Callum Healey: Helping agencies win with AI

Pauline Bertry 0:13

Hi Kalum, welcome to the dan. Thank you so much for finding time to join me. How are you?

Callum 0:19

I'm very well, thank you, Pauline. Yeah, thank you very much for having me.

Pauline Bertry 0:22

Thank you so much for joining. I mean, I'm super excited for our conversation. I generally like to start with asking people about their stories. And I mean, in your particular case, it's also super exciting to me because we haven't worked together in the past. So I'm I'm all ears and like super curious to learn more about you, what you are doing, your past, your present, and maybe how you're thinking about the future.

Callum 0:48

Absolutely, yeah, I'd be happy to share. So I am a recent psychology graduate, and this was in the summer of last year. And as people can probably tell from the title of this video, this is very much about AI and agencies. And that seems like quite an interesting pivot. You know, psychology and AI are almost on opposite ends of the spectrum. I

Pauline Bertry 1:11

Mm-hmm.

Callum 1:11

was always more interested in the science of the mind, so studying neuroscience. Brain chemistry. I've always been interested in the why behind pretty much everything in life but people are at the forefront of what makes up life. And so studying people what makes them think the ways that they do, which leads to behaviour, that was always very interesting to me. I didn't really want to take the traditional route of becoming a clinical psychologist, though, as I'm quite blunt and focused on the science of it all. I was more interested in the sort of quantitative facts of the subject. And so data analysis was a real sort of opportunity for me to go into. I then, throughout my degree and using AI as a nice helper along the way, realized that AI is almost the best data analysis that could exist. In the the sense that it can ingest so much data at in such a quick time. And process it so coherently, understand connections. I was listening to a podcast once and they said that AI compared the anatomy of a wasp's nest to a nuclear bomb because it just made those connections and saw that there was patterns there. And so very clever analysts. So I decided to learn the beast instead of doing the it's almost like a mathematician using a calculator. I'm like studying the calculator of mathematics. I'm now studying AI, the technology. And while it does seem counterintuitive for psychology and AI as a career, there are a lot of overlaps. I kind of started off my journey studying what's called machine learning and deep learning, which is all about the architecture of these AI systems, how they were created, how they go about learning. And we modeled AI's intelligence after the human brain, because that was the only intelligence that we were aware of. Brains in general, every animal in the animal kingdom has a brain, and that was where intelligence came from. So, as a very simple example, one of the first architectures of AI was called an artificial neural network. And of course, humans have brains and full of neurons. And there are many similarities here to I won't go too much into the jargon about neurobiology, but there's such a thing called neurotransmitters in a synapse. And There are such a thing called action potentials. These very much translate to artificial neurons in the sense of weights, where a neuron lights up more and is more likely to fire. This is based on mathematics, but the weights can almost be compared to that action potential. And then they have a bias, which is like a natural threshold. And so that dictates whether these neurons turn on and off. And AI, at this case, has now. These these weights and biases are called parameters. They now have I think five trillion was the most recent chat GPT model. And

Pauline Bertry 4:16

Yeah.

Callum 4:17

after my learning, I now help agencies. And so my dad, Gareth Healy, who's also been on the show, he has been experienced in the world of marketing agencies for 30 years now. And so I thought that there was no better mentor to get me into. a certain audience than Gareth. And in this sense I'm using AI and helping that target audience and target market target market effectively embed AI into their operations to promote efficiency ultimately, but also it's a form of change management that over time is more important as technology grows. So that's that's that's kind of where my story is at the moment. I'll I'll let you get a word in Paulina.

Pauline Bertry 5:05

Yeah. No, no, I mean of course the the podcast is about your story. So if I'm silent for the entire time, that's perfectly fine with me.

Callum 5:12

Well, that's great. We'll see how it goes.

Pauline Bertry 5:15

but I'm curious a little bit, I mean there are so many things I'm curious about, but let me start with maybe asking you to kind of walk me a little bit through this transition you were going through when you graduated and you started running your own business. I mean I I understand you have a mentor and that facilitates things, but I know this must be like still a very challenging, very challenging exercise, right? And I'm I'm curious to learn a bit more about like what was happening for you in in this period of time. Like how were you finding your own angle, your own value proposition? Like why did you decide to launch your business straight away out of university? Like what What's the story

Callum 6:02

Yeah.

Pauline Bertry 6:03

there?

