Mastering Timeless AI Strategy: A Deep Dive into Data Readiness and Industry Transformation with Chris, William, Andrew and Ana

Mark Smith: 0:30
Welcome to the Power Platform Show. Thanks for joining me today. I hope today's guest inspires and educates you on the possibilities of the Microsoft Power Platform. Now let's get on with the show. You're back with the Ecosystems Podcast Today. Like we attempted last time, we're actually going to talk about a white paper that's to be released, or has been released, by Andrew Andrew, why don't you just kick off today with telling us about this white paper? How did it come about and who's the target audience and what are you trying to address with it?

Andrew Welch: 1:08
Yeah, yeah, thanks, mark. So the white paper is Crafting your Future Ready Enterprise AI Strategy and it is a Microsoft and Cloud Lighthouse white paper that came out on the 16th, I think, of January. Organizations are right now going and starting to think about how am I going to roll out and infuse artificial intelligence throughout the organization. Many are doing so right now in a really, I think, fundamentally unstrategic way. Right, it's like let's go turn something on, let's go build an AI-infused workload to the loudest problem that we heard.

Andrew Welch: 1:56
Right, we're not taking a strategic approach to it and I think, really importantly, many organizations are not doing or maybe unprepared to do, some of the hard work around their data platform, whether it's consolidation of data into data services that are addressable by AI, or whether it's some of that data readiness, work around data hygiene, governance, indexing those sorts of things that are going to be necessary to make these really powerful AI workloads possible.

Andrew Welch: 2:26
And I started thinking about this a while back when the partner at a firm that is my client said oh, I want to do this amazing thing and it should be really low hanging fruit, right, because it's AI, and I think he just assumed that this was going to be easy. So it occurred to me that organizations have a lot of work to do to begin to do this in a strategic way and to kind of correct some of the particularly data mistakes that they've made by kicking the can down the road over the last 30 years when it comes to actually having a clean and proper data platform. So that's what the paper is about. It sets forth five pillars upon which to build an enterprise AI strategy, and the whole idea is to help organizations become future ready for the AI that we understand today and the AI capabilities that are coming in the very near and increasingly rapid-paced future.

William Dorrington: 3:27
Well, that's a wrap for the podcast. It's been an absolute pleasure.

Ana Demeny: 3:33
I think Chris is like and it's a wrap, and we're done.

William Dorrington: 3:39
We managed to repeat a summary of what we said in the last podcast. We're done.

Chris Huntingford: 3:44
So I have a question, right, like when you say data readiness, all right, now I'm slightly kind of playing devil's advocate here, purely because number one, I feel like data is a different thing to everyone, right, and number two, like perspective around things like security is quite important as well, like. So, when you say data readiness, you've got things like data hygiene, data distribution, that type of thing. What do you mean by data?

Andrew Welch: 4:12
I intend in. So I think the intent here, right, is data in a pretty broad form. Right, it is the you know, and a lot of people think only about the data that's really sort of close to their own job, right? So for some it's structured data sitting inside of a database, right? So it could be your ERP or your CRM or a bespoke application that you've built. In some cases we're talking about documents, unstructured data, so the files, the photos, the videos, the documents, etc. And in a lot more cases we're actually talking about data that's sitting inside of a spreadsheet or an access database or some sort of third-party application that is in no way addressable by AI. Today, this stuff is just littered across most organizations, and the larger the organization, the more of a mess I think it is.

Chris Huntingford: 5:06
All right. So my next question is we're thinking around like structured and unstructured. What about things like phone call recordings and that type of thing? That's an important one, right? Because one of the things that I've started to discover about organizations is the moment we start talking about data, you're right to have broken it up into structured and unstructured, but, like there's, I feel like there's more to it than then from from a data landscape perspective. The example I'll give you is that insurance company is recording hundreds of phone calls. Okay, where does that fall? Is that unstructured transcriptions and things as well? Like I want to, I'm trying to figure out the multimodal thing Like?

Mark Smith: 5:43
is that getting? Add to that digital exhaust, right? What are the things that you do? Or the telemetry data that is formed out of a specific time it's done, a specific location it might be at, or all that other stuff, all the other data that's still been accumulated but wouldn't be seen that oh, I need to access that today by an individual in the organization, but could be very powerful at uncovering behaviors, patterns, opportunities.

