Ian Mullane - KeepMe.AI & The State of AI in Fitness

Everybody, welcome to the Future of Fitness, a top-rated fitness industry podcast for over four years and running. I'm your host, Eric Malzone, and I have the absolute pleasure of talking to entrepreneurs, executives, thought leaders, and cutting-edge technology experts within the extremely fast-paced industries of fitness, wellness, and health sciences. Please stop by futurefitness.co to subscribe and get our interviews with summaries delivered straight to your inbox. You'll also find our free industry report on artificial intelligence. Five industry experts, five different opinions, and tons of valuable insights for free at futurefitness.co. Thanks for listening and on to the show. Back for its fifth year, the Connected Health and Fitness Summit is returning February 7th and 8th in 2024 to explore the industry's most lucrative opportunities and hottest trends. Come learn how legendary players and innovative brands are implementing preventative health and longevity-focused initiatives into their businesses. Discover how to successfully tap into Gen Z consumers, meeting demands for holistic and wellness-centered fitness offerings. Stay one step ahead of your competitors and learn how to take your brand to the next level by leveraging artificial intelligence into your framework. Secure your next investment by keeping up to date on where investors are prioritizing their spending and realizing white space opportunities. Build relationships and share ideas at the Women in Connected Fitness and CEO Founders Forum. The Connected Health and Fitness Summit is bringing together the strongest brands in fitness and beyond for two days of quality networking and curated content to get you ahead of the game. Go to Connected Health and Fitness.com to download the agenda and find out more. Use code FOF10 for an exclusive 10% discount on your event ticket. That's connectedhealth and fitness.com, code FOF10, and do not wait. Seats are very limited. All right, we are live. Ian Mulane, welcome back to the Future of Fitness. It's been a minute.

It certainly has, sir. Thanks very much for having me back. It's great to be here.

Yeah, I was uh you and I were just trying to sort it out. I think it's been about three years, right? Since uh it was either pre-pandemic or right at the beginning of the pandemic. We we uh we had a great conversation and um you know you've done uh an excellent job of the white papers that you release and the education that you're putting out and into the industry and trying to keep people um you know informed uh and adapting to what AI is doing is is is really critical. And you know, when we last did our interview, whenever that was three, four years ago, AI was just kind of this mythical thing, right? It was like no one really knew, like, ah, whatever, that's gonna be you know 20 2040. And now it's like uh in Q3 here, 2023, AI is just almost become ubiquitous in in everything. It's like, well, if you're not playing with it, then you're not playing the game. So it's it's been fascinating. Let's just start with that because I want to set the table, just give you an update, like give us an update over the last three years if you can, and like what's been going on with you uh and keep me uh AI.

Um, well, you you the the last three years have been eventful, I think it's fair to say, for a whole bunch of reasons. Yeah, the business itself was founded round about starting in 2019, and 2020 obviously came along, and um, one would suggest that that would have always been a difficult period of time, but for us it it actually wasn't. I think that we found ourselves at a good juncture where we were able to support the customers that we did have and were able to continue growing over that particular period of time. So if you remember our premise, our premise was that within each operator there is a set of data which we always believe to be predictive and to be valuable in the context of both member attention and in member and sales. So, how can we utilize this to be able to generate better conversions and more sales and how can we retain more members? So ultimately, from a financial perspective, increase financial performance. Um we've been, I would suggest, successful over the last few years. Um, we have enjoyed considerable success with north of 50 customers in 13 countries now. We add three or four customers on a monthly basis as a whole, and we've refined the offering. So, in as well as that specific offering where we also try to drive value is with quick deployment, meaning that we are not um involving our customers with lengthy implementation periods. Generally, it's done within a four or five-week period, and we've worked extensively on making sure that it has a low ongoing resource cost, meaning that it doesn't have the necessity to have dedicated resources that are just trying to work it. So now I would comfortably say that we are on a path to generating a strong customer base around the world and one that we've been eternally grateful to for providing us a lot of the input, which is now shaping the product development in the way that it has been going.

Yeah, gosh, you I mean you found yourself right in the middle of it. I'm sure you saw it coming, right?

Uh with AI and um I'd rather be lucky than good, Eric, to be honest.

Yeah, yeah, yeah, right, right, right, right. You know, maybe for the sake of the conversation today, because I want to get into a couple things. You know, obviously the updates that you're doing uh with Keep Me, and you you gave me a nice little demo right before this. And it's really impressive. Uh the white paper, um, your most recent one, I was reasoning over it this morning. Uh data and AI, the ultimate executive briefing, and people can find that on your website at keepme.ai. Uh also, you know, let's let's start with this. There's a couple definitions that I think uh would help a lot. Um three of them in particular machine learning, uh, artificial intelligence, large language models, right? Those get thrown around a lot. Can you, and I'm not looking for like a you know, Webster's dictionary uh um definition, but just to you, like, how do you define those three things and what's the differences between them? Because I think that'll help set the table for the things we're gonna talk about today.

