00;01;00;29 - 00;01;21;28
CHET
Today. We had Jon from Ohio on the podcast. He held many roles throughout his career and interests me. His first ever job involved building fences. His advice to leaders is that no task is beneath you. Strike. He held many positions throughout his career, and interestingly, his first ever job involved building fences. His advice to leaders is that no desk is beneath you.

00;01;21;29 - 00;01;41;09
CHET
John shared how his company, Akil, is bringing together predictive analytics and generative AI to serve customers. And he gave a couple of bold predictions for the future of Gemini, including how it'll influence the way we make decisions. You are going to learn a lot from this one. John, welcome to the inspired execution. Thank you for joining.

00;01;41;11 - 00;01;43;17
JON
Thanks for having me. Happy to be here.

00;01;43;20 - 00;01;53;03
CHET
Your career has been in tech, but your first job involved building fences. Right. What did you learn from that experience?

00;01;53;06 - 00;02;14;07
JON
Well, I mean, I grew up in Montana, right. And, you know, for whatever reason, my parents, they had horses. They were really into horses. You know, everyone thinks horses are great, but it turns out horses are a lot of work. And particularly in the winter in Montana, there are a lot of a lot of work. And, you know, you've got different pastures and you've constantly got to rotate them between them.

00;02;14;07 - 00;02;38;01
JON
And then, of course, like you have to section those pastures in different ways. So you're you're building fence by literally digging holes with with a purposeful digger in the ground and putting in putting in posts. And, you know, it's a it's a lot of really hard work. And, you know, I think that kind of sticks with you over time, you know?

00;02;38;04 - 00;03;08;06
JON
You know, I just sounds like I think it's important that you never think you're above any particular job in general. And you know that that really like sort of set most of how I approach things emotionally, even like from is like, you know, manual labor, like monotonous labor, like, you know, that all this stuff gets done and you need to be willing to do that and understand how all that stuff rolls up into bigger picture things in order to get something accomplished.

00;03;08;08 - 00;03;12;02
JON
So that's kind of that's why it's on my LinkedIn for whatever it's worth.

00;03;12;02 - 00;03;13;09
CHET
So it's formative.

00;03;13;09 - 00;03;14;06
JON
Experience.

00;03;14;08 - 00;03;18;17
CHET
Especially since it was very child labor and it probably had a great impact for you.

00;03;18;19 - 00;03;21;15
JON
Right. It's really to take advantage of that. I'll tell you.

00;03;21;17 - 00;03;25;20
CHET
Which is which is the way it should be, right? I mean, come on. Right. You know, it's so.

00;03;25;22 - 00;03;29;00
JON
But I'm not I'm not monetizing my kids enough somehow as.

00;03;29;02 - 00;03;44;18
CHET
Well. They really should start now. Yeah. You know, there's a difference when you grow up in a farm, right? I mean, it just, you know, it's not like, you know, this. There's not like, you know, you have help sitting around, right? It's not like you have a labor pool that says, Come on in. Right. Yeah, that's right.

00;03;44;18 - 00;03;51;18
JON
That's not like raising these city kids. I don't I'm worried they're not going to, like, be as big as hard as they should be, you know?

00;03;51;20 - 00;04;22;11
CHET
Yeah, it's it's funny. The one thing that you that you learn is that I say this. It is sex, which is, you know, you know, we we we watched, we wash dishes and we do windows, right? And we do anything that it takes to get to make the mission successful. Right. And so, yeah, including at times when you're tired and you just have to actually put your shoulder to the wheel and get it done, because that's what the mission requires and that starts very early.

00;04;22;11 - 00;04;42;29
CHET
That is not something that you can put in your, you know, in your mission statement or put it in your values. Right? It is values and behaviors. It is something that you either inherently as part of you, I call these shallow dishes. It's an part of you, or you learned it over a period of time. But it's not something that can be taught.