Chris Huntingford: 6:12
That's the whole Microsoft Graph conversation as well. Like when you start plugging contextual data against structured and unstructured, you're starting to get like a really good view around what's going on. But I think, like with AI, can we look at that as from a full picture perspective?

William Dorrington: 6:27
And that is the beauty of it, isn't it? It's one of the benefits that large language, multi-modal, native multi-modal let's keep building that sentence up Large language models can really bring to bear, because you can now start absorbing. As long as you have an intelligent data platform that allows you to reach into these different facets of dispersed, siloed and segregated data, you can now start getting into that. But that is what I would hope you meant by data hygiene is bringing those data points together.

Ana Demeny: 6:54
But I think you can start really easily Like. I think you can start very simple Because, whilst we want to be sophisticated and talk about like, multi-month you know data with like with your inbound calls, conversations and transcripts and stuff like that the reality is that, in my opinion, even if we have like, if one contact looks the same everywhere in all the systems an organization has, then you're one step closer to be able to use AI Like. I feel like you don't need to be that sophisticated and in fact, I think that if you sell sophistication to that level to an organization, they're not going to buy it. They're going to be like okay, fine, that's our strategy in five years. Thank you.

Chris Huntingford: 7:46
On that point, right. So one of the things that I found and this is where and I think you know, when we did one of the previous podcasts, we spoke about readiness for AI Like one of the previous podcasts, we spoke about readiness for AI Like one of the things that I found which is important is that you get some organizations who are like and Andrew, this is where I liked what you, what you'd put around like data consolidation across the whole spectrum. You know there's. You've got to start somewhere. But when you start looking at the complex orgs especially law firms, insurance companies, those types it starts getting really interesting. They want it, they want the depth. So when we're going in those are the different levels, we're not doing anything. In the middle it's like, oh shit, we're here, or like we need we want to understand more.

Andrew Welch: 8:26
One of the things that I've observed over the last year of some of the early adopter organizations really start to put workloads into production that have an AI component.

Andrew Welch: 8:37
Is that, because most organizations have spent years and years and years not tending to their data estate, really kicking the can down the road and saying we'll let future us deal with the lack of investment we're putting into our data platform today, right, deal with the lack of investment we're putting into our data platform today, right. What I've seen a lot of is folks will build an AI-infused workload and they're putting it into production and they'll scurry around to get the data that they think this particular workload needs. Stick it in, often, stick it into its own data lake, right, like this, is the data lake for this workload right? And that's because they don't. There's no hope of going back, and you know, by the way, it could be lake, it could be blob storage, it could be wherever you're, whatever facility you're talking about. So right now we're in kind of a band-aid era where, I'm sorry, a plaster era for most, for for most of you well, I finally understand what you mean where folks are just doing just enough with their data in order to make the immediate workload function.

Andrew Welch: 9:53
And one of the things, one of the points that I make very early on in the paper, is that there are really, I think, two key principles behind your AI strategy. One is that your AI strategy needs to be able to absorb tomorrow what we don't understand fully today. Right, and relatedly, your AI strategy needs to, the best investments you can make in AI are fabric architecture, where you've got data in one lake and you're using that data to feed your AI workloads, but the organization is deriving benefit and return on investment from that consolidation in other ways as well, such as enterprise search or such as improved performance of analytical workloads. So those are the types of investments I think are really wise right now, because it's not just an investment in AI, but it's an investment in multiple other needs that most organizations have.

William Dorrington: 11:06
Absolutely. And one thing I just want to qualify, because we have different listeners and even ourselves, we have different thoughts because of our scope of perception and our industries we work in, and what I mean by that is we started talking about complexity. Now, complexity is very dependent on, first, your perspective, but also the industry you're in. So, actually, what Chris discussed around bringing in cool recordings and images and that is absolutely warranted for the legal industry, you know and high policy. And when you start looking at other aspects, such as high policy in central government, where you have people like you know that take care of the environments that need IoT aspects feeding in, you know, water level, quality, water quality itself, pollution that will need to come back in along with other stuff so they can enforce it's a complete necessity for their business.

William Dorrington: 11:54
But if we went and spoke about that level of data feed, that level of complexity to a charity client, nfp or even some commercial, yeah so go. Well, wow, that's that's, that's the future, that's universal way, and you're like back to them, that's the most basic principle of value you can provide them to make their lives easier. So I just wanted to cap off One person's complex is the other person's necessity, and it all depends on where you're working, where your client is in the industry you're in and the regulations you have to follow.