I I think that uh probably the the best way of looking at it is to say that machine learning and large language models of generative AI are branches of ultimately artificial intelligence. Okay, so um in in our context, we've always utilized machine learning predominantly what we would call prediction or classification problems. Like what will happen? Will this happen? Um will that member churn or will they not churn? Will they buy or will they not buy? Machine learning is an excellent tool to be able to deliver that level of prediction. Um, the reason being, or how, if you if you'd like to understand, is that you would take a data set and that data set within it would have all of the outcomes. So did they stay or did they go as an example in a membership database? You would take 80% of that data, the historical, and you would train the machine learning model on that data. And it uses decision trees, right? Um, it uses, you remember, yes, no, moving down different aspects all the way down to the actual solution. On a we use a methodology called random forest, and as technical as it may sound, it will very clear be why it's called random forest shortly. The random forest will look at the data set and it will put every single permutation of a variable together to see if it had any impact on the ultimate outcome on that side. So for 43 data items, 42 data items, which is what you would expect to see in a standard club management system, it will generate 8.8 million decision trees just for that one member that will get us to the outcome. Now, it does it randomly and it does 8.8 million trees, therefore that's why it's called random forest. There's nothing complicated in that particular aspect of it, but how accurate is it? Individually or statistically, you're not going to find that any one of the outcomes in itself is going to drive the the accuracy of the model. But when you consolidate them together, so it's called an ensemble learning method, you are going to get anything between 92 to 95%. That we have customers that you've been on the website that are reporting a 98% accuracy. Now, that's not will I leave tomorrow. That's I joined in January, will I be here in the following December? Right? That's a very strong, and that's that that is the definitive here. Because for an operator, that then gives me this incredibly long runway to be able to change the outcome. But more importantly, it allows you as an operator an understanding of what the outcome's likely to be, even before I do. If I've joined you in January and I'm being flagged up on a platform as leaving in December, and you come to me in February and say, We think you're gonna leave. I've just joined, right? There's no way I'm leaving. You know, what are you thinking about? I've just paid my membership dues, but in reality, the signal is already there in the noise, and the platform's able to do it. So that that's machine learning. We use that for a whole host of classification problems within keeping. Large language models. Okay, so thank you, ChatGPT. Um, I I the analogy I've been using a lot recently is that um it's a bit like when I was using the internet back in the early 90s on the trading floor. There weren't many of us using it, we were all viewed as pretty geeky, and the reality was that to be able to extract value from the World Wide Web was very, very challenging because there was uh you know you need to know what you're doing. And then Netscape came along, and suddenly we had this thing called a graphical user interface thrown on the front end called a web browser, and the next minute everybody could extract value from the internet. AI is having its web browser moment at the moment in the form of Chat GPT, right? Because what they've done is they've been able to put an interface, in this case just natural language, on the front end of an incredibly sophisticated AI model and allow anybody, regardless of their technical capabilities, to have the ability to drive that value up. So, what's a large language model? A large language model is effectively a super intelligent parrot, but it's gone beyond the capability of just being able to parody to what we've said. It's now got to the stage that it can have a conversation with us as if it was a human. And it's able to do this because of the incredibly large quantity of data, in this case text, that it's been trained on. And that'd be the internet, that's books, that's articles, and a whole host of stuff. Which means it's now got the capability to form conversations, generate conversations, and engage on topics of any nature whatsoever. It doesn't have um, it doesn't have any conscious, it doesn't have any feelings about what it says. It's not a human, but it has the capacity to engage, and because it uses natural language, by definition, it's an interface which we can all engage with, and that's why we're seeing some of the success that it's been having.

God, it just sounds so futuristic, but it's not because it's here.

And that's uh it's uh you use it and I use it every single day.

I use it every day. Um I play with it all the time. In fact, I woke up last night at 4 a.m. Umly because I was thinking about an invoice dispute I had with our home builder, but then that dovetailed, as you know, as things do at three or four in the morning into like AI. And I'm just sitting there thinking about AI. I'm like, God, what about air traffic controllers? Like, is that job done? Like, what a you know, when you look at these complexities that we have that, you know, humans like we gotta sleep, right? We have stress, we have families that stress us out, right? We have all kinds of things that happen um in our lives that affect the way it performs, but AI doesn't do that, it doesn't need to sleep. It doesn't need it can get handle like that to me was like the perfect, and I was like, oh my gosh, write this down. This is like the perfect AI application. Is like all these human mistakes that could possibly happen, we can now take away and just put in the hands of AI, and you know, everyone can be way more efficient. And I think that's you know the kind of stuff that just blows my mind at four in the morning.

I have to say, the the I'm a pilot and the the uh ATC thing, I had not actually given any thought to that one at all. I mean, I'm probably a bit lazy in that regard that um, you know, I I've got uh as we've had the chat before, I I've got uh uh children, uh young adults now going through education, and um I I would not be encouraging law to any of them in any shape or form. I would be struggling even to look at accountancy to any straight or form. Um I I think medical definitely is going to benefit rather than be impacted uh by any great degree because I still feel at this stage there's going to be a need for people to still open up the individual to actually perform. But I think the diagnosis stuff is starting to get done. And yeah, I I won't share too much detail, but I know of an individual recently who has had um just through their interactions with a large language model, has been able to turn around the outcome for their treatment just by being able to engage with some variances with their practitioner, who quite frankly was closed off to every single possible outcome. And all this individual did was when they got their blood back, they pushed the entire blood work into the large language model and said, What would be the outlier possibilities for my diagnosis and potential treatment routines? And just being able to understand that better and then go and engage with their practitioner who probably thought they'd hired somebody on the outside, has given them a different outcome, and I'm delighted to say a successful outcome as well. So lots of exciting stuff going on in this area.

