AI’s Enterprise Evolution: Enhancing Cloud Applications with Data-Driven Intelligence

Machine-generated from the episode audio and not hand-corrected, so names and technical terms may be imperfect. The audio is authoritative.

Welcome to the CTO Advisor podcast. I'm Alistair Cook, hosting the show for today. Today's episode is going to be looking at some AI added to Cloud. Who would imagine that AI would be in the Cloud? Was triggered by a post from my colleague, Keith Kirkpatrick here at the Futurum Group. Welcome, Keith. Could you let us know just a little bit about yourself and what it is you do with the Futurum Group? Sure. Alistair, thanks so much for having me here on the podcast.

I do appreciate it. I am a Research Director here at the Futurum Group. My area of focus really is enterprise applications. I take a look at it from, how's AI really impacting the development of applications across the tech stack? Everything from your large applications that everybody seems to touch like your CRM software like Salesforce, as well as the CDPs, which bring all of these data together from different parts of the organization. The line of business applications, which are the retail applications that you see when you go to check out at the supermarket, and they're punching in information there at the point of sale.

Really, what I try to do is just assess, how is AI being utilized in the software to really improve its functionality, and then assess, is it doing what they're promising? Just by way of background, I spent many years as a technology journalist going back into the mid to late 90s, covering PCs and things like that. I've since moved up the stack into enterprise apps from there. But again, thanks for having me on the podcast. Thanks, Keith. I think that moving up the stack is an important aspect of what we've been seeing.

It's a phrase we've used for probably the last 20 years, having myself come from IT infrastructure, hands-on building service. The other thing that was really strong in there was this whole idea that AI is a thread through things. I liked that you talk about AI as bringing value to the actual application that's out there, and the AI itself is not the thing. It's the thing that makes the thing better, to quote Bobby from Cloud Field Day. You particularly wrote a research note recently about Salesforce adding AI to their Consumer Cloud.

Is that Consumer Cloud? Consumer Goods Cloud. There it is. Retail focused application. The Consumer Goods Cloud is something that's been there for a while, and it's something that helps that point of sale retail piece. Can you run us through just a little bit about what that Consumer Goods Cloud is for and why AI is being added to it? Sure. If you think about the way various consumer goods are marketed and sold, increasingly, we've been moving from this mass retail approach where you and I go to the store and there's a pair of pants, and well, we look for our size, and that's traditionally how we bought things.

Now, with the advent of personalization, and really, this advent of all of this data, they're able to actually market to us based on who we are as people, in terms of what we like, what we don't like, and how did they do it? Because they have so much data that is captured on each one of us, from me going to a website and clicking on ads, or hopefully, going into a store and picking up something and being able to track that. What's going on here is they're trying to really add on to that.

In addition to capturing data from clicks about what I like and what I don't like, there's also just this wealth of data out there that are captured in databases that say, okay, here's Keith. He's a 50-year-old guy who lives in the United States. Typically, other people like him tend to like black jeans because it makes him feel like he's 10 to 15 years younger, and he likes ones that are cut more athletically because it makes him feel young and hip or whatever. They're able to take all of that aggregate data and then overlay it with the personalized data that they already have about us to help them craft more targeted campaigns to me.

So when I'm on a particular website, they can actually make sure that the right ads are delivered to me or that the right personalized campaigns. If I'm on Amazon, they go, well, he tends to buy things, and people like him tend to buy things in pairs of both blue and black so he can match all those outfits and keep his wife happy, all of that kind of stuff. So what this is doing is it's allowing these organizations to pull in all of this other sort of syndicated data, combine it with the personalized data they already have.

And then, of course, the other thing when we're getting into the whole AI thing, and particularly generative AI, is it allows these marketers to interact with that data in a very natural way instead of having to build, to go through clicking different boxes of different attributes. You can actually query that data in a very natural way, saying, hey, what does a 50-year-old guy who lives in New York, who wants to be cool, what are the kind of things that he likes? I mean, I'm oversimplifying it, but really that is the goal here, is to make it easy for organizations to interact with and access all of this data they have on all of us to create more personalized campaigns that will hopefully keep me coming back over and over again to buy those products.

And so this is that continuation of the more data you hold about your customer, the better you can work out what they want, the less meaningless ads you put in front of them and the higher return on that investment and getting content in front of your customers. Right. And you know, the other thing to point out here, of course, is that in the past, a lot of times marketers would rely on third-party cookies, meaning that they would track you from site to site to site.