Callum 6:03

That's a great question. I I think ultimately it starts out looking at the current state of the job market. I think

Pauline Bertry 6:11

Mm-hmm.

Callum 6:12

thirty years ago, perhaps if you had a degree, I finished with a first class degree from Manchester Metropolitan, you would almost be guaranteed a job in that sense. And admittedly I didn't look too hard. I spent maybe two months applying for countless jobs, but never quite got

Pauline Bertry 6:28

Still.

Callum 6:29

Yeah, it's still a lot of time. And I was doing this along a part-time job. And as a lot of teenagers do as they grow up, I you know, I've experienced different sectors of kind of teenager roles in the sense of retail, hospitality, factory

Pauline Bertry 6:44

Yeah.

Callum 6:44

work, labouring. I've kind of dabbled in all of those. And I think as well as the job market being tough and me just going, I'm gonna make something for myself here, because I mean, just on that point. I'm I'm a very big believer in the fact that it's never been easier to start your own project and your own business because life is so connected now. And you can market offers and deliver value to people all across the world. And there's so many different outlets where you can let people hear about your voice, you know, almost shouting from the highest rooftop you can about, you know, the value you're trying to create, the target audience and whatnot. And I think AI made that it kind of accelerated in the sense that if you were to learn new skills and you needed to ideate on certain strategies and you wanted to learn how to set up a business, me sort of studying AI, it's almost my partner in crime in sort of getting

Pauline Bertry 7:42

Yeah.

Callum 7:42

me into the start of business. I think my personality trait as well is I'm quite stubborn, strong-willed, and competitive. I I have always hated losing and that almost kind of puts me in the sense of I will not stop until I win. And so I I've got quite a a high tolerance for for sort of hard work and almost psychological pain in the sense that I just love challenge and overcoming adversity I've just learned in life that if you apply yourself for a long enough period of time, it's almost a guarantee that you will succeed. when I was starting my business, I was working more than eight hours on weekdays for six months and I didn't earn a penny. But I sort of was Thinking of new offers and how c how I can approach certain markets, doing free work for people. And then on my first month, I earn multiple thousand pounds. it's almost like the the miner in the cave where he's he's he's near the diamonds, but he turns away right at the last second. It's like I've always kind of understood that as long as you stay in that game and and work hard, you can achieve. And with my kind of personality, I've never much like people telling me what to do. As

Pauline Bertry 9:07

Okay.

Callum 9:07

much as as much as I like understand when I'm in the wrong, of course, I'm willing to admit that. But I'd much rather be in charge of my own time and my own kind of autonomy. Even though I work more now than I would in a nine to five, I'm doing it for me and it's

Pauline Bertry 9:22

Definitely.

Callum 9:23

kind of it's key. It's kind of my purpose. So it is I'm enjoying it. It's it's a s bit of a strange analogy as well. I grew up playing a lot of video games and there's a category called RPG and this me this is kind of all based around the idea of progression and getting better.

Pauline Bertry 9:40

Yeah.

Callum 9:41

Even if that's through like very tough challenges, like massive boss fights, etc., you you get the picture. That's kind of just instilled into my mindset that I just kind of love failing and failing, but trying and seeing that little progression along the way and I think that's entrepreneurship in a nutshell. It's almost I I just get addicted to that kind of progress over time. And I'm enjoying the journey. Yeah. It's going really well.

Pauline Bertry 10:07

Yeah. That that's exactly I mean, I'm not a big gamer, but that's exactly the analogy I was given quite often when we were talking about building people processes inside agencies or inside organizations, is that you need to make it feel like a game. Then people get addicted, they stay and they play. Makes sense. I love it.

Callum 10:28

Yeah, exactly. Yeah, I think it's great. it's all about that progression for me. I I kind of I've I learned very early on that external rewards in the future can never truly exist because you can't experience the future, you can only experience the present.

Pauline Bertry 10:44

Yeah.

Callum 10:44

And so in any kind of endeavor in my life, I I like to focus on the day itself and how can I improve

Pauline Bertry 10:52

Yeah.

Callum 10:53

my position from where I am, what's the the next best move. And if you even if it's something like sales you're doing on on LinkedIn, if

Pauline Bertry 11:02

Yeah.

Callum 11:02

you get more impressions or one money one month you earn more money than the last, or you you deliver a great client project, you know, that's that's kind of enough for me. I like that present I like that progression in the moment rather than thinking too far ahead on like the end goal because I do think it's the journey that's the important part and it's the

Pauline Bertry 11:22

Yeah.