Yeah, yeah. And uh allow me to get us back on track here. I dove us off a little bit. But the um in your white paper, you mentioned, you know, data is gold. Uh, and I believe it. I guess the big question is, and this will probably, you know, spurn to many other points of of conversation, but how do we take the data? How do we make it gold? How do we actually take what we already have, right, and turn it into more revenue for gyms? I mean, that that seems to be the at the surface layer at this point where we're at, that's where people want to are wanna focus, right? It's like, well, how do I turn this into more revenue?

Yeah, I think the uh I I have in my limited years, if you say the last four or five years, I have seen a considerable change in the capability of people to engage with data, first and foremost. So initially, one of my reservations about our sector as a whole was that there was very little currency placed in some of the core applications having APIs that would allow data to flow out and into other tools. And my concern, and I sit on the UK Digital Futures group with um uh UK Active, and my voice concern there was an inhibitor to digital transformation was that the tools that were coming into the marketplace, whether they be for our sector or for the more wider marketplace, were going to require data in a structured format. And there were no there was not too many of the club management software providers at that time that were giving those capabilities. I'm pleased to say there's been a change in that, and uh now I think that it's probably a primary thing where when people are acquiring a new club management software platform, the first thing they're looking at is do they have an API? And that's going to be critical. Um, the world of large language models is going to, and I've I've uh you know, you and I have just taken a look at how that will change. Because what has been required in the past is to be able to take this data in into a structured format and then to be able to predict how the user would like to use that data so that you can show it to them in output. With large language models, what's going to be possible, as I've said, I've just shown you, is the capacity to take in this data and then to run natural language questions over the top of it. So rather than going to a dashboard and seeing, you know, which is great, by the way, there's nothing wrong with dashboards, but looking at a dashboard which is fixed with the metrics that we chose there and then to believe to be the north star of the business, we can go to these dashboards and then we can start to understand what the underlying variables is. What so, for instance, if we're to have a look at what our net retention is, if we're to have a look at what our sales conversion is, phenomenal. But those metrics, they're one-dimensional. You can't really do a great deal with them. You had great retention last month, why? We had poor conversion last month, why? Well, why in the conversion? Because of our social media ads um didn't perform as well as they recently did. Why? Right? And that why is just a constant that can be delivered into a large language model and it can start delivering more and more. The way I'm looking at and the way we've built it in ours at the moment is rather than presenting visuals anymore, which tend to be one-dimensionalized, is we will place a narrative underneath. And the narrative is probably providing a level of insight that even if it was possible for an operator, an individual, data analyst to find it, it would take so much work for them to do so. And because we're logical creatures and because we're very much pattern recognition creatures, we're look, we we've got, in many cases, we've got biases which are reflected when we try to review data and understand how we can perform better because we believe it must be, and then we go into the data to find it. In the world of a large language model, those types of biases are eradicated because large language models, just like machine learning algorithms, they don't care about your expertise, they don't care you've got 20 years' experience as an operator or that you're the best in class. All they care about is what it can see in the data, and then it offers up its opinion for you then to make decisions. So, how do I think it will directly impact? There is no operator in the world now who shouldn't comfortably be able to predict within a nine-month window which of the members will not be with them in nine months' time. If they then choose not to do anything with that, that's down to them. But if I'm an operator and I believe that I can within with that for that type of uh runway, I would expect that I've got the capability to at least change the outcomes for quite a few of those, particularly nowadays with the you know, the automation tools and the rest of it that are available as well.

Yeah, it's it's interesting. What you were showing me, Ian, right before this, um made me think like, you know, my experience as an operator too, it's like, and this was a while ago. Um dating myself, I don't need to do it. So it would just say it was a while ago. The uh the the platforms that I use as far as CRMs from um you know Zen Planner to Mind Body to Um Wattify to all of these, right? I went back and forth all the time because I never really found what I wanted. But the the the tough part was like I never could get the dashboard or the data and the analytics that I personally wanted. Um I have uh you know a good colleague of mine, uh he's been on the show, Dan Waimura, he's the CEO of PushPress, and you know, he talks about um you know having a really solid North Star metric. And you know, I'll I'll let him tell what his is, but I'm like, gosh, you know, it's it is really um important to have you know one or two metrics that are the north, you know, if you if you're following this correctly, and I think his is like uh you know check-ins per month per client, right? Like that's what he recommends for gym. Because if you do that right, if you can get that to a certain percentage, your gym will be wildly successful. So it's like taking all of the, like you mentioned, a narrative, right? Like taking the data, building a narrative that's workable. Um, how are you guys approaching that problem? Because I obviously you have a very slick solution that you that you have right now and probably working on more.