And we're going to see that sort of go away because there's an industry-wide push to reduce that sort of tracking, which is sort of invasive, you know, for people knowing where I've been everywhere on the internet. And they're relying more on data, you know, first-party data about, you know, the interactions that you have while interacting directly with the brand. And that's where, you know, the ability to really dive into that, understand it, make sense of it, and as well as bringing in that syndicated data of sort of a persona of someone, you know, who's 50 or whatever and lives here and, you know, has an income of this and all of that, you know, being able to meld all that to really create a real persona and profile of a consumer, there's a lot of value in that.

I think drilling to one of the technical points in that, that shift away from using client-side cookies where it's a very fragmented collection of information, even the information about me will be on my laptop and my phone and my tablet potentially as separate pieces. That shift to centralization makes it so much easier to actually get value out of that data because you can look at trends across multiple customers far more easily if you centralize that information. What do customers or what do organizations that might want to actually use this?

What kind of steps are needed for onboarding for getting to the point where you can actually extract this kind of value? Well, there's a few things. One of the most important steps, and this is something that I think a lot of organizations overlook, is most organizations hold an incredible amount of data on their customers. The problem is a lot of times it's scattered around the organization and there is just no way to get at it because it's held in your individual point-of-sale application or it's held in your marketing database.

Unless you have some sort of a CDP, which is a customer data platform, Salesforce's Data Cloud essentially is one of those. Unless you have a system or an application to harmonize all of that data so there is one single source of truth about me or about you, linking all of those disparate data points together, it is very hard to really have a complete and accurate picture of who you are and what your interactions have been with the company over time. One of the big challenges with the third-party cookie thing also was that I might be online and I might stumble upon a site and inadvertently click on an ad or just casually click on something that really has nothing to do with what my real preferences are.

That's happened a lot. It also happens with these made-for-advertising sites that are designed to trick people into basically clicking through even when they don't want to just because it's a way to generate cheap clicks. So that was also a big problem. But I think to get back to your point, what organizations need to do is make sure all of the data is available, accessible in one place, and that is what Salesforce is doing with Data Cloud. Then in terms of using their consumer goods cloud offering with their Einstein AI, this is basically an application where they are making sure that they're able to pull all of their marketing activities together in one place and make sure that they're able to integrate all of that data and use that Einstein AI to query that data and really make sure that they're able to interact with it in a way that is just much more simple for, let's say, a marketing manager to devise a campaign as opposed to having to go through and write essentially what would have been a very, very difficult query to do, linking across all of those disparate systems before you had that sort of generative AI approach.

So the first stage is get all of your data in one place, and presumably there's often a whole piece of data cleansing and data preparation in there, which is a really tough part. And then the ability to use a large language model, a natural language querying to extract intelligence out of that vast amount of data, and that's where adding the AI features, the Einstein component into the consumer goods cloud. It's going to enable more business-oriented design rather than technology-oriented designs of how we built that application out.

That's a great point. If you think about what generative AI does, it really, and I know everyone in the organization hates the term democratizes because it's been one of those overused terms, but it really does allow folks, you know, who do not have sort of an engineering degree or a programming degree to get in and really interact with data in a way that is natural and that ensures that what they have here in terms of a, you know, marketing strategy or a sales strategy or a commerce strategy can actually be executed in the way that they actually envision it and making sure that they're grabbing the right data because that's the problem, one of the big problems, and I'm sure you know this from your background with application design, you know, there's a lot of time-energy cost that is lost sort of translating what the end business process demands and what is actually delivered in that whole back-and-forth process.

Yeah, and there's a whole heap of challenges around writing complex join SQL queries across this content that historically means you need a developer in order to build the application logic, but the developer is not typically a business person who understands the business process and business requirements, and this is absolutely that terrible word, democratization, but, you know, the accessibility of the advanced data analytics, essentially, from asking real human questions. If I want to make better sales in the Pacific Northwest, what sort of product should I have in the Pacific Northwest that don't actually suit a man living in his 50s living in Long Island?

Right. But that element would be much harder to ask of a SQL query than it is with a generative AI. Of course, that generative AI has to be hooked up to your corporate data. The foundation models, as Keith has been going through, Keith Townsend has been going through his 100 days of AI, those foundation models are a way of interacting with humans, but you need to tie them up to the corporate data, and again, that's what this consumer goods cloud enablement is doing, is you centralize that corporate data in one place, make it easier to use AI tools to query against it, and then you use the natural language AI tools to make this accessible to somebody who is more business-oriented, and hopefully that's going to shorten the time between a business requirement for good data or for a new campaign until that gets delivered, and having spent some time teaching DevOps engineering courses, that shortening of the latency between a business requirement and the delivery on that requirement is fundamental to delivering value, so I see a really strong parallel there.