Callum 11:23

happiness of the pursuit is something that Jimmy Carr said, you know, rather than the pursuit of happiness, I'm I'm kind of relishing the the progression along the way and the journey that I'm on and entrepreneurship. There's many different sort of turns and unexpected directions that you might find yourself in when you're in control, which I just love that novelty.

Pauline Bertry 11:43

Yeah, I I agree. I mean I I'm I relate so much to that. But then I mean it kind of very naturally leads me to to my next question about what are you working on now. So I you you mentioned a bit AI and agencies, but like what what what are you up to? What are you doing with them?

Callum 12:06

Yeah, absolutely. So as I mentioned, Gareth, who has he had a lot of or has a lot of experience in the agency world and was a former CEO of a marketing agency. And for the last ten years he's been doing his own consultancy. And so I learned very early on while studying the actual architecture of these AI systems, it should never be a standalone initiative, in my opinion. And I always think that It purely is a tool to amplify human competence and not just human competence, the competence and the quality of the underlying systems that it has. And so I decided and me and my dad were discussing together that he's already auditing and and working with agency A founders on a retainer based system and and sort of audits, like I said. P uses a system, he's written a book. It's called the Standout Framework, where we look at agencies on eight levers of their kind of

Pauline Bertry 13:04

Mm-hmm.

Callum 13:06

makeup. So we've got sales, team, ambition, numbers, development, operations, uniqueness, technology. And we look at all sides of these, and that's kind of the experience, the framework that my dad has made for his experience in his career. And I decided that being a consultant as a 22 year old doesn't carry a lot of weight. Because people are saying, Well, you've not really had much experience. You've just graduated. Why should I listen to you? And so

Pauline Bertry 13:33

Mm.

Callum 13:33

I'm almost leveraging Gareth's authority and experience and learning from him daily using the same market as his. But my skill and expertise is very much in AI and understanding these systems,

Pauline Bertry 13:47

Mm-hmm.

Callum 13:48

helping teams effectively embed them. And so I've taken my Gareth's existing framework. And it's all about accelerating it rather than having my own framework for AI and my own kind of

Pauline Bertry 13:59

Yeah.

Callum 14:00

system. It's going, how have agencies succeeded? What are the common strengths and weaknesses and patterns that people have seen and Gareth has seen in his career, of course, and made a framework out of it.

Pauline Bertry 14:13

Yeah.

Callum 14:14

That doesn't change, in my opinion, now that AI's here, because AI is purely

Pauline Bertry 14:18

Yeah.

Callum 14:18

a way of either It it's it's an ex it's an accelerator in the way that if there was weaknesses in that certain operating model, those weaknesses would be sped up as well. And they can kind of lead to a lack of control and the sort of the weaknesses and the risk in that existing system can go out of control. Whereas the strengths in that existing system get even stronger because they're quicker. And I think speed is one of the ultimate values in any business, you know, whether that's speed of decisions, speed of something like a deliverable to your client, you know, and it's only effective when AI is speeding up something that has a high quality to it. And so I help agencies initially for that same reason, I don't just dive straight into a one size off implementation. It if that was

Pauline Bertry 15:12

Mm-hmm.

Callum 15:12

possible, I probably would do it because it's a product size service, a little bit more sustainable. But I very strongly believe that to get value out of AI, it completely depends on the context of its user. And

Pauline Bertry 15:25

Yeah, of course.

Callum 15:26

this this is a side note on just the people side of AI. And perhaps there might be people scared of AI taking over jobs. I would say that is very much not true in the sense that it's the people who can use that who learn to use the technology effectively that will replace the people who don't. And The effective way to use the technology, in my opinion, just as a broad kind of thing that I've noticed, is that it's about what makes you have skill and expertise and how can you design AI systems that speed up the knowledge that you already have. And so a a lawyer using AI is going to use AI much more effectively for law than a doctor would for if they were trying

Pauline Bertry 16:12

Yeah.

Callum 16:13

to use AI. For law. You know, it's all about I was actually working with an agency last week and I designed a system for them on Claude. We're using a Claude skill. And it was to do a weekly carousel, like a sales initiative on LinkedIn and various other kind of social platforms. And this system would research their certain criteria based on it was like a seven page carousel, and they had certain variables for each page, like what the research should consist of. Got the research, the design, which are used their hex codes and fonts and whatnot, and then it designs it and pushes it to Canva or Adobe or something like this. And they they thought it was brilliant, but the there was five people in the room and questions started to arise, and one of them was saying, But we only focus on research in the last two weeks, for example. Or the text is slightly too long here and it's too much too much jargon in the text. We want to strip that back. Those Questions and comments are exactly an example of expertise because they know that

Pauline Bertry 17:13

Yeah.