Well, I think the I mean, you know, Dan's completely right. I mean, that you know, you you need to take a look at some of those core metrics and things, for instance, in a studio context, even and then in a fitness operators context, attendance never moves away from being a core variable because if they're not attending, they're not engaged, and if they're not engaged, they're not gonna renew it, right? So, you know, not none of that is is open for question. I think where AI starts to drive in is that it uh A will see prior to that non-attendance, because there will be aspects in there, for instance, it'll see that because you came in from a particular lead source and because you're a certain age, a particular gender, and because your location is a certain place, and because your average attendance has dropped over the previous two weeks compared to the previous two weeks, you know, I can talk logically to try and explain the point. But the reality is that when a machine learning algorithm is looking at it, it's looking at every single data item and constantly putting them together to see what it is that's actually pushing it through. I think that's where there is a significant value. If I am engaging with members when they have stopped coming, the probability of me re-engaging them, I don't even I haven't even tried because I've never had to do it, because I've always used this as a method, um, is probably very low because I've stopped and you're reminding me that I'm not doing it. Some come back, there's no doubt about it. Yeah, definitely you can come back on those things. But what you want to be able to do is catch that as early as possible. And you may only be catching it two weeks before non attendance starts, but it's probably the most crucial two weeks in the next nine months of the membership on that particular side of things. The other aspect is that I think all of these tools are only as valuable as the action that is taken with them as well. So, as I've said, I can flag up to you who's not going to be there in the next nine months. Time and I can give you plenty of runway, but you've got to do something about it. But then the question, and this is where the AI comes into it again, is what do you do? Right? And then how do you do that on a personalized basis? Right? Because, you know, we we predict these guys are not going to be attending in four weeks' time. Well, let's just send them out exactly the same vanilla email to say, hey, why don't you come and join us for this class or that class or whatever? It needs to be better than that. You need to make sure that you are communicating around a value proposition that they're likely to want to engage in in a format that they may want to, whether that be in video, an email, whether that be in text, whether it be on a WhatsApp or something else along those lines. It needs to have the gender definition within it as well, where appropriate in that particular aspect of it. So, you know, don't go and send me a picture of a buff 23-year-old male in, you know, taking into account my age, right? It's not gonna it's not going to be the appropriates on that side. When you marry, give you an example of how we employ. Let's say, for instance, you uh want to do a PT campaign. First thing the machine learning can do is it'll tell you who will come to you for that PT campaign. So it'll tell me that Eric will buy PT if he's given an offer. Okay, so I pitch Eric on a 10-pack. Eric gets that 10-pack, he's not done PT before and he's thinking that's a commitment I'm not prepared to do. So Eric passes. He goes to Ian and they pitch Ian on a one pack. Well, Ian was going to buy it all day long. If you pitched him a 10-pack, he was gonna take it. So the way we utilize this here is that not only does it identify the people with the highest probability of the purchase, but it also assigns the probability of how much they will purchase. So that if Eric is my says, get served a one, right? He will purchase, but serve him a 10 and he won't, but serving a one, he'll purchase it, but on the 10. Then, having done that, you then use the automation to serve the correct offer with the correct visual at the correct time, with the correct method. That in itself is content optimization. If you do that, it's very difficult to see who will not perform better than we want to do a new PT campaign. Let's throw out the PT offers for the next 12 months and see who wants to take it.

Yeah. Wow, that's powerful. It really is. I mean, just the ability, it's not just okay, who's ready for PT, but how much PT can they handle, or you know, subsequent side programs or you know, supplements, or whatever it may be, who's ready for you know, uh a recovery pack of cryo, whatever. I mean, there's there's a million applications for it, I'm sure. The uh the data sources, I mean, that's a really big thing. I mean, I I've heard a lot of people talk about AI, and um obviously I'm I'm fairly fascinated with it. You know, the um crap in, crap out, right? Like you get bad data, you're gonna get bad results. Um so let's just, you know, as as mundane as it may sound to talk about data because it's everywhere about us and everyone else. Um what are the what are the primary data sources that we have now? And obviously there's whatever whatever's in the CRM, right? There's whatever's publicly known, like someone's, you know, location or whatever you can scrape the internet and find about people. But what are some of the big data inputs that are usable, right? Not just you know, um, sources of data, but things that we can actually apply within these models. Today I'm joined by Curtis Christofferson on my very first micro interview on the future of fitness. Kurt is the owner and founder of Workout and Innovative Fitness. He is a 20-year veteran in the wellness space, and he's got the success to prove it. He's also a long-term friend and colleague. Kurt just recently launched a newsletter called Healthy Ambitions to get back to the industry that's given him so much. His goal to help other entrepreneurs scale their wellness businesses just like he has. All right, Kurt. What is one actionable insight you can provide our listeners that has been incredibly valuable to you in your business career?