You know, and the other thing, if you think about the way that business has become much more granular in terms of, like you said, let's talk about regional differences, if you have a merchandising department that's a person who is responsible for, let's say, the Western United States versus the Southeastern versus international, they're going to obviously all be competing. In the past, they would have been competing for the same resources, the same developer resources to accomplish very specific things. The list is, you know, this long and growing.

Now, with these types of tools, they're able to actually get access to the data and develop their own applications as they need to meet their specific requirements without sort of sitting in that long queue of, you know, other priorities. And that is, if you think about the speed of business today and the way that, you know, many organizations are taking the approach of let's respond to what's going on in near real time in terms of, you know, it could be any kind of current event, a weather event.

If there is something like, you know, obviously we have a presidential election coming up this year in the United States where things are moving fast and furiously and things change quite a bit. There are certainly organizations that will want to be able to respond to that very, very quickly. And tools like these enable them to do that in a very granular and very specific way to meet very specific personas and actually even individuals. So there'd be elements of the current weather. There's a cyclone that's currently hammering Texas and having the ability to respond to that and get stocking levels in the right places ahead of it, make sure that there's enough sheets of plywood to put over windows in a place that isn't necessarily hitting cyclones this time of year every year.

So having some of that intelligence as well. So feeding in not just from data that you have about your own customers, but also more of that environmental election, the progress of the election, the locations for different candidates' rallies would also be impactful for retail. There'd be a whole lot of real-time event driving into these decision-makings. And combine that with the trend towards much more personalized marketing. We actually want to make sure that the things that are being put in front of Keith are the things that are Keith's interest, not just everybody in their 50s who wants to appear younger.

Isn't that a ruin? No. But I think one of the other things to keep in mind with all of this, particularly around generative AI is though, and you alluded to it before, generative AI in and of itself, there's not a lot of value there. Where it really provides value beyond the novelty factor is when it is properly grounded within a very specific corporate of data and making sure that as the generative AI models run, that they are being monitored to make sure that the model isn't drifting, to make sure that there isn't bias being introduced there.

Because typically when we think about bias, it's the usual suspects with respect to protected classes. But what I'm talking about here is, if a corporation is using this for commercial purposes and they're not really watching what's going on in terms of the data that is being spit out or spit back to them based on these models, the model can drift. There can certainly be inaccuracies that may impact the way that the data is being returned to them if they're not being careful with it.

And that can actually have real consequences. At a macro scale, it's things like if the model starts to favor clothes that are much brighter colors than are fashionable in an area, then that can lead to overstocking of stores with a product that doesn't sell. That's exactly the kind of model drift impact for business. I also like that you came back to generative AI by itself isn't useful. It's great for making headlines of, here's the latest video we can produce, here's a recommendation to put glue on your pizza.

But none of these things are actually solving business problems directly. You've got to couple that generative AI into a business process, into real business data to actually get usable predictions out of it. Particularly as we start seeing the currency of data, the up-to-dateness of data being vital, the foundation models that are trained a year ago, two years ago, have no idea what's going on in the election. You have to bring additional data in to feed that and to update that information, that awareness.

And that timeliness of modern information is going to be crucial to these business uses, particularly in the marketing. You can't rely on what we knew two years ago to market to an individual. I know I've changed in the last two years, but my situation has changed and my personal preferences have changed. Marketing has to stay up-to-date with that data. And again, it's that timeliness and completeness and quality of the data that we're using to drive these AI models. Yeah, you know, the other thing you mentioned that I think, if we think about some of these foundation models, the sort of conventional wisdom is thinking about, you know, your very large open models.

That is not, you know, in my research, what I'm starting to see quite a bit more of is the use of, you know, more purpose-built or, you know, even small language models, whatever you want to call it, that are specifically built and trained around a specific industry or function, task, what have you, with the idea that, you know, you don't want to waste all of that compute power, you know, using a, you know, like open AI model, a Chet GPT-4 or whatever it is, to do something that is very specific to, let's say, you know, a mining operation or something like that.

I'm trying to think of something as far away from the, you know, traditional public model. But what, you know, I've had discussions with a number of software vendors that do focus in on very specific industries and are taking the approach that, you know, there is certainly value for large language models in very generic, basic tasks. But when we're starting to get into very specific industries and functions and things like that, it makes more sense to have a smaller model in terms of, you know, number of attributes and what have you, but that is more focused and is trained and then, of course, grounded in very specific data to make it a much more efficient way of using generative AI.