Callum 17:14

workflow. They understand the process and what makes it good because they've done it for so long. And so they can then put that knowledge into the AI system. I made it because I didn't know anything. well, I knew a lot about the agency, but not too much about that workflow. They just gave me some examples

Pauline Bertry 17:29

C of course.

Callum 17:30

and the brief. But what makes an AI system powerful is the user's understanding of that. process and the thing that they are so good at before AI was there. It's about how can you put that in the system yourself. And that yeah.

Pauline Bertry 17:45

That's so true. I've s sorry, I just want to add one point here. I was reading a research that basically the they run as like a survey or a conversation with different people using AI for different use cases, and what they realized is that people would generally be overwhelmed by the quality of the output produced by AI outside of their area of expertise. And they were underwhelmed by the quality of output produced inside their area of expertise. So for example,

Callum 18:19

All

Pauline Bertry 18:19

like I'm not an accountant, but when I talk to AI about the taxes and like what are the different elements I need to consider in my paperwork, I'm like, wow, that's amazing. But when I do like design or product strategy with it, I'm like, Maybe you

Callum 18:39

Yeah.

Pauline Bertry 18:40

you could do you could do a better job. And I'm pretty sure it's gonna be the other way around for my accountant. So I think that like what what you are mentioning is super valid because depending on how much expertise you have, better you can become at leveraging EI for your specific expertise.

Callum 18:59

Completely, yeah. And it's very simply you can it's analogous to a mathematician using a calculator, like I said before.

Pauline Bertry 19:06

Yeah.

Callum 19:07

The quality and the the skill level of the mathematician is going to determine whether they get to the calculation rather than the technology in between the steps from A to B. There's actually a pretty straightforward reason for why for what you just said there in terms of, you know, if it's your expertise, you're underwhelmed, if it's not, you're overwhelmed. And The way that these systems are built as revolves pretty much entirely around calculus. And

Pauline Bertry 19:36

Yeah, of course.

Callum 19:37

it's the way that these especially LLMs in this sense and pretty much all of the AI systems, what they're doing when they're outputting text is basically just predicting the next word to generate over and over and over again. And they do that by ingesting the context of what's in the chat and their training and whatnot. But if you see this on a graph the way that these AI models are thinking, you might have ten words and the word riverside, like the you know, this recording of the podcast, that might have an eighty percent chance or an eighty percent likelihood that this is the next best word. And then you'll have nine other words that all have, you know, twenty b or it'd be in this case fifteen and then zero point zero two and zero point zero point zero one and it it goes down in that way. And so The way that these models work is by picking the most likely answer. They default

Pauline Bertry 20:29

Yeah.

Callum 20:29

to the mean, they default to the average. And so if your expertise

Pauline Bertry 20:32

Yeah. And this is why

Callum 20:35

is above average, it's going to almost drag you down in that sense to to to that baseline, you know.

Pauline Bertry 20:41

Yeah yeah. Th this makes sense. And this is why like you're you like when people are talking about AI creativity, you don't have the same result in like ChatGPT or Claude because they're not using the same the same metrics to like what they consider a good practice. But I

Callum 21:01

Yeah.

Pauline Bertry 21:01

wanna I want to go a bit deeper on on this topic. I mean I know you you've been doing this for quite a while and I assume like every agency is like very different and they have their own processes, etc. But what are the like what are the typical patterns that you see or the typical use cases that you've observed where for like most of the agencies there is a potential for as you call it acceleration or enhancement with AI systems.

Callum 21:37

Great question. And from my experience in the last year of working with various clients in various different countries, which is something I didn't mention before, but you know, you can start a business now, and I had a client in Hungary, for example, and it's like yeah, that would never have been possible in history. So

Pauline Bertry 21:54

Yeah.

Callum 21:55

good time to start a business in that sense. But yeah, I actually often see that agencies have a very low proficiency and very low maturity.

Pauline Bertry 22:05

Mm-hmm.