Wow, powerful question. And I could have so many answers to that one. But you know, I'd say that you know, the biggest element of scaling your business is hiring the right people. And without the right people, you can't scale. And, you know, one of my token rules is hire people that you admire. Hire who you admire. When you do that, chances are the people that you admire have some level of experience, education, or impact that you might not be able to provide. And so when you hire people that you admire and the value that they bring to your organization, inevitably they're gonna help you scale your business no matter how big or small it is. When you hire people that are unlike yourself that contribute value that you can't contribute, you're winning. And so my biggest rule, hire who you admire, whether it's the things that you admire about them around their education, experience, impact, or even how they lead their personal life. I think when you surround yourself with great people, you're winning the day.

Oh, that's awesome. Kurt, thank you so much for that. And if you guys want to learn more, please check out the Healthy Ambitions newsletter. You can go to Curtis Christofferson.com to subscribe and learn more about it and all the great content he's putting out there. Thank you, Kurt. Awesome. Thanks.

I think the the primary remains the club management system for a great deal of what I would call the fundamental structured data. So uh the payment data and the attendance data and the initial sign-up and the source data, that type of aspect of it as well. Um, that in itself can generate a great deal from a um from under, I mean, for instance, you know, if if keep me sales, for instance, is looking after the the sales aspect and then passing it into the membership before the member into the club management before it comes into keep me on a membership side of things, you are effectively enriching the data. So all of the stuff is captured up at the the sales side, what lead source did they come in, what campaign did they come in, how did they uh find us, did they do a trial, did they do a visit, all of that type of stuff, and then it goes to a certain goes into the club management system, and the club management system then enriches it around with data such as their attendance, their sign-ups for newsletters, those type of things, and then you know the the membership side of Kiki can take it uh uh a little bit further. Um, logically, other areas that can be utilized would be uh things like MPS platforms, they're useful, they're useful for uh um for for a data source. I'm not gonna comment on them relative to their capabilities around retention. I've always suggested that people who use MPS platforms are 10 times more likely to stay with you whether they give you a zero or a ten because they're still engaged. It's the ones that aren't giving you any MPS scores. I need to worry about, in my opinion. But there's that. Um, wearables is you know, I I I've said it in uh the future fitness paper, which I I wrote before the the data and AI one. Wearables are going to play an enormous part in the operators' fitness proposition going forward. Not because they are going to be tying themselves to a particular one, but because the data is is becoming more and more freed. So out of my Apple Health and my Google Health at the moment, I can record and hold so much more information, including stuff that's coming from my Whoop, coming from my Aura, coming from my Apple Watch, all of these data sources. What that's allowing is for the models to further personalize the content that's delivered. And when I say content that's been delivered, I'm not talking about um, do you want to buy more PT or we've got a new class that's coming in. I'm talking about customizing on a member-by-member basis club newsletters. I'm talking about posting automatically relevant content that, for instance, if I am identified as a vegan who's got a uh a current goal of trying to build muscle, right, then having particular programs which are pushed my direction, which meet that criteria, which are then in turn further flexed by the fact that seeing my health data, they can see how satisfied how satisfactory I am doing and whether there needs to be any modifications. Um, I showed you the content creation as well just beforehand. One of the biggest challenges operators have is not just around the content that they send out in the form of engagement emails, but also being able to put together relevant content for their particular type of audience. I mean, let's face it, you know, a CrossFit box is going to have a very different audience than a Planet Fitness to an Equinox to one of the one of the cycling studios. One of the challenges in there is how do you consistently push out content which is relevant that is going to keep people engaged, keep people within the community, and keep people understanding the importance of the product and the offering that you've got. And I think the large language models, again, are going to play a significant part in that because they will take away that fear of the blank page and allow content creators to start building much more extensive but also personalized content that can be utilized for education.

And that that's one of the things that uh it's it brings up a really good point that I ask the question all the time. And maybe you can look at it, you know, give us your insights into the future where else is going. Because wearable data is very powerful. I love that you have the whoop and the Apple Watch combo. Um good, good for you. Rocking, rocking the super combo. Um the uh right now I feel like we're we're we're doing the obvious right now with AI, right? Like we're working on retention, we're working on uh increased uh you know value per member. Um, all these things that are really important from operation standpoint. Like we, you know, first of all, we have to run successful businesses, then we can solve the world's problems. You know, if we're worried about revenue, then we can't do anything. Um but I'm most in I'm most interested in is like, well, how do we get better health outcomes for more people? And you touched on a lot of that as like, well, personalization is is huge, and that's where data comes in and helps personalize that fitness journey for people. Um so when you look at like, I guess I know it's a it's a broad question, but how are we actually gonna deliver better health outcomes for people who maybe um you know weren't atypically engaged in what we're doing as an industry?