Ana Demeny: 12:22
But at the same time, the tooling right now doesn't require organizations to go into very complex data models or very complex realities at all. Like they can get quick wins straight away. All they need is a little bit of knowledge and open mindedness.

William Dorrington: 12:44
That's all they need, really all depends on their need, absolutely, oh, I agree.

Andrew Welch: 12:49
Well, so so there's a actually there in in the paper I make a distinction between what I call incremental AI and differential AI. In AI that is going to bring greater efficiency, speed, precision, accuracy, whatever. To a task that a human probably would have otherwise performed. Okay, so most of the co-pilots are in this business, right? I listened to another podcast recently that was talking about how doctors and nurses are now taking inbound requests from their patients and they're letting you know. Ai is analyzing those requests and then writing the response and putting that response in front of the practitioner to say, yes, this looks right, send it off and it's saving. Saving doctors and nurses hours and hours every day. Right, right, but that would have happened anyway.

Mark Smith: 13:54
Do you know who owns a big chunk of that IP for that doctor medical scenario?

Andrew Welch: 13:58
Who does?

Mark Smith: 13:59
Microsoft when they acquired Nuance. There's a product called DAX, and DAX is incorrect. What it does is it's been shown to shave about two hours a day off a medical practitioner's note-taking time Two hours a day, and the accuracy is like 99.x in what it can do. So it knows all the medical terms, it knows all the pharmaceutical drugs, it knows all the. It has that. So when you know how there's a joke in the medical industry or anybody that tries to read a medical person's notes're unreadable, it's eliminating all that. It's eliminating the error of. Oh no, that's not what my little squiggle meant. It meant that they can just dictate it now and it and the app just works on any mobile device. They've got feeds into the cloud behind it, into the server, and so, yeah, you want acquisitionance. Acquisition. That little DAX product is another massive bit of IP Microsoft has in the space.

Andrew Welch: 14:59
That's interesting and I wasn't familiar with that particular component as part of Nuance.

Andrew Welch: 15:06
But you know, I was listening to and we'll get back to differential AI in a minute but I was listening to this podcast.

Andrew Welch: 15:13
Like I said, it's a New York Times podcast and they were talking about AI a year on and this was around the birthday of ChatGPT, and they almost sounded as if they were lamenting that, oh, we have this brilliant thing, this thing that was supposed to be brilliant AI and all we're doing is saving some white collar people a couple of hours every day.

Andrew Welch: 15:38
And I was pretty floored by that level of pessimism because, as much as I think, when we think about AI, we want to jump first to the sort of amazing groundbreaking. We could never have done this in a million years if it were humans performing it. But the fact is is that across the world, around the world, most societies are leveling off from a productivity perspective. So if we can save an entire profession let alone the fact that, by the way, we have in most countries we have a shortage of medical professionals, so we can save an entire profession two hours a day, or even an hour a day in the aggregate, the productivity gains that that means for society are enormous, so I'm not nearly as pessimistic that, oh, all we're doing is saving some people some time. That's big.

Ana Demeny: 16:29
But it's life-changing for a lot of people. In fact, today I read an article that between 2020 and 2022, in the UK alone, there have been over 300 deaths in women who are pregnant. Most of those women had kids before, so no one actually thought that they were at risk, and even when they tried to, I don't want to say blame, but even when they excluded COVID from these deaths, from these maternal deaths, they still were left with like 8.5 in every 10,000 pregnant women in the UK alone. That's because, unfortunately, medical personnel is very limited, exactly like you said, andrew and they just don't have time to deal with any potential complication that may arise.

Ana Demeny: 17:31
So here's where AI could really help us. It could really like if we were to use it, even in rudimentary use cases. You know where we're saying oh, if a person ever suffered from like a high blood pressure, then let's put them at risk for thrombosis, for example, which is like a medical term for blood clots in pregnant women. Then all of a sudden you have like a short list of people who can be at risk and all of a sudden you can deal with less. So we don't really have to. When you're talking to incremental AI versus differential AI and everyone wants to do generative AI and they want to do the cool things. In reality, humanity can easily benefit from the most simple use cases of artificial intelligence with regards to data, and data doesn't even have to be complex or sophisticated or like super relational nothing like that and so can organizations, right.

