Episode 163: From Guinness Price Pub Calls to Business Model Change: AI Agents in Practice with Matt Cortland

so today for calls question time i'm really excited to be speaking with matt courtland hi matt hey there how are you very good very good matt now i got really excited when i saw the press and the media about my two favorite subjects coming together guinness because i've got a place in Ireland and AI and you created the Guinness Index. Could you explain, those who haven't heard about it, what you did using AI for the Guinness Index? Yeah, so I noticed that especially in Ireland, the cost of a pint really varies very widely depending what street you're on and I mean what venue you're in and I decided to try and figure out what the cost of a pint actually was.

And I created an AI agent that was a voice agent that called every pub in Ireland that had a phone number, had a chat with the person who answered the phone and asked a simple question, can you tell me the cost of a pint of Guinness? And then with that data point, then kind of made an index called the Gindex, where that is logged on a map and people can add to it. And it's kind of like the initial data set and people can kind of adjust and add as the price changes. And yeah, it's kind of had some interesting insights around how pint pricing changes.

And I also married that data historically with what pint pricing had been tracked at for, you know, the last 20 years or so. Yeah, and I was fantastic. Yeah. And I was really excited when I saw that because, and so the stuff I've been doing at AI and that over now the last two or three years is seeing how ai could do things that would be too costly and too difficult and that to do for humans right you know getting humans to go into pubs and and sign up and on an app but then get put the price in yeah we could have done it or humans phoning up the pubs so so you had a voice agent that you trained and said as you said to to call up those pubs.

So give us a bit of the stats maybe about how many pubs they called, how many calls got answered. What was the reaction to people when they were speaking to an AI out of the blue? Yeah. So initially we started with Ireland and we called over 3,000 pubs.

About 2,000 of those pubs answered the phone or had a working phone. And we were able to get the price from about 1,000 of the pubs, right? So the kind of the stats whittle down as you go. Generally, people just, I think, just answer the question, right?

So we called just, you were to say, hey, do you have a table of two available? Are you serving a roast today, right? It's just kind of a simple question. So there, we just, it was relatively straightforward to get that response.

We, you know, we did the same thing as well then afterward in the UK, and we called almost 30,000 pubs, right, to give a sense of the scale. And we were able to achieve even better results with the second time around from a research standpoint. Right. Good.

Good. And what sort of reaction were you getting from people? Yeah, I could see that most of the time it was a simple thing. But how do you think people react when they get this call from, you know, were they able to tell?

And when they did tell, What sort of interaction did you have with those few that were interacting? Yeah. So from the data, the data suggested fewer than 4% of people realized they were talking to an AI agent and not a person, right? So that was one of the data points I was trying to collect in that.

And of those 4% that realized most people just still gave the answer, they're kind of bemused. A few people were annoyed, right? But for the most part, people were just kind of just okay with it. Yeah, yeah, yeah, yeah.

Yeah. And I guess in the work that I've been doing over the last couple of years, you know, I've seen these sort of use cases and we've demoed in the workshops we've been doing with the interaction with voice, with the interaction with video and sort of stuff like that. And, you know, going back two years ago, certainly it was quite glitchy. Now I think it's got a lot better.

But also I think how we think we will use it would get better. so can you think of other use cases now because we're looking at things like you know if you're a tyre company or an automotive company for me as a for me as a customer a user for example if i want new tyres for my car it's oh right i've got to get a website or i've got to phone up my local independent garages have they got the tyres have they got one for a ford fiesta so i can see that it could really help people.

I wouldn't want to call 1,000 higher places, but I might want to call 10 in my area rather than the one or two that I know. So I'm interested in where you think other use cases could be applied for this type of voice and AI. Yeah, I mean, so what you're describing there is kind of like how ChatGPT has agent mode, right? You can give it a task and it goes and tries to find answers and what voices is just another like tool in the tool belt of an agent to be able to get an answer for research for pricing or for anything you want to know reservations I mean really voice is a very untapped area from an AI perspective because as you said, it hasn't been as developed until recently with the advent of Eleven Labs and other voice AI companies.