I think I I am an unabashed provider to an operator rather than to the consumer. So keeping that as my direction within the business, what my role and my business's role to do is to keep members with the operator for longer. Right? Just by that simple task, I am improving health outcomes because of that. Now, it is not just about the simple approach of saying because we've identified they're not going to be with you for much longer, you need to rush out there and find a way to keep them. It is about equipping operators with the tools to be able to engage with the wide variety of people that come to them. If I use a large language model and I am a PT and I want to understand what are the variables I need to consider when I'm dealing with a vegan practitioner that's looking to go into a bodybuilding competition, for instance, or if I am looking at an ex-athlete who's looking to come back in from rehabilitation and is choosing to spend the time in the weights room when I've got a good understanding of whether it be a reduction in shoulder movement or whether it be that there's an ACL injury. I've got the capacity to use these large language models, which we integrate within Keene, to start to understand both the premise and possible solutions for discussion. I'm not suggesting they replace ever, but I have yet to meet a PT or a person that takes responsibility for people's health outcomes that has the breadth of knowledge necessary to deal with the society that walks in through their door in all its very many flavours on a daily basis. And I think that what this is going to be able to do is to provide a I mean, I do it myself on a regular basis. I use the Kick Me Tool, um, whereby um I've got my own aims. I've got and and I I'm telling it some of the most esoteric details. It probably doesn't need to know, but it will be what equipment's available to me, what my actual aims are, whether I've got any struggles with a knee injury, whether I'm coming off a post-hike weekend, for instance, on that side, right? Um, and what my current schedule is relative to my diary in the week, whether that's going to involve any jet travel, whether that's got any jet lag, and what I know I take things too far. I wear a CGM as well. But the um I will also have a look at that, but I will even have a look at what is the probability that I'm keeping my hydration on point during that week, taking into account what my schedule has. And the differences that it generates. Now, I go to my PT and he gives me my program, right? But I am utilizing his program as one data item into my overall outcomes, which I'm looking for on that particular site. Why does any of this matter? Which you can probably hear from my voice, it's got me incredibly engaged in my own outcomes because I've got more control over them because I'm finding myself easier to educate about what I should do and the impact of my actions. And I think we all, any of us who've worn an aura or a whoop, have now realized that alcohol plus sleep does not go. Right? That's you know, one particular aspect, it's it's been a disappointing part to life, but it has been a reality. But there is so much more that's starting to come out in this particular. So, you know, for instance, you know, I'm now at a stage where I can I've been able to move by 15% over a 12-month period my HRV. And I know you're a big practitioner in that particular area yourself. HRV is impossible to move on most cases. It's a really, really difficult metric, particularly when you're my age, to be able to push that up. But all of my outcomes are now at a level where when I engage with a health practitioner or a gym practitioner, I've got a better understanding, but also I'm finding that the ones that I'm working with, they've got a better understanding of how to engage with me. What I'm trying to do with Keep Me is to put in hand those type of tools, particularly in the generative AI side of things, to allow those uh providers, those service providers, to be able to engage with their audience on a different level than maybe they had before, where it's not just generic, they're able to personalize it, they'll be able to, you could quite simply say in the keyni context, these are the 10 groups. I would like to have an individual blog post done for each one of those, and I want to automatically send out to them based on the criteria which we have from the club management system. And that then means that not only are we posting 10 blog posts within a short period of time, but the relevance to each part of that particular member base based on age, based on gender, based on outcomes, based on diet, based on practices, based on likes, is all taken in place. That, so I'm told from the literature suggests how you build a community, it's how you build loyalty. And in my hope, right, that when you combine that with the rest of the kidney tools, if I can keep people in the gym for longer, if I can help the operators to have more people build fitness and health as a habit, then the outcomes are going to improve. And all of those other aspects, whether it be type 2 diabetes, obesity, or whatever else, will start to move down.

Yeah, yeah, well said. And it's it's been really interesting too. I I've noticed as wearables become more and more mainstay. Um, and like I'm taking this month off drinking, which is really funny because everyone's like, I'll go out and I'll like, no, no, iced tea, I'm good. And people look at me like, oh, he must have a problem. Why isn't he drinking? Right, like it must have gone too far. I'm like, no, no, I'm just trying to see how it improves my sleep. And unfortunately, it does improve my sleep and my HRV. And I say that unfortunately because I like to have you know a bourbon, right? I like it. Um makes me happy. But it does have the effects, and that's your bummer. So, but a lot of people now are focusing on their sleep. They're just aware, right? The awareness around like, okay, what affects my sleep? How many hours of sleep should I get? Am I getting deep sleep? Like all these things. I've noticed some really powerful trends, and people are worried about like, well, is it too much data? Is it gonna make people paranoid? And I'm like, come on, you know, that that's that's an individual person's uh you know reaction to things. Like if you're gonna get paranoid about you know, your your data, um, you know, like, okay, I know I'm tired, I don't need a wearable to tell me I'm tired, right? And stressed out. Be like, well, okay, relax. Like, you know, if you already know, then it shouldn't be a surprise. Anyway, I digress. I think that's it's really powerful about what you're saying. And I think um, you know, putting more personalization, getting people more engaged through data and their own personal data into the process of becoming healthier is is really important. And that's that's what's you know, it's it's these aha moments that people are having. Because let's be honest, we all love to think of ourselves that we're unique, right? We're all snowflakes, we all have our own journeys, we all have our own stresses, and the data, you know, through wearables and all these other sources will tell us exactly that. It's like, no, you you can, like you mentioned, if I hike 30 miles on a weekend, um, I go to my trainer on Monday, that data and information and insights is extremely useful, right? Because that's like, okay, let's slow it down today. We're gonna do some recovery work, we're gonna do some mobility, we're gonna do some things to get you back on your feet so you have a great week. Um, so it's it's really exciting time. You know, one of the things I want to look at when you kind of zoom out um into our industry, um, who do you think is using AI really well right now? And we can you know be specifically to operators, like who do you think is really leveraging it?

