honest, I think the whole pandemic situation and COVID, there were some inefficiencies that got built into a number of the businesses because we were all experimenting with ways of working around, okay, we're gonna highly virtualize in the space of just a few weeks. What does that look like? And what kind of organization shape do we need? And what do we do with data and cloud and virtual technology to enable that?
Shared data = valuable data
Satyen Sangani: (39:03)
You once wrote in a blog that you wrote on the Deloitte website that data gains value when we share it. That seems like a profound statement.
Frank Farrall: (39:11)
If you have data in a vault, in a database hidden away, not accessible, that's like having a product in a warehouse and not out on the shelf. Now, I'm being a little facetious around PHI and PII and sensitive data that you've gotta really be careful about how you manage. But if you have something that can lead to insights, or what we're seeing more and more, particularly with our financial services and our pharma clients, media would be this as well, it is like they've got some data and they can offer it as a business, as a service. And then with all the different cloud platforms and things, you can offer that through those different marketplaces. There are entirely new businesses that are being built up based on the data that organizations have about demographics or clinical trials or customer behavior and things like that.
Frank Farrall: (40:03)
If you've got that squirreled away in a database that you use, and maybe you use it for analytics and decision making inside your organization, you actually can put that out there, and again, in a safe and secure and appropriate way and monetize that. In a lot of cases, you can link that with other partners. You can create new businesses and models and things like that. So, to me, it's like if you get it out there, you can then unleash the value. If it's hidden away in a warehouse, then it's not gonna reach its full potential.
Satyen Sangani: (40:29)
Does that value accelerate as it's shared more?
Frank Farrall: (40:34)
I think it does because you can add to it. Also, if you get success with this product, more people are gonna want it, and you're probably gonna invest in it more, and you're gonna think about, okay, how do I augment it? And then what we've seen with a lot of data sharing situations is you get a two-way partnership going and there's actually a third... Maybe it's weather data. And so, if actually, if you brought weather data into this offering, boy, it would be even more impactful. So then you have got three parties that are selling that out as a channel and then it just rolls and rolls. And then once you start to realize that, again, coming back to the generative AI piece, there's about to be some launches of plugins and things where you've got things like weather and demographic and whatever, data and applications all being able to be brought together and then be delivered through an app. I think that's just gonna enhance the value of data. And the more it's shared, the more value it becomes, in my opinion.
Satyen Sangani: (41:28)
Yeah. There's so much nuance there, but the idea that there's sort of initial unlock and sort of progressive value changes the economics of how one thinks about data and maybe the framework for how you just think about data more generally. Before we close out the podcast, I wanted to ask one final question. You've got...
Frank Farrall: (41:43)
The (near) future of generative AI
Satyen Sangani: (41:44)
This perch within Deloitte where you've got exposure to so many customers, so many different data technologies, so many different AI technologies. What are you most excited about — and less trends than naming names of partners — frankly? And you don't have to say Alation obviously, but I know of course that would be at the top of your list. But what are you truly excited about over the next 3 to 5 years and why? And give me three things.
Frank Farrall: (42:09)
So we've talked a lot about generative AI, so I'm gonna go there and I'm excited but I'm also, “scared” is probably too strong a word, but nervous around, we have seen in the last six months the kind of window into what could be around autonomous coding and computers that can generate award-winning imagery and things. There were a lot of people that thought, “Well, AI will take out rote routine tasks. Knowledge workers are fine. Creatives absolutely have nothing to worry about.” Well, that's actually now been proven wrong, really, in the last kind of six months, or maybe a little bit longer than that. So what I do think is exciting is there is a lot of base-level work — coding, writing, and things like that — that will be able to be taken away. Now, what I hope then is it 10x's everybody, it frees up people to do their best work and takes out some of the rote mundane activities and things like that.
Frank Farrall: (43:05)
So I think that has really exciting potential for us from a future of work. And then I do think too, as the technology gets better and better and better and better, as it exceeds human knowledge and the like, being able to solve some of the challenges around disease, some of the questions we still have unanswered in physics, I think that's very exciting, and I think we're on the path to that. Now, anything that can be used for good can also be used for bad. And so I do worry about some of the negative consequences of that. I do worry about talking about prompt engineering. There will need to be a mindset and skills shift from individuals. And then also, in a lot of cases, in most industries, and consulting and professional services is absolutely like this, you learn your craft and your profession through doing the rote and mundane and frankly not very satisfying activities.
Frank Farrall: (43:54)
So I do worry, how do you learn the underlying basic concepts if you just jump right into prompt engineering and things and you don't learn some of the basics. I think there are some significant things that need to be worked out there. But on a whole, I think the potential for a positive outcome, particularly positive outcome very quickly — we are not gonna have to wait a decade for a lot of this stuff to become mature and really impactful — that to me is what is most exciting. And this is all gonna be fueled by data and the knowledge of models and how to bring this together and deliver this through technology. So that's what I'm looking forward to.