So from a research standpoint, right, hard to get data points, this is really good for, because, I mean, as I learned after, I mean, to me, this was kind of an obvious way to go about this, but I'm using AI tooling like 12 to 14 hours a day, right? So yeah, let's just call and get the answer. But from a traditional research perspective, I think this is relatively novel because this type of scale and insights is like not really been able to achieve without an enormous sum of money or a lot of man hours.

And so if you think of like, okay, that cracks open the door, what else can we do there? And where might we be able to find just hard to get information? But yeah, from a personal consumer perspective, right? But also from just a larger business perspective, I think there are a lot of options here.

Yeah, yeah. Yeah. And I think we're touching on a point here that a lot of people sort of say, oh, AI is going to do people out of jobs. And there might be particular cases where it does people out of jobs.

But I'm more excited by the point you touched on there that, you know, the AI potential now to do things that we weren't able to do at a cost or with the human number of hours that we're that we're going to be using gives this sort of augmented opportunity. opportunity because then you know you can decide to go to a particular pub or you could decide to get your tires done in a particular location or you know and whether that's an agent which is a computer talking to you know a website or whether it's a voice one or it might be some other sort of interaction it brings us all sorts of new opportunities so in terms of your more broader things that you do what sort of things are you working on that that are now using this augmented capability from AI?

Yeah. So as an AI engineer, right, I was, with the Gindex, I was just trying to build in public, right? You kind of have an idea for something, you show how you made it, what tools you're using, and you make videos and tutorials about it. It's how other people learn.

It's about how you learn, right? So this was just one project that I thought was interesting. And I guess other people thought it was interesting too. But for, you know, from a more general perspective so i work as a like a white glove consultant to go into companies and help them to implement ai tooling and workflows that work for their organization given whatever it is that they're using already so um you know i'm i'm working with a comms agency right now to help them build like an ai first ai native like comms and pr agency um i'm building a ticket booking system for an immersive experience company, right, that kind of uses agentic AI in different processes.

Really, like my interests are kind of working on interesting things that I think are, that I can really make a difference with. And then also side projects that I think are fun, that other people might find to be fun too, that kind of bring insights into what's possible with today's AI. Yeah, yeah. And I agree.

And I think we're on similar sort of parallel sort of areas, because like one of the examples that we've tried is I'm a founding member of an investment, angel investment group. So going back two, three years ago, we said, well, surely AI can read the pitch decks and say it's me reading through lots of emails and reading the pitch decks just to find out what's the size of the venture, how much investment they're looking for, have they got the tax incentives that are relevant and stuff like that.

AI could do that two or three years ago. And we then went and tried it with, okay, I've done some background research that AI can do as well, very well now, you know, searching companies, doing competitive analysis. And then the audio bit or the video bit, you could get the AI to ask the entrepreneur a bit more information. Right.

Because, you know, you can't expect all that in the deck or the background. It might be, oh, Matt, it's really interesting what you're doing. Tell us a little bit more about what you did in the past just to build that richer part. and then I think the the point I'm getting to is AI could then change the business model because if AI can look at the pitch decks the AI can do a lot more of that it lowers the cost from before where you know a lot of that research would have been done and the investment managers wouldn't have bothered investing less than a million pound or five million pound because of all that background work.

Now AI can do a lot of that work and it's more cost effective to invest a lower amount of money and then get your AI to monitor that business, what sales they're doing, what traction they're getting, and then invest more money as they move through the process. So I think it's really interesting for people now in lots of different sectors, whether you're a tire business, a pub or whatever to think where can I use AI to improve my current processes, do things I couldn't do before, like the Guinness index, but also then changing the business model in the sector.