00;04;43;01 - 00;05;13;07
JON
Yeah, and I agree that things like culture can be stated, but it's really it's really what you do every day. And how you write. It's like it's not like what you say you do. And, and, you know, like we're a remote first company. So I find this type of attitude is even more important to train screen for. When you're hiring people or finding people you work with or build companies with in remote environments because you know they're there really is a need to be like self-starting and willing to engage.

00;05;13;10 - 00;05;41;10
JON
There's also, I will say some times where you can get like too focused on like doing a task like that and pull up and look at the bigger picture, you know, like, I'll give you a quick example. I did like every single support chat that happened in the company for like the first three and a half years and for good reason, because like, you know, I would learn when people were getting stuck in the product and what worked and what didn't and be able to sort of guide our direction more effectively.

00;05;41;10 - 00;05;55;24
JON
But, you know, it got to the point where it was like distracting me from all the other stuff I needed to do because I was so focused on doing right. So so there's like a there's a bit of a tradeoff, like a willingness to do whatever and delete and to show what the right way to do things is.

00;05;55;24 - 00;06;02;28
JON
But, you know, keep in mind, like you have to maximize your impact at the same time. So there's there's this there.

00;06;03;01 - 00;06;32;11
CHET
As as I as I say, attitude matters. Yeah right. Because in so you've had you've had a lot of different experiences This is on your last comment. Right. You, you you do dishes you get you know you do fences, you've had a wide range of roles, right? Engineering, product marketing as we were talking before we got started. What is the one thing through all the experiences that you can tie together that has really helped to be a founder?

00;06;32;14 - 00;06;58;04
JON
Yeah, it's a it's a good question. I think I've always had like a hunger for understanding, like why and how decisions were being made and like and trying to like, make sure we were always making the best decision for our collective success. So there would be like I basically break that into two key things. One has always been like really interested in winning collectively as a team.

00;06;58;07 - 00;07;23;23
JON
And then too, I've always been like really curious about like the how and why of an organization making decisions or driving direction. So, you know, when I was an engineer, I was always wondering why the product managers were deciding to go one direction or another. And and like many times I saw things that I just couldn't agree with happen, or I couldn't understand why those decisions would get to those points.

00;07;23;26 - 00;07;48;11
JON
And so so my curiosity about that, like sort of naturally drove me to taking on those roles and digging in and figuring out like what was going on. And and, you know, that that then drove me to the next thing and, you know, it just ended up like, you know, I, I didn't want to I honestly didn't really want to take over marketing when I, when I ended up in the seat, you know, like actually like in my family, my dad's a professor of marketing.

00;07;48;11 - 00;08;20;18
JON
My brother's a professor of marketing. I was the engineer and was I was like, as far separate from that as I could get. And then, you know, when I did get involved, I did it because, you know, I thought I could make a positive impact overall to the to the success of the team. And what I found when I got involved is just like anything, you know, there's a whole lot of process optimization and intelligent decision making that goes into it and and it's the same type of thinking you can bring to any kind of problem, as long as you're as long as you're trying to, like, drive for a good outcome.

00;08;20;18 - 00;08;37;06
JON
And then you understand the inputs and control degrees that you have. It was it was actually quite similar to the other things that I done, like product and engineering in some ways. And I will say upfront, I'm not a creative marketer, you know, I'm not I'm not the person who's going to come up with the brand campaign or something like that.

00;08;37;06 - 00;08;42;19
JON
But but a lot of marketing, particularly in B2B businesses, is is very, very much data driven.

00;08;42;22 - 00;09;04;22
CHET
So for sure, for sure, I would it seems like the one thing that continues to help you and that springs together through all your is this you have this I it this massive curiosity on how things work in a specific way. And I actually I actually referred to it in a different way, which is I'm, I continue to be a student.

00;09;04;24 - 00;09;27;03
CHET
Right. And I continue to think that I'm even though I've led companies for a while and led functions for a while, I feel like I'm just starting to get good at my craft, right. And and just being just just making sure you have the discipline of learning and getting really good at it. And that comes from curiosity because you're not copying and pasting what you learned yesterday.