Andrew Welch: 18:50
so, and this is where that differential versus incremental ai question I think is really important, because differential AI those are we could have called this moonshot AI or aspirational AI those are, I think, more of the things that we sort of dream about predicting deep minds. Alpha fold right, that has famously predicted the folding of proteins and I'm far from an expert in the science here right, but this particular AI has done something that humans could have labored for many, many years and still probably never really achieved, right. But I think that organizations need to understand that they don't need the flashiest or the sexiest or the most groundbreaking AI workloads in order to derive value from their investment in artificial intelligence. Organization, balancing the portfolio between incremental and more differential type artificial intelligence helps you manage risk that a single investment is not going to work out, because some AI investments won't work out either because the technology is ready, or you don't have the data to make it work, or it's not particularly applicable. So kind of balancing the risk of your investment across that spectrum of complexity is important.

William Dorrington: 20:17
So a few points on what we've discussed. I could not agree more and I think starting with the simple allows you to adopt the complex as well, because it allows to build confidence. So I think there's no mistake in the strategy of Copilot within Microsoft to show people how to leverage and get AI and think about leveraging AI. I think customer service co-pilot is one of the best co-pilots I've seen, along with a few others, because it's very practical, it improves the experience of the content and it improves the experience of the agent. It helps them both out. You know the web chat functionality, grounded on your knowledge articles and your various SharePoint lists, to then monitor the web chat while you're being asked questions and presenting the agent with that. I think this is the answer, kind of you can have that up and running within a day if you really wanted to go for it.

Ana Demeny: 21:04
right, and your knowledge articles are there right, and you will search the internet as well, so you don't even like absolutely.

William Dorrington: 21:11
and and the other thing I want to comment on uh, just because it is, it's a passion topic. So my, my background was medical genetics and the protein folding one is very impressive, especially if you've ever tried manually looking at something like that. But the other one that I use, case that I've seen and I think I may have commented this on a previous podcast, but as I'm here only one in 10, I can't remember and that's around cardio.

William Dorrington: 21:34
Yeah, exactly, there's always an excuse and they're all terrible. So cardio oncology is a really good example where AI is starting to come into play, which is with oncology obviously meaning cancer, the drugs and the medical practice towards cancer, which is, of course, quite a sensitive topic for most as well, and it affects most people. The medicine behind it is getting so good that actually it's not the cancer that kills you, it's the effects of the medicine, of prolonging your life with cancer that gets you, and that's only through cardiac events. So cardio-oncology is the study of the entwined between those two points, of both the heart and the cancer you're going with. And what they've realized is if they can start tracking episodes. So every time you go in and you get your blood pressure done, you get your vitals done, that gets put against your record and then you keep getting given the same drugs and the doctor may occasionally adjust those drugs if they feel it is appropriate, but a cardiac event will occur.

William Dorrington: 22:33
But they've noticed if they get loads of that data together. And it's not complex data, it's very simple data. It's very simple data. It's just, you know, andrew's coming in and he's had these checks. Will's coming in and he's had these checks. They're the same checks measuring the same things, but they can start tracking over time who's had adverse effects and as soon as you get enough data on that, you can start tracking the microchanges based on the medicine given. To adjust the medicine and it's recommended, then you need to change this patient's medicine to x, y, z based on these and these vitamins, and that's another really good example of something that isn't entirely complex but is will have huge benefit to humanity.

Ana Demeny: 23:10
It's the same in the study of diabetes or nutrition. You know, zoe. You know zoe, the they. But they do a lot of that and all they do is they gather data and then they apply some predictions. But you know, grant, I'm not that smart. I have no idea how to study genetics. I don't know how idea. I have no idea how to study nutrition or how things affect us as humans. That's just way too complex for me. But what I can do is like apply some of the low-code tooling that we've got right now. Organizations can do that straight away, even from reports Like today.

Ana Demeny: 23:51
I was talking to Andrew because I was actually watching another podcast. I think it was from Nick Dolman and Ulrike. Really really good, great shout out for you, mark, in that, by the way. Very good, how was it really? Yeah, yeah, yeah, awesome, awesome shout out. But they were talking about data responses in Fabric, where you can hook up some Power Automate to your Power BI reports, and I, incidentally, went to actually look into that a bit more. You can predict whether a certain record is going to be, you know, cold, warm and hot, and it's going to alert you when things are about to change. So even that it's a sort of prediction and artificial intelligence that we could use straight away and all of that is low code, like honestly, you don't have to know. I mean, it's great if you do know some genetics and you know actual machine learning and stuff like that, but if you don't, it doesn't mean that artificial intelligence is like oh, I will let 10 years from now me look into that.