Um well, uh logically I'm gonna point to any of my customers because for obvious reasons they are from that perspective. Um I if I was to pick one off the top of my head, one which I I feel um utilizes it to great effect, it would be uh an operator out in the UAE called Well Fit. Um and they employ it across a wide uh uh side of their mostly on the operational side, mostly around the retention. But the retention is not the North Star. Engagement is the North Star that they're actually working towards there. So how can they make sure that um they I mean they increase the I I forget the actual stat you find on the keyme side, but they inc they increase the average attendance by three or four X over the particular period, their engagement by 10 to 12 percent, right? So it was a considerable uh element in there. I think if you if you look around um at the AI practitioners, I think it's more important that it is the ones that are utilizing automation rather than AI. And let me tell you why. Because the again, the the the AI aspect has always been phenomenal at giving you the insights, but the challenge is what do you do with the insight? And the insights themselves can't be delivered in at scale generally unless you're using some type of automation. I have found it easier to get adoption of AI within an operator than uh automation because automation somehow triggers off this fear of job loss, it it seems to figure off this um concern that um that the importance of the individual will not take place, and that's an area which I try to contradict quite a bit. Um, someone like a word, an average customer on Kimi will have between 38 and 47 automations in play on their platform, some of them have gotten north of 100. That means that every hour of every day, between 38 and 47 things that should happen are actually happening, right? And there's a big difference there. I sat, particularly in the early days when we were building out the tool, I sat in many a room and I went through some very elaborate member journeys. And the CEOs who may, and I know one of them does listen to your podcast, will smile when he hears this, where it was drawn up in this beautiful schematic on the board, and it had as many, it had as many outcomes as you could possibly imagine. If they don't do this, then it will go to this, and if they do this, it'll go to this, and if it doesn't go to this, it'll go to that. And he very proudly laid it out to me and he said, That is why I feel we've got it covered. And there was a snigger in the room, and he turned around, he said, Well, who sniggered? And he went, Boss, this has never happened ever in our history. We wrote this down, but do you have any idea how many platforms and people are involved in delivering that? He said, I'm not even sure. The first stage gets delivered more than 50% on an annual basis, right? Automation is where that aspect is taking over. And what automation is doing, it's giving certainty that core actions are taking place. Am I engaging with people at the right time? Am I congratulating them from hitting a milestone? Am I welcome them to the club? Am I reminding them to download the app, right? That type of aspect. It's making sure that when I generate content, it's going out to the correct people at the right time as well. Right? So don't start telling the people that are on racquetball when the new Pilates schedule's coming out. That type of stuff. Again, automation that is done on that side. Of things. That's valuable. But it's also then meaning that the individuals that are on the marketing teams are able to engage in the much higher value aspect of building the user personas, their member personas, understanding and finding new catchment areas, looking at how they can do that activation, or as importantly, individuals that are on the floor being able to engage in conversations on a basis rather than having to worry about actual task management. So I think we'll see, we will see an increased use of AI across the sector. I very much hope for obvious reasons. But I think that it what I am personally excited about is when Keep Me Creator comes out in the next week. Um I'm really keen to see what people end up utilizing that for. I've got a very, you know, at the moment it's going to be about uh blog content, sales and member engagement content and those type of things. But I'm going to be very eager to see how people start to utilize this to be able to put out personalized content at scale across their membership phase.

Yeah, it's really exciting. And that's the funniest thing about releasing tools like this, right? Is you get to see in the field how people are actually utilizing it and how they maybe they're doing things that you didn't expect. Um so that's that's uh that's gonna be the fun.

And it will always be that, you know, Eric, that that is the thing. It's you know, the the unintended purpose tends to end up being the only purpose that matters. So if I ever look at, and anyone who's in the industry will laugh at this, but if I ever look at any of the successful aspects of development that's gone out, it's generally been a variation of the initial intended use case, which has caught fire and then moved forward. If I ever want to look at any of the white elephants that have been designed that didn't work, that I designed in most cases, it was because it was done from a position of not looking at it from that particular perspective. But yeah, I'm I'm truly excited to see how that will adapt. We you know, we put a lot of time in putting in a big variation here, which is basically that an operator's brand voice has learned before it starts to work on it, but the model actually learns from its website and its blogs and all of those type of things. It's gonna be great for content providers to finally be able to say, throw it at me, right? Give me the ICPs across all of our members, and I am now going to start generating content for each individual sector in that particular thing, whether it be their, whether it be by their age and agenda, whether it be by their interest, whether it be their outcome, whether it be the actual part of the business which they're engaged in, I am now going to be able to generate proper personalized member content for each and every one of them.