00;09;27;05 - 00;09;48;18
JON
Yeah. And something about like, I guess maybe it's my personality. I think there's other people that are a lot like this. If I'm not learning, if I'm not faced with new challenges, then and I get kind of bored and jaded. So I always kind of seek it out and, and I will say, you know, doing your own startup is probably the best and fastest way that you can learn.

00;09;48;21 - 00;10;08;24
JON
Like, hands down. You know, I've got my MBA and, you know, I went through like and it was an entrepreneurship focused program option that I did. You know, I think my recommendation to people would be that if you if you have the ability to try actually building something yourself, there's no better way to learn.

00;10;08;27 - 00;10;15;25
CHET
That's that's awesome. So what's the elevator pitch for Akhil?

00;10;15;28 - 00;10;40;26
JON
Yeah. So you know, for for agencies and you know, these are the, the people that we help with our product who are looking at their advertising spend data or their audience data. We make it incredibly easy to slice and dice the data, to build reports that are alive and to build, you know, and to do it with natural language.

00;10;40;29 - 00;11;05;17
JON
So you can just ask for what you want to know and get an answer immediately. And then we allow you to build custom machine learning models to predict things like we're turning to advertising, spend our lifetime value, and use those to optimize your ad campaigns. So you know, you get better reporting like you get an analytics service with your clients and then you get to optimize all that by making better decisions on the data that you have.

00;11;05;20 - 00;11;30;08
CHET
In this in this paradigm and the approach you have, it seems like you're doing your your, your converging what people would have called predictive A.I., the stuff that happened. And if I can do it, you know, before, you know, before Chatbot came out to the world and people started thinking about how to use, you know, large, large language models.

00;11;30;11 - 00;11;47;24
CHET
And you're also bringing in General RBI. Can you talk a little bit about that? Because you've been you know, you've been doing the the decision science stuff for a while. How are you how is that going and how are you, you know, converging it with what I would call generative by techniques?

00;11;47;26 - 00;12;13;10
JON
Yeah, it's actually pretty fascinating because we started by building the AutoML Engine and and like back in the beginning, we had a lot of debates about do we call it machine learning? Do we call it. Yeah, I, yeah, I think AI largely has one in terms of like the public facing language and so that's for sure use. But you know, if you're going to be pedantic about it, like you're building machine learning models to predict outcomes and you're doing it a really easy to use way.

00;12;13;13 - 00;12;30;28
JON
You know, we can I come from Sonos in consumer tech and I'm a big believer in building user experiences that you don't need to get certified for or take or put a badge on your LinkedIn to use. You could just figure it out on your own. And so the first bit of our company was building this auto email engine.

00;12;31;00 - 00;12;48;14
JON
And then what we realized is like, okay, we have a really competent AutoML engine, but now we need a way to get data out of the systems that lives into the air and and the predictions out of the AutoML engine back in your systems. And so we built a ETL and reverse ETL system that connects on both ends of that and is pretty permissive.

00;12;48;14 - 00;13;14;04
JON
And then like after that, we realized and I'm getting to the generative API question, I promise that like just pulling the data in from your system, there was a lot of work you needed to do to transform or prepare that data before us. Ready to build? ML On top of and you know, we were basically staring down the problem of needing to build like every transformation function you could do in Excel or some sort of sequel equivalent.

00;13;14;06 - 00;13;38;03
JON
But, but you know, we needed it to be really easy to use. And unfortunately for us, you know, language models came along and got sophisticated enough to write that transform code. Yeah. And so even before B came out, we had the ability to take natural language to data table transformation and then build rapidly. And for that, like natural language to report or natural language to it.

00;13;38;06 - 00;13;56;13
JON
And of course all of these like then leverage the entire system. So if you ask a question to get a report, we build a data pipeline with the ETL system. So as that data updates, the report updates and all kind of just came together that way, you know, we we were at it for a long time before the market shifted to really be ready.