Callum 22:06

Do with AI. And it was quite surprising to me because I think anyone who's interested in AI, perhaps people listening to this podcast, all the algorithms on their social medias, their network, their friends, seems like they're all innovating all the time with the newest AI technology. And you can feel like you're behind, you know, there's a new model you're not caught up on, or a new

Pauline Bertry 22:28

Yeah.

Callum 22:28

way of setting up a system. What I've seen most often is that teams Are getting efficiency out of the tools, but for very basic use cases, like using

Pauline Bertry 22:40

Mm-hmm.

Callum 22:41

Chat GPT, just the chat feature, not uploading any sort of context documents, which is called rag in this sense. And I can get more into that of why context is so important. That's a massive thing. But it seems like agencies and team members in the agency are trying AI out and they realize how effective it can be. But they don't know where to start. And that

Pauline Bertry 23:04

Mm-hmm.

Callum 23:04

is a massive reason of why the kind of bottom of my funnel, when I begin to work with agencies, I'm always saying, unless you know exactly what you want, and I kind of agree that this is the right thing to do, we do an audit of the whole agency. Because AI whereas a calculator is just for c calculations and mathematics, AI you can practically use in any task. Because it's just interested as much data as possible. And so people don't know where to start. It's almost a paradox of choice. Like, what's the most effective thing that I can do? What's even possible? And so I do the audit to get that element of clarity and prioritization and go, okay, here's a roadmap based on our conversation and the answers to my questions and the frameworks that you have. How what should we do first? And Pretty much always the first thing that I see is that there's a lack of governance around the tools. And there's

Pauline Bertry 24:03

Mm-hmm.

Callum 24:03

there's a lot of implications for this going forward. And I think the longer and the the more rapidly AI tools get developed, the worse this will get, because they'll become more capable,

Pauline Bertry 24:14

Yeah.

Callum 24:15

more destructive. But I see free plans, first of all, people using free plans that aren't an agency plan, or even personal plans, which still aren't very good.

Pauline Bertry 24:24

No.

Callum 24:25

And This is completely makes them at risk to shadow AI, people using it without the founder knowing or leadership team knowing. If agencies

Pauline Bertry 24:35

Mm.

Callum 24:35

are working with regulated data that they have in client MSAs that they should keep very secure and private, and one of the team members puts that in a free Chat GPT chat or a free Gemini chat, that data is not safe. It's

Pauline Bertry 24:51

Yeah.

Callum 24:51

there's that old adage if the product is free, then you are the product. You know, these tools are there to gather data from its users. And when you're

Pauline Bertry 25:01

Yeah.

Callum 25:02

on paid plans and especially more bulletproof business team plans, you can turn this model training off and regulate it further

Pauline Bertry 25:09

Yeah.

Callum 25:09

further. You know, in my case in in the UK, it a lot of these models comply with UK GDPR. And even then, you know, I I have clients who were going, We're not going to connect Office three six five because there's too much sensitive data in here. Let's just create a secure environment on our own. But completely governance is the thing. I would almost always say for agencies, if you're running an agency listening to this, you would you want a company plan. And for every AI tool you have, you want the company plan that everyone uses. Because not only does this give you control and you're able to under you're able to have full knowledge that the settings are configured correctly, it's also The only real way you can convey AI usage to your clients, because you can say, we have this club, say if it's Claude, we have a Claude plan. This is the name of the Claude plan, this is the type of plan we use, these are the members on it, these are the settings we have configured. All of that could be in a client MSA at the beginning of the of your relationship.

Pauline Bertry 26:13

Yeah.

Callum 26:14

If someone's using a personal Chat GPT plan, It isn't a Claude plan for one. So there's different implications there. Claude doesn't have any data storage in the UK or Europe. It's purely US servers. Claude have a maximum or a minimum retention policy where it's held in their servers for at least 30 days before deleted.

Pauline Bertry 26:34

Yeah.

Callum 26:35

All of these AI applications have different implications. And even if you did use a personal plan of Claude, but you had an agency plan, it's still not got the same settings. So it's you can't be completely transparent with clients. And so definitely

Pauline Bertry 26:48

Compliant. Yeah.

Callum 26:50

getting that company wide plan is the first start. And there there are many more to come, but if you if you have any comments on that before I do do a monologue. Yeah.

Pauline Bertry 27:01

No, the that this makes sense. I'm I'm also curious to learn a bit more about like the automation I like elements or like you were mentioning the the content generation support systems. like what what are the typical kind of patterns you are observing there?