Satyen Sangani: (44:30)
Okay. Well, we'll take that one. Frank, this was a really fun conversation. It's been just great to be at your perspective. Thank you for joining us.
Frank Farrall: (44:36)
Great. Thanks for having me.
Satyen Sangani: (44:43)
Today's interview reminds me of a quote from Carl Jung. He said, "To ask the right question is already half the solution to a problem." And in the world of gen AI, asking the right questions is both an art and a science. For Frank, question engineering is just as crucial as knowing the answers themselves. Before search engines, we relied on memory and exploration to answer our questions, but search technology made instant answers totally accessible. Now, in the age of AI, we must learn how to ask better questions to get the best results. Our relationship with knowledge is ever-changing. Today, crafting artful questions is crucial. It's also exciting. What new doors might your queries unlock? Thank you for listening to this episode and thank you, Frank, for joining. I'm your host, Satyen Sangani, CEO of Alation. And Data Radicals, stay the course, keep learning and sharing. Until next time.
Producer 2: (45:35)
This podcast is brought to you by Alation. Subscribe to our Radicals Rundown newsletter. You'll get monthly updates on hot jobs worth exploring, news we're following, and books we love. Connect with past guests and the wider Data Radicals community. Go to alation.com/podcast and enter your email to join the list. We can't wait to connect.
Frameworks and the Art of Simplification
Advisor, Consultant, Thought Leader
Perfect is The Enemy of The Good with Ameen Kazerouni
CTO of Orangetheory Fitness
Asking the Right Questions
Strategy & Analytics Ecosystems and Alliances Leader, Deloitte
Building the Company You Wish You Could Buy From
CEO of Qlik
Start with Story, End with Data
Founder of Qubole and Creator of Apache Hive
From the Outskirts to the Center
EVP & Global Head of Data Analytics and AI, LTIMindtree
Get Out of the Building!
Co-founder, Sudden Compass
The Scientific Integrity Crisis
Dr. Elisabeth Bik
Microbiologist, science integrity consultant
Data Governance: Any “Dummy” Can Do It!
Dr. Jonathan Reichental
Author and Founder, Human Future
Humanizing AI: Authentic Storytelling
Jepson (Ben) Taylor
Chief AI Strategist, Dataiku
Your RFP is Useless
Duca Family Professor of Technology Management, UCSB
Host of Cautionary Tales and Author of The Undercover Economist
How Extreme Focus Launched the Modern Data Stack
George Fraser and Taylor Brown
DataOps & The Data Catalog
VP, Principal Analyst, Forrester
Turning Data Librarians Into Supercomputers
Senior Director of Learning & Communities, Alation
From Data Strategy to Execution
CDO, The Very Group
Multiple Sources of Truth:
Decentralization and the Data Mesh
Creator of the Data Mesh
Truth, Data, and FAIRness
Chief Data and Analytics Officer, Ingka Group
Using Data to Fight for Human Rights
Data Scientist, HRDAG
The Data Behind Dating
Author of Don’t Trust Your Gut
The Beginning of Business Intelligence
Founder, Business Objects & Managing Partner, Balderton Capital
Knowledge Graphs 101
Founder & CEO, Stardog
Radical Data Politics
CDO at Tableau
Data is a Weapon
Retired US Army General
The Data on the Chief of Data
Founder & CEO, NewVantage Partners
The Death and Rebirth of Data Privacy
Michelle Finneran Dennedy
#1 CPO and CEO of PrivacyCode
Data Quality Is a Risky Business
Co-Founder and CEO of BigEye
The Science Behind Quitting
Former Lead Science Writer at FiveThirtyEight
Your Path to CDO Superstardom
Co-Founder & Director, Carruthers and Jackson
Super Chickens & the Productivity Pecking Order
Dr. Margaret Heffernan
5-time CEO and Author of Willful Blindness
Attacking Data Literacy
Chief Data Strategy Officer at ThoughtSpot
Your Guide to Corporate Controversies
Co-Founder and Executive Editor, The Information
Making Big Data an Asset in Medicine
How to be a Data Cheerleader
Co-Author of The Chief Data Officer’s Playbook
Data Governance: It’s the Final Frontier
Principal at Teknion Data Solutions
The Secret to Data Storytelling
Cole Nussbaumer Knaflic
Founder & CEO, Storytelling with Data
The Reign of Specialists is Over
Bestselling Author of Range
Emerging from the Data Dark Ages
Principal Data Strategist at Snowflake
Forging a Culture of Data Governance
President & Principal, KIK Consulting
The Bazaar in the Cathedral
Co-founder and CTO of Databricks
Subscribe to the Data Radicals
Get the latest episodes delivered right to your inbox.