00;13;56;13 - 00;14;17;26
JON
I think we were actually kind of early. And, you know, they say timing is everything. But, you know, we're like we're sitting there like, of course everyone's going to be using ML in their marketing team at three years. And if and you know, like looking back, okay, like if you're an engineer in marketing, you might think that, yeah, but like actual adoption curves are much cleverer.

00;14;17;29 - 00;14;42;11
JON
Yeah, you know, the onset of like general awareness of how all this stuff matters has been pretty important. And so now I think a lot more people are thinking about how and where to use it. We use it almost strictly for actual data table manipulation or chart creation. So, you know, we don't have it do any sort of any sort of creative content or anything like that.

00;14;42;11 - 00;14;47;13
JON
It's all about working with data and extracting answers from your data.

00;14;47;15 - 00;15;05;01
CHET
Are you are you are you finding that marketers would prefer to have their interface becomes prompts? Are you you think that is as that happened or you think it's going to happen or it's not? That's not for your space.

00;15;05;03 - 00;15;35;21
JON
No, I think the prompter of an interface is going to extend beyond just marketing. Actually, I think I think generative AI is going to change user experience across almost every massive. And I think, you know, there's such a big learning curve to understand how a user interface in a product delivers value for everybody. This has always been how it is since since we all started using computers, you know, all the way back to like to dos, you know, like command line interactions.

00;15;35;23 - 00;15;53;20
JON
And it's all been like either visual abstractions or like, like language abstractions to try and let you do a task. And now, now those abstractions, you know, you don't you don't have to learn something in order to get what you want, but you can just ask for it. So we're already threading in language models in all parts of the user experience.

00;15;53;23 - 00;16;12;05
JON
I think everyone's going to do it and I think great. And just the ease of use and simplicity and speed means that it's not just going to be marketers or people who have like learning barrier access to some type of before, you know, even the people who like are the data scientists or in our team are now using the natural language because faster and easier.

00;16;12;08 - 00;16;29;17
JON
Yeah, there's a lot of work that goes into making sure the interpretation is done correctly. Right. And that the that you understand how your prompter is interpreted. Right. So yeah, so it's not easy but, but once you get that right, it's faster and easier than any other way of interacting I think.

00;16;29;19 - 00;16;45;10
CHET
Yeah. It's so if I, I have thought this is this is awesome by the way it really well said if I think about 2024 in the beginning of the year we said this is going to be the year of production, the AI and I've been very clear, right? So my statement has been.

00;16;45;13 - 00;16;45;28
JON
That.

00;16;46;00 - 00;17;13;20
CHET
Every company will do incremental use cases on generally by this year, every company that, you know, they may be like, you know, a couple of coal mining companies that may not do it, but even they will do something great. The question is when do the transformative use cases actually show up? Right? Is that a is that a you know, the snbs will start doing it by the end of the year and then the larger companies do it in 25 or you think it might happen sooner.

00;17;13;20 - 00;17;22;07
CHET
Right. And transformative defined John, as something that affects their PNL, not just I'm serving my customers a little better.

00;17;22;09 - 00;17;45;24
JON
Yeah, I have a maybe interesting take on this. And, you know, like some of the ways we've built back here reflect this tech before the, the big hype around like beat and and you know, some of these like interaction language models many AI projects were done top down because they were looking for that sort of big transformative change in the business.

00;17;45;27 - 00;18;05;25
JON
And they ended up being like really slow. They often got stuck in like data hygiene space where like guy, you're trying to pull information together from ten different sources and clean it up and then use it to make decisions. You're trying to get thousands of people on board with a new direction or way of doing things. And those things are complex.

00;18;05;25 - 00;18;30;22
JON
They're not easy, right? Yeah, I think I've always kind of been of, of the opinion that that the change is going to come from a thousand bite sized wins instead of like big overarching programs that you do. And so and I think we're seeing that, which is like everybody is going to get some percentage more efficient at their job by being able to leverage these tools.