Ah, wonderful. Um, last question for you, because I know we're starting to butt up on time, and it's your Friday evening, which I I uh I appreciate and respect. Um thank you. Do you have any legitimate, and we'll keep this in the scope of the fitness help of industry, but do you have any legitimate fears about AI, what it could do to damage or shift the industry in a negative way?

Not the industry. I mean, I do overall, obviously, I think you know the the capability in the in the hands of a bad actor is logically going to be concerning to some degree. I think primarily because um I think primarily my bigger fear is in the areas around deepfake and uh that that that type of thing, but particularly particularly when I look at the United States and I see how partisan that is these days, it takes such a little fire to stake something off there. And when I look at the operators within some of the you know particular camps around the world, the capability for a bad actor to be able to utilize the momentum of a particular political practice with a deep fake would be overworry from an industry perspective. Um, no, because uh I I think uh um as a sector, the sector is not particularly fast at adoption of new technology as it comes through. So I think that it will probably have a period of lapse. It'll be a little bit behind on on aspects of it, so I don't. Um so I yeah, overall, I I'm I'm positive. I think that um what we are going to see over the next 10, no, not even 10 years, over next, if you'd have told me, we've been working with large language models since June 22 now, right? Before Chat GPT came out. If you'd have told me then what we're doing now, what we're doing now, not what the world is doing, what we're doing now, I would have said 2025.

Right?

And I got that wrong, right? If so, and and when I look at what we're looking at doing in eight weeks, 12 weeks time with the other stuff I've shown you again, I wouldn't have thought that was even possible this time last year. If it was, it was a long, long, long way off. So I think that the biggest thing that we're going to notice is the rapid speed that some of this stuff is going to come through. And one of the biggest changes which is going to take place is that these things are mobile phones, which are these incredibly powerful computers, they are still designed with a necessity to do two interfaces, right? The interface that the computer needs and the interface the human needs. This is going to change quite dramatically because when we move to a large language model element of it, there's going to be no necessity for all of these blocks, there's going to be no necessity for all of these individual compartments, which we call programs and apps. We're going to move to one area where we are effectively dealing with a processing instrument that's got the capability of taking care of any of these types of instructions in natural language, where by definition it can knit apps together, right? It'll be able to bring things together where I could ask a question and say, you know, what type of run should I be doing today? And it looks at my Apple Health data, it looks at my straw there, and it looks at my track app, and it then decides where I am and what the optimal time would be for me to go out and run this particular trail at this particular pace. And by the way, Ian, make sure that you've got one liter of hydration in because you've been below average for the last two days, right? That's where the difference is going to be. How long is that stuff going to be um uh away from? Um, you know, I I think that we're already seeing a radical overhaul in user interfaces already. We're doing it at Keep Me, so I I know that to be the case. Um, my expectation is that we will not recognize the mobile phone in its present form beyond its physical size and its screen dimensions within 18 months. 18 months would be my my uh that aspect of it. And that's not within that's not with an insight of of where it's been, it's just that I could see how it could happen.

Oh man, crazy times, Ian.

Like this is just yeah, but not good, good times.

Yeah, I think so. It's not boring. I'm not bored, that's for sure. No, and uh, you know, whatever it is, it's happening. So uh, you know, get on board. Yeah, exactly. Um well Ian, thank you so much for for joining me. It's it's been uh it's always really uh great to catch up with you and get your insights. You know, AI is something that um I can't talk enough about. Uh I just think it's it's so uh it's so seemingly so sudden, it's so impactful, it's gonna change the way we do everything. And it's uh, you know, I I I just don't think you can talk about it enough. And I'm that guy at dinner parties that makes things weird. And um, I'm cool with that, right? So um yeah, real pleasure. Uh Ian, if people want to get a hold of you, uh you want to send them somewhere online, where would you like them to go?

Yeah, well, keep me.ai, I think is uh has got all of the white papers and a whole bunch of resources, including the the previous podcast that I've had the pleasure to do if you could self as well. Um, and then me personally, I'm always interested in people's views on both the direction they'd like to see products like keep me take and also their views and concerns over AI in general. I'm happy to engage with everybody on that subject. And I very simply can be found at Ian at keepme.ai.

Awesome. Thank you so much for joining me. Ladies and gentlemen, Ian Millane.

Thanks, Eric.

Hey, wait, don't leave yet. This is your host, Eric Malzone, and I hope you enjoyed this episode of Future of Finis. If you did, I'm gonna ask you to do three simple things. It takes under five minutes and it goes such a long way. We really appreciate it. Number one, please subscribe to our show wherever you listen to it. iTunes, Spotify, Castbox, whatever it may be. Number two, please leave us a favorable review. Number three, share. Put it on social media, talk about it to your friends, send it in a text message, whatever it may be. Please share this episode because we put a lot of work into it. We want to make sure that as many people are getting value out of it as possible. Lastly, if you'd like to learn more, get in touch with me, simply go to the futureoffitness.co. You can subscribe to our newsletter there, or you can simply get in touch with me, as I love to hear from our listeners. So thank you so much. This is Eric Malzone, and this is the future of fitness. Have a great day.