Because, as I'm sure you're, you know, do a lot of research and talking with people about, this stuff isn't free, no matter what people seem to think. You know, compute is not free. Yeah, I think costs to value, right? Businesses are very happy to spend money as long as they're going to get a return on that investment. They're very reluctant to spend money if there's no sign of return. I think using the right tool is always central to that. And as you say, a lot of the business use of AI, the actual business value doesn't come from the large language model.

The large language model is often a way of interacting with a human, but it's not the bit that does the business process. A lot of the AI that's done in business is more traditional pre-generative AI type models. And we definitely saw, it's essentially a collection of statistical regressions in a trench coat, as Justin Warren is inclined to say. It's an analysis of some specific data that doesn't require that large language model. But what we see is that combination though, where you want to democratize and make access easy for more business-oriented people rather than requiring the programmer, large language models are awesome for that.

And I do quite often come across cases where the large language model is used as the design piece, as the way of coming up with the design of the actual AI, the more conventional AI that is doing the business value. And so quite often when people are describing the use case of a generative AI, it's around designing something that's not going to use generative AI. It's writing the code for you. It's coming up with the parameters that might suit building a much smaller model.

So yeah, production tends not to be all driven by large language models. And I think the other thing that's really important to mention with all of this discussion with AI is that I think organizations are starting to realize the benefit of making sure that AI is not used in a vacuum, meaning not just using it for this one specific task here and this one over here. It's about connecting various processes together. So it's acting on the same data, and you can actually derive insights across the organization or throughout the organization, because really that's where you see greater efficiency.

You see real insights that are bigger picture in nature, as opposed to looking at a very sort of small segmented space, which really there might be a little bit of value if we're talking about the big picture issue, which is how are organizations going to justify their investment in all of this technology? It's going to be because the technology is being applied across as much of the organization as possible. So incremental 1%, 2% increase here, here, here, here, and here adds up to a much greater number.

And so a systematic view, a whole system view of how your business operates, how it interacts with customers, how it interacts with suppliers, that more holistic view. And again, it comes back to you need that centralized information about how all of this is working. You need to collect all of that in one place where you can work with it. I think that's a reasonably common story that we've had of wanting to have a central source of truth, and that there's this sort of innovation happens around the edges of that central source of truth, because it's easier to make changes in a smaller object.

And then that innovation gets pulled back into that central source. So I think there is a tension between wanting to rapidly innovate independently and that centralization, cohesive, holistic view. And so businesses should be aware that there will be a tension. There is no nirvana where this all just works perfectly together. There will be changes over time of how you approach this, and that different stakeholders within the organization are going to want more independence and then more centralization over time. But it's not a sign a check here and your business is perfected.

It will come as no surprise to a mature leader, but that's the situation. Right, exactly. And I think that's true with whether we're talking about Salesforce, whether we're talking about ServiceNow, whomever it might be. Obviously, their goal is to position this stuff as out of the box. And the reality is that is not usually the case. And it really does come down to having a well-thought-out strategy, you know, looking at your data, looking at what are the objectives, the business objectives that, you know, an organization might have, and then looking at realistically, you know, what is the projected time to value for this?

Because there is, you know, none of this stuff really works completely out of the box like that. That's just a fallacy. And I know that's, you know, the marketing departments are going to be angry I said it, but it's reality and organizations, you know, cannot think that just because everyone slapped a generative AI label on everything that suddenly everything works like magic. It still requires time, thoughtfulness, preparation, and, you know, quite honestly, a lot of hard work behind the scenes. I think the thing that will differentiate these tool sets that come in is just how much effort you have to put into doing the things that everybody else does, which you want to avoid, you want to minimize that, and how much of your time you can focus on doing the things that differentiate your organization that make your organization unique.

And that's where you should be able to spend the majority of your time. You should expect to spend time building out the thing that makes your business unique in these tools. Keith, we're heading to end of time. Any final comments? And also, where can people find more of your insights and engage with you as well? Sure. Well, you know, sir, I think, you know, final thought on this topic is, you know, there's a lot of really interesting stuff going on.

I don't think that we have really even scratched the surface of possibility. We're still in, I think, you know, to use a baseball analogy, we're probably still in the first inning when it comes to AI at this point. But, you know, certainly I expect to see, you know, real innovation continue ahead. In terms of where can you find me, well, I have a weekly podcast called Enterprising Insights, which you can find on any, whatever, wherever you get your podcasts. That comes out every week.

com, where I write regularly. So please feel free to check out all of that there. And, you know, hopefully we'll be seeing me with you again sometime in the future. Excellent. Well, thank you very much, Keith. And thank you for joining us for this episode of the CTO Advisor Podcast. We'll be back again probably next week with more interesting content around, obviously, probably AI and how it impacts the enterprise infrastructure.