00;18;30;25 - 00;18;49;14
JON
Practically whatever you do break, even if that's just correcting your grammar or helping your email better or whatever it is. And, and and that's going to hit the piano like directly, but it's going to hit it because it's going to change your efficiency across the board, in some cases more than others, and in some cases faster than others on it.

00;18;49;20 - 00;19;16;28
JON
But but it's going that way for everyone. And so as leaders of organizations in the space, you know, we've kind of thought about what necessary primitives of being an a company in the space. And it's like if you're going to build AI for people, you have to use AI to build out. So so you monitor everyone like is aware of the tools that there are certainly is trying new tools and making sure that they're driving efficiencies by a large, large store.

00;19;16;28 - 00;19;28;13
JON
In short, I think it's a it's a your wins are going to come from a thousand small applications. You'll have your big wins. But but I think broadly speaking, it's a rising tide that will drive efficiency to everyone.

00;19;28;16 - 00;19;45;04
CHET
I tell A that's a super interesting way to think about it. So switch gears, talking a little bit about the future based on what we just talked about, do you see a future where AI surpasses a subset of decisions that humans make today?

00;19;45;06 - 00;20;11;13
JON
AI surpasses it. It's an interesting question in the language. I see a world where I definitely see a world where it surfaces. You know, we already see this, where it surfaces, the information necessary to make a decision that you maybe didn't find before or didn't have access to or couldn't figure out on your own because it was too early to look at that.

00;20;11;20 - 00;20;43;17
JON
That's a very near-term future thing. Is there a time where you let it then make the best decision given those constraints? That's a little harder because a lot of times the best decision to make once you have access to the best information, is kind of a business specific decision. And it depends on the way I think about AI and machine learning is really it's a pattern recognition engine and it's core and you know, it either duplicates a pattern for you or shows you what the pattern is.

00;20;43;18 - 00;20;44;28
JON
You can understand what the performance.

00;20;45;04 - 00;20;45;27
CHET
Correctly.

00;20;46;00 - 00;21;12;04
JON
Once you understand those patterns. The decision making, though, is figuring out how to take advantage of those patterns to best, you know, bring your business success or accomplished objective, whatever that is, that takes, you know, internal data, internal knowledge, competitive knowledge. It's very rare that all of that information is encoded in these core models. So I think there's still like a it's more like a partner with a with a decision maker than a decision maker.

00;21;12;04 - 00;21;15;29
JON
But yeah, if it had all the information, maybe.

00;21;16;02 - 00;21;41;28
CHET
And maybe there a different way to state it would be will will I be able to make better recommendations based on the amount of data it can process and, you know, based on and do that consistently across many, many domains and make sharper recommendation patterns that human beings would write because they're not proven to an absolute. And the answer to that is like, Yeah, all right, yeah.

00;21;41;28 - 00;21;52;28
CHET
And then and then decision making, obviously if they if it does their recommendation 600 times, it's probably going to get the decision. Right. Right. This is the Yeah.

00;21;52;28 - 00;22;14;15
JON
I mean there's just that last thought loop which is like keep in mind that it only has access to the information it's looking at. And if the Yes, if the environment that you're operating in is changing rapidly, you know that that might not actually predict the future. It might just, you know, be reflective of the past. And and so like, yeah, you know, a big thing happened with machine learning models just covered, right?

00;22;14;18 - 00;22;34;21
JON
You know, you have these ML models predicting a bunch of things when when a systemic change happens like that, those models no longer were valid and they had to be retrained on new information as as the emerging environment changed. And one of the reasons are AutoML Engine has automatic retraining is because of that because you know, your data environment evolves.

00;22;34;24 - 00;22;54;15
JON
But I don't think there's any question that it's going to make decision making better and faster. And in many cases, it'll make recommendations that, you know, you may not have even seen or been able to service. It'll be optimum. It's just, I think, important to keep a think loop on it, to look at it and make sure you don't know something about the environment that it doesn't.