So are you excited or seeing those types of things as well with organizations now could see new opportunities Yeah absolutely What you described is in order to have that mindset you have to have like a peek into what possible right And it a little bit overwhelming for organizations that aren using AI because I mean honestly the landscape shifts so quickly and there so many things coming to market where like how do you get started But once you realize, okay, let me sit down from a baseline, from a research perspective, let me build context, explain what my business does, how it does, what metrics we use, what industry we're in, feed in our P&L, feed in as many inputs as possible, and then give Claude, right, Claude Code, an open-ended research task and say, okay, based on everything you know about this, and your expertise as an analyst, how do you think that these things can be changed?

How would you advise me, right? And so that base thing from a research perspective is really, really powerful and it lets you see opportunities and ways of doing things that maybe you just didn't know about or hadn't been exposed to and hadn't thought of yet. Yeah, yeah, yeah. No, I think it's, I've seen that over the last two to three years as well that until you get those wow moments So, you know, that's what I've been having over the last three years.

And when we run these programs, some say, wow, how did it manage to analyze that document? How did it manage to create that image? How could it have that conversation with me and then have those insights and that from the background? We get all those sort of wow moments.

But until you've had those bit of experiences, it's very difficult to say, well, how do I apply that to my business? And that's right. So we've got to help people go on that journey to of learning. right yeah i mean and i think that apis are the key to like agentic ai right having data sets that have apis that are like input so an ai can plug in to a data set for companies house for example is a great example it's a great api you can get demographic information you can get information by postcode about a market you can get information i mean the amounts of insights you can get and And then AI is great at connecting dots across things that maybe you hadn't seen.

But it really helps to identify market opportunities, I think. Yeah, I think that's a good point. Because taking the example we did with the investment pitch decks, yeah, it's going off and find that information that would have taken us too long to do the Google searches and stuff like that on. but also having the context of like in our case 10 years worth of previous pitch decks and knowing where they got to it's like you know i've been a member for 10 years but i can't remember that we saw something similar seven years ago and dennis was one of the ones that invested in that business and now the ai can know that sort of stuff that would have been part of an organization you know you would have had to have kept a database or you'd have you know the people would have had to have remembered and sort of stuff like that now the ai can do that and then the humans can take over in the next bit right yeah and i think the i mean the context of training that model for whatever it is that your organization values or like the metrics that you find successful companies have i mean it's something that i mean i i found and i tried to test with anything that i build is like, how do you break it?

Right. So, and also how do you make sure that whatever system you make, whatever AI agent you make is not racist, is not homophobic, is not all of these things. You just kind of have to barrage, like get at it and try and make sure that it is aligned with whatever it is that you want it to do very carefully. And I think like the area of AI safety in AI research is critically important for understanding what's going on underneath the hood a little bit.

Because if you're trusting an AI to baseline investment opportunities, you need to make sure that it's doing the right thing. And the best way to do that is to give it the best context you can. Yeah, yeah, yeah. And I think those risks and guardrails and that is one of the key parts that we sort of cover within our Henley strategy and implementation sort of program.

because understanding the basis of how the system is starting to think, then how we work with it I think is so important before you get down to the actual getting your hands on the keyboard or getting your hands in programming things. So one area now which I guess you're being impacted or getting the advantage of is using things like co-work and getting the systems to do the coding and do the developments. Could you give us a bit of an insight in how development now works for you as opposed to how it might have worked five years ago or something like that?

Yeah, so it's a major shift with how software development has changed. That's really rattled a bunch of cages. But now code is no longer really written. It's generated, right?

So it's describe what it is that you want and that code is made to do that thing. And having a base of understanding of computer programming is critically important, right? Because you able to peek under the hood and say no don do it like that Please do it like this Or oh what if you tried from a problem solving perspective right Like what if you tried this instead of that So that foundation is really important and really powerful But I mean code is being generated Websites are being generated Apps are being generated.