Andrew Welch: 25:13
No, it's still right now. Absolutely that's a good string for us to pull. I think probably in another episode is how and I went into this a bit in the paper right is how the role of low code in scaling AI across an organization. It's not enough to just consolidate your data and build an AI workload right but you have to scale it and actually infuse it into the culture of the organization, and low code, I think, is going to be a huge part of that. That is a string we should absolutely pull.

William Dorrington: 25:44
And I would go one step further and connect an extra string to that once we've pulled that, which is, I think, going into the well no code, the querying a pool of data, of simple data simple data a complex data doesn't matter by using a research canvas such as project sophia that we've seen coming along to ability to state what you're trying to find out and then have that bring a ui actions and dashboards to you to then interrogate, receive insights that you do not need to be a specialist on to receive. I think that is also the future of where we'll start seeing intelligent applications and intelligent research and BI going, and it's already there.

Chris Huntingford: 26:25
Oh yeah, one thing I'll add quickly is that when you look at the state of what's going on with low code, anyone that's adopted low code in any organization is actually in a really good place to adopt AI, and low-code in Microsoft platform is the AI extensibility tool right. So anyone that is using low-code right now congratulations, you're an AI extensibility expert. Crack on.

Mark Smith: 26:52
Nice, we're drawing to a close on this session. We're drawing to a close on this session, and one thing I just wanted to highlight right up front in the start of that paper was future ready, not future proof, and I think it even goes back. Why it really resonates with me is the whole concept of ecosystems is around creating an environment to address the unpredictability without knowing what it's going to be before you get there. Right, creating a tool set and infrastructure and environment of where things can grow, whether they be ideas, whether they can be better ways of doing business. So I found that resonated. I think.

Mark Smith: 27:32
Another talking about threads that we need to consider pulling is then, across your data estate, addressing the role-based access, no matter where that data is sitting, or making sure that people are not touching stuff. And you know, a simple example I use for this is I don't want my colleagues to know, uh, my payroll data. Right, it's still organizational data, it's there, but I don't want my colleagues knowing it. Necessarily, you know, I don't want them to be shocked at how much it is that I'm getting paid compared to them, right, and so we still consider that organizational data, but I feel that there's a whole thing around. How do you make access to data no matter whether it's sitting in SAP system, an Excel spreadsheet, an access database, a Oracle ERP that the access is consistent across? It's not that just somebody remembered to configure that system right, but how do you get that consistency of rights to access, read, write, access, read, write, update, etc. In the platform?

Andrew Welch: 28:41
that's a huge area, I think, where actually the the data platform and the application architecture is not, is probably not sufficiently caught up to what we really need it to be right now. So I think, on the one hand, it's going to be a huge investment for technology vendors such as Microsoft or Amazon, google, et cetera. Going forward, I think that Microsoft is holding more of the cards that will make it very successful at solving that problem, so I think that definitely worth pulling on.

Mark Smith: 29:17
Excellent, and with that we'll get Chris to wrap the show for us.

William Dorrington: 29:24
Cool.

Chris Huntingford: 29:24
Well, thanks everyone for coming along. Andrew, I love the fact that you put this paper together. I think it's really, really interesting, and the five pillars are definitely something that I will be using in my day-to-day and talking to customers about. Yeah, absolutely, and I also enjoy the fact that you've made it understandable, which is very difficult because in this day and age, right now, I'm actually busy responding to an email trying to make it understandable, and I've used a couple of your points at the bottom, you know, saying, hey, go and read this. So, yeah, thank you very much, a. It's a very valuable piece of information and I'm excited to see it, or see everyone else, or let everyone else see it at least. And, um, yeah, thank you everyone for coming along. It's been epic awesome.

Mark Smith: 30:06
I'll add to that. If you want to be entertained while you read this document, use a screen reader with snoop dog. That's what I did. Snoop Dogg, he read it to me and, yeah, he had the good smoke. Hey, thanks for listening. I'm your host business application MVP Mark Smith, otherwise known as the NZ365 guy. If there's a guest you'd like to see on the show, please message me on LinkedIn. If you want to be a supporter of the show, please check out buymeacoffeecom forward slash nz365guy. Stay safe out there and shoot for the stars.