00;22;54;17 - 00;23;09;18
CHET
Yeah, no pressure. That's that's really well said. I feel like we could be talking about this for another, another 2 hours. So I'll ask you one one more question before we move to Rapid Fire. One bold prediction for generative AI in the next two years.

00;23;09;21 - 00;23;11;24
JON
A bold prediction.

00;23;11;26 - 00;23;28;03
CHET
Interesting know. I love the I love the point you made earlier, which is, you know, hey, listen, it's not going to be one big thing that changes the panel. It's going to be a bunch of really small things and that will affect the DNA. I love that. This is yeah, I think you need to look at it.

00;23;28;05 - 00;23;48;06
JON
I think it's going to find its way into our daily work in a in a much more substantial way. And I don't know that that's a bold prediction, but I suspect everyone's about everyone. I think on average over three years could get about twice as efficient at doing whatever job they do if they can start to leverage the tools.

00;23;48;06 - 00;23;51;12
JON
So let's call that like a2x efficiency gain.

00;23;51;15 - 00;23;53;25
CHET
In three years. Using that is bold.

00;23;53;28 - 00;24;06;21
JON
You know, in a lot of spaces, maybe not all, but a lot. You know, I don't I don't know that it makes my dentist more efficient. But like, I think a lot of jobs, a lot of technology jobs are right.

00;24;06;24 - 00;24;18;20
CHET
Now for short for sure. This is John. This is awesome. All right. Moving to Rapid firing. So quick questions, quick answers. What's your favorite way to use Jenny in your in your daily life?

00;24;18;23 - 00;24;30;02
JON
It's just exploring data for me, but that's what we do. So like being able to like take our take our CRM and ask a question about like our audience and get an answer right away. That's how I like to use it.

00;24;30;04 - 00;24;35;17
CHET
What's the problem in Maddie's pacing that you would have? Jenny I solve now.

00;24;35;19 - 00;24;54;26
JON
I efficiency mismatches, which is kind of a strange answer, but I think a lot of resources are underutilized or under allocated, and the standard of living could collectively be improved if we if we acted with more efficiency, you know, even at even just like food allocation, like two populations and stuff like that.

00;24;54;28 - 00;24;59;05
CHET
That's awesome. What's one common misconception people have about AI?

00;24;59;11 - 00;25;14;19
JON
I think a lot of people think it's like Sentient right now, and I don't really agree with that. I think it's like AI regurgitating patterns right now. That's not to say that it will eventually get there, but you know, people, people have this like idea that it's more than it is. It's, it's a tool right now.

00;25;14;21 - 00;25;32;10
CHET
Yeah. No, for sure. One word you would use to describe great tech leaders, one word inspirational, great one words to describe how you feel about the future of IQ.

00;25;32;12 - 00;25;36;15
JON
I excited.

00;25;36;17 - 00;25;56;13
CHET
I was it was great to see that smile, you know, with the word excited, right? Yeah. My point it is, you know, as you know as you would you would agree building startups is hard. If you're not excited about it, you shouldn't try it, right, because it's there. But as I tell people, there are far better ways to make money, right, than just go off and, you know, do a startup, right?

00;25;56;13 - 00;26;00;18
CHET
Because you put your blood, sweat and your tears into making it happen.

00;26;00;21 - 00;26;02;18
JON
So yeah.

00;26;02;21 - 00;26;30;02
CHET
So John, this has been awesome. I truly, truly had a blast and I love your insights there. It definitely comes from someone who's actually been living it right and has a perspective on it and obviously thinking about how it's affecting the folks that work at your company as well as yourself. So I think our I think our listeners are going to have a blast and we really, really appreciate you joining us.

00;26;30;04 - 00;26;32;00
JON
Well, it's been fun. Thanks for having me.

Inspired Execution

Jon Reilly

Founder & Co-CEO

Akkio's Jon Reilly on Partnering with AI for Better Decision-Making

Episode Transcript