Callum 27:24

Mm, okay, yeah. Good question. I would say that once the boring governance is out of the way, like I just said, you know, topping out all the client contracts. It takes time

Pauline Bertry 27:31

Yeah. And it it takes time. It takes time, I know.

Callum 27:35

when it's completely necessary. And it's like if you don't do it, you're building your data security on sand. You want to spend a lot of time building a stable base. Then when your team members want to innovate and build new stuff with AI, they can do it in an environment where there's no risk, which is completely what we want. I think that's actually where team adoption comes in because I focus a lot on the people side. How can we incentivize AI usage and train people on certain systems? And because of what I said about prioritization when it comes to AI, I do think my the best method of training that I've discovered is actually automating one of their specific workflows. So the people

Pauline Bertry 28:18

Yeah.

Callum 28:19

who aren't very technical with AI and they're not sure quite how to use it. If they are they are the experts on their own workflows, they know the certain pain points, they know the steps from A to B, hopefully, all of these things, output quality. And so rather than there being a sort of common pattern that I've seen with with certain types of automations, it's very much been what is the context of your agency and how can we make a workflow that best represents what your work what your workflow is. I will say in general, it's more about people learning custom systems. So let's say Chat

Pauline Bertry 28:59

Mm.

Callum 29:00

GPT has custom GPTs, Claude has

Pauline Bertry 29:02

Yeah.

Callum 29:03

projects and skills, Gemini has gems, the list goes on. That's a very important starting point because once people realise that even prompting, which might be controversial, but I don't think prompting is very important at all because I think purely it's the context that you put in these AI systems that is the value. And as we discussed earlier, the people who have that expertise and strategic

Pauline Bertry 29:27

Yeah.

Callum 29:28

judgment have that context. So that is the value. And it's about codifying these workflows into the AI itself. R aside from that, I think that time tracking is always a very polarising topic in agencies, something that

Pauline Bertry 29:45

Interesting.

Callum 29:45

it's quite hard to manage. And even for companies, so I I work with a media planning agency and they they earn commission, for example. So they didn't really need to keep track of their time. But I think that when it comes to AI, there's a set it's it's a lot easier because of its speed to over service or p be a perfectionist. And so

Pauline Bertry 30:06

Yeah, that's true.

Callum 30:08

Time tracking systems that make it very seamless and automate that system of time tracking can allow you to sort of it it can generate you reports to leadership team, for example, and say, This is how many hours were saved this week on this new AI process. This is where the these these are the hours that were freed, these are where you can put them, instead of them leaking into sort of low ROI meetings and admin and and whatnot, just to sort of get through the day because Unfortunately, as much as AI speeds up processes, it's not going to make the days the workdays shorter. That's something I found out very quick. If anything, it's making them longer because the opportunity cost of not working is now higher. There's there's there's that much possible. But I I've done time tracking automations for three agencies now. That's quite a common one because in general, it's like what are the systems and the processes that we have we know we should be doing. that we really don't want to do and is very manual and

Pauline Bertry 31:08

Mm.

Callum 31:08

strips away from our creativity and and strategy skills and people skills that we thrive on rather than just sitting looking at a spreadsheet and analysing loads of data. Automations for time tracking is one of those because time tracking is quite a hated part of the industry, although it's very important. I've speaking to a CFO that me and Gareth are quite sort of familiar with and he was comparing an

Pauline Bertry 31:34

Yeah.

Callum 31:34

agency to a car manufacturing business. Like a a car manufacturer knows the price of every nut and bolts in their model, which allows them to then sell it for a suitable price and and make perhaps make repairs. In a knowledge-based business like the agency, people are that raw material. People are the nuts and bolts. And the cost rather than money is time that you spend on processes and As these processes get sped up with AI, it's much harder to track them. And so automating this is is very important. I'd say that that's a massive one I've seen.

Pauline Bertry 32:11

Yeah, that's a big one, I agree. I have

Callum 32:13

Yeah.

Pauline Bertry 32:14

one more question, because you were mentioning a few times about the context and how important this is.

Callum 32:19

yeah, absolutely. Yeah.

Pauline Bertry 32:22

And I would love, I mean, I know you have a very special approach to like building this context, which is teaching AI is like teaching a new employee. And I'm very curious to learn a bit more about how you see that and like what is the right way for a company which is as as we already kind of acknowledged, not necessarily very deep on AI and they have they're not technical, they haven't done this before. So how do how can they leverage AI in the best way to build this context, to build this knowledge, and to teach their new employee?