They're not being written in the same way that traditionally have. And that's got software engineers in a bit of a pickle because, you know, you hear a lot, AI is taking jobs. I think, you know, I heard it recently. I don't know where I heard it from, but I don't, I think people who use AI are taking jobs.

You know, AI is, yeah, AI is a part of that. But really, it's everyone having an understanding of how to use these tools, because you still need humans. You're always going to need humans in these jobs. but if you're the one who knows how to use the ai correctly and well you have a major advantage yeah yeah yeah yeah and i think i would draw the parallels there to that the investments of the type scenario that i'm talking about because you know coding is is on the leading edge and has been happening now for a couple of years but like in my investment example yes you know the people that used to do the google searches and do the the pitch decks and reviews their work is going to be supplemented and augmented and made faster than that with the ai but as you sort of said it's the investment funds that change their business models and engage with the entrepreneurs for example and with the investors make sure they can do the real things rather than just spending their time producing reports and producing analysis and producing financial plans those things now can be done a lot more efficiently and probably a lot more accurately and with a lot more context, as we've touched on earlier, than what we can do in the, you know, and so it's about augmenting.

It's about enhancing what people can do to deliver real stuff, I think is really important. Yeah, I mean, the years I spent really becoming very good at Microsoft Excel, right, are still are still useful because then when i generate a pnl or i generate something i can look formulas and i can say yeah that's correct i know what's going on or like oh no hey there's an error here right so that base of understanding is still very important yeah yeah yeah yeah and i've had so so many frustrations with you know excel and it didn't quite copy a formula right that you put into somewhere but it always required that bit of knowledge of sort of saying no it can't have a 90 profit margin which is why is that or or what you know so there's been an error that the human has done or now the ai can do a lot faster and and whatever because it didn't have the right context but you do need that understanding you do need that um that that knowledge to be able to to know when it's not right and then understand the direction it's going right you can interact with it a lot more yeah i mean i think one of the more the more interesting observations that i'm seeing like currently right as i'm using some of these tools is i say okay hey please complete this analysis and it's like oh well no i can't do that i don't have access to that tool and i'm like oh no you do and it's like oh you're right i do i'm like oh but i can't do that because of this like no you've done it before do it again and it's like oh you're right i'll do it again so it's oh it's almost like my sheer force of will of saying no you can is having the ai correct itself to then actually complete the thing and it's like well why do i i have to be as insistent as possible for it to believe that yes it can do this thing or it has another way around it and it's it's very it's very specific i don't know it's it is becoming quite weird and that isn't it and and how we you know um make it human like and interact and that with it i know when i when i first started using gpt i thought oh yeah it feels like i'm interacting with the person or you know whether it's a lower level intern or it's an expert on a particular topic and i ended up creating names for mine grace peter and tony with an i because i asked it i said well give me three names that are male female and gender neutral and and for a little while i was starting to think okay well grace is going to do this but tony will do something else so have you got into that mode or are you keeping it arm's length a bit like i i experimented with um with OpenClaw, as a lot of people have, right?

And you give your claw a fun name. So my claw's name is Rooklaw after RuPaul. So I've got Rooklaw and she just does things that I need, but she was eating a lot of API credits. So I've parked her for the minute, but yeah, I mean, it's all right to anthropomorphize things, right?

But it's important to remember, always say please and thank you because when machines rise up, you want them to say, hey, we remember you. Yeah, yeah, yeah. Well, I do. It just feels more natural to me to simply thank you, to be honest, than that when you're doing some of these things.

But it's fascinating, fascinating. So, Matt, it's been a great pleasure talking with you. Great to get insights in the Guinness Index, but also the broader things that you're doing. So it would be good to share the sort of stuff of that with our audience.

It would be great to keep in touch. No, thanks for having me on and talking about this stuff and for logging the cost of a pine at your local in Ireland. I appreciate it. Thank you.

Good. Thanks, Matt.