Callum 32:59

Yeah. Yeah. I I think that is the most important thing I've discovered about AI. I I think if if there was one takeaway out of this video for anyone listening, it would be the importance of context. And this might be more compelling to me because of my psychology background. People have the biological hardware of a brain and they have the the way that that brain learns and develops is through experience from the environment and stimulus from the environment. Then it grows, learns new skills, etc. AI is very similar in the way that it has its computational hardware with these algorithms, learning algorithms, the initial data that they've been trained on to understand the meanings between words, etc. But then it's learning and training de is data. And that's AI's version of the environment for people. And so if you don't train AI on context it in on your context specifically, it's like hiring a super fast intern with internet access. They've got all this data, they're very quick, they can make coherent arguments some basically any field. Like we've said, that's all quite diverted to the mean, which so it has a base level understanding of every sort of industry or most industries, but It doesn't know anything about, in this case it would be agencies. It doesn't know anything about who you are, who you serve, your mission, your values, your brand voice. SOPs, which I've I've quickly realized isn't too too much of a used word in agencies because the the creative work is a bit more fluid in its operations, but SOPs in the sense of okay, what are the repeated workflows that we do in the agency?

Pauline Bertry 34:50

Hmm.

Callum 34:51

And the list goes on. And it's basically like you said, the employee analogy is you want to train AI like you're onboarding a new employee, because that's the way that you can get AI to work for you rather than picking a system that works for everyone. You know,

Pauline Bertry 35:06

Hmm.

Callum 35:07

it's simple if you went to an off the shelf AI tool, and of course you use these tools to then create custom systems, that's when they become valuable. But when you don't configure them, A a ten year old could ask AI about what Lego set they should buy. A an eight-year-old might ask the best flowers for their garden. It's designed to be used by everyone, which is great and it's it's important in that way. But you wouldn't want to hire something that is good at everything for your agency.

Pauline Bertry 35:36

Yeah.

Callum 35:37

You want some you want it would be an employee in this case, but you want a system, you want technology that works specifically for you rather than for Anyone. It's like that whole thing, specialization,

Pauline Bertry 35:48

Yeah.

Callum 35:49

you know, drack of all trades, master of none. AI is a master of nothing until you give it the value that you've learned and your expertise into the systems. And so that is completely the first thing that I always do with agencies. It's like. And that that's great for training as well. Like I said about prompting before, people can get overwhelmed by these prompts that are 200 lines long and they have big handbooks.

Pauline Bertry 36:13

Yeah.

Callum 36:14

Prompt libraries that they should use for each task. Once you set up the system and set up an environment that knows your agency and your workflows, AI can teach people how to use AI. And

Pauline Bertry 36:27

Yeah.

Callum 36:28

ironically, that's how I was learning about AI in the beginning. You know, I'd ask, what's the best course I can do? You know, please explain what a recurrent neural network is in more detail and give me analogies and whatnot. You know, AI can teach you how to operate and interact with itself, which is quite profound and amazing once you realize this is true. The only way this is possible though is if you ingest the context. If you ask

Pauline Bertry 36:52

Yeah.

Callum 36:53

what what is this, you know, what is this workflow in my agency, it would say, I have no idea, or it would fabricate some lie because it hasn't been trained on that data. There's a process called RAG, which is retrieval augmented generation. It's basically you introducing new data into AI. So it's trained on Billions and billions of data sets across the the internet, everything it can find, that's how it can speak and interact with people coherently. But it it isn't trained on the data that lives in your head. Your kind of IP, your personal skill and your your knowledge of specific niche workflows inside your business. And so once this environment is set up, and there's a lot about this on my page and on my LinkedIn that I've been talking about recently, but it's for example, a Claude plugin. And this is a system, there's a an application in Claude called Claude Cowork. And the analogy I give to this is chat is your kind of search engine. Cowork is an employee because it learns instead of dragging in PDFs every time you open in a new chat and dragging in new context, you can internalise this knowledge into cowork. And then every time you interact with it, it understands all of the context about your agency. And this just allows you it's AI to work with you rather than against you. It's simply whenever you have any kind of query, the more it knows about your situation, the better it can help. And it's very true of agencies. And in the sense of context, it has a nice advantage. And like I said earlier, it's not that common for knowledge workers in the creative industry to have their processes mapped. you know, twenty steps

Pauline Bertry 38:37

Yeah.

Callum 38:38

in meticulous detail. Here are the output policies, here are the the human verification steps, here's what data we use. It's quite fluid. And I think this is undoubtedly a reason why agencies do so well, because they they offer each employee has a different way of approaching the problem and that allows

Pauline Bertry 38:56

Yeah.

Callum 38:57

for more innovation. It it's very good in this way. But AI doesn't have that. It doesn't have our understanding. And so You almost need to train it like it's an alien. And yeah,

Pauline Bertry 39:07

Yeah. I like it. I like this analogy.

Callum 39:10

completely. It's it doesn't have knowledge of especially knowledge of what doesn't exist. And the reason why I say it's an advantage is because it can force agencies to actually map their processes. And th this isn't a sort of desirable way to spend a Monday afternoon, for example. It's not people want this fancy AI system that looks amazing instead of sitting down at a table together and picking apart their processes. But I think that the only way you can effectively make custom systems out of AI is by mapping the process in brutal detail. And through

Pauline Bertry 39:52

Yeah.

Callum 39:52

my learning at university, there's two main types of data. You have quantitative, which in the AI sense, you would ingest every single file relevant to the task to give AI more context. And this is simply how I'd go about setting up a custom system. So you'd bring in everything, every document that's relevant to the task. And then the thing that people miss is qualitative data. And so that's of course the the nuance that is captured through conversation and people's kind of epistemological viewpoint and

Pauline Bertry 40:26

Yeah.

Callum 40:26

you know their own understanding of a process rather than kind of a a proof of a fax. You know, it's more kind of subjective viewpoints and that can offer a lot of value. And a a very simple use case you can do with AI, if you want to set up a custom system, you start a new chat, get in all of the quantitative context, say, I want to build this system for my agency, but then you can get it to interview you. And there's a lot of transcription tools that you can use. I use granola myself, Notion do it, Gemini do it. And I advise agencies to get together in a big meeting with their team who understand the workflow best. You can ask AI a simple prompt to kick off the interview process. I would like to start this workflow. You've got my files already here. Please ask me, you could say a hundred. It depends how much time you have, really. But you could say

Pauline Bertry 41:15

Yeah.

Callum 41:15

ask me a set number of questions to give AI as much context as possible about this process. And you can then answer those questions in a meeting one by one. with your team members in the agency, that transcript is then recorded. And undoubtedly then certain pain points are going to be brought to the surface and y there'll be a pr a pretty clear step-by-step process, you know, what good looks like, what bad looks like, you know, how long the out how long the workflow should take. Everything possible context-wise that you could understand about a workflow can be answered in that interview process along with files that represent the the workflow. At the end of that you can then go, you've now got all the data. Please, well, make a context file. So MD is usually the the thing that a AI reads most effectively, but it'll have that context in the chat. You can go, please build me a custom GPT, a Gemini Gem, a Claude skill, etc. And then it has that in brutal detail. And then when you're training team members on this Like I said, I use Claud plugins, you can use Claude Cowork and a s skills as they're created can just appear on people's seats automatically, which is great for team adoption. If you have someone struggling with AI, they can load up a system and go. Perhaps it's a process of generating first drafts from research and the the sale the skill or the system, the GPT, whatever platform you're using, they can go into that shared system and go. Can you teach me how this workflow goes? And then it goes, Yeah, absolutely. Here's all the context that you Exactly, yeah. So it it's

Pauline Bertry 42:59

How do I work with you? I love it.

Callum 43:03

it's it can be a very helpful tool for team adoption in the way that as long as leadership of almost injected AI with their skill and expertise and knowledge about their agency, it's that's their knowledge. It's not AI that's teaching them, it's almost the leadership that's teaching them, but through AI where people can ask a thousand questions if they wanted to, or you know, they don't need that founder involvement. That's a massive thing that we design AI systems around. How can we make it so the founder isn't constantly asking or getting bombarded with questions about workflows? A lot of the

Pauline Bertry 43:41

Yeah.

Callum 43:41

time it's because company wikis and agencies are collecting dust in a corner somewhere or in a file that people don't access. When this context and the wiki in this example, which can be one of the quantitative files you use to set up a custom environment. Once it's in AI, people can just log into the AI team plan and go, what you know, that question that they would have asked the founder, you know, what what's allowed in this workflow? You know, what does good outputs look like? And they go, Yeah, here you go, that's all ingested because it's in the environment and therefore it becomes an employee. And it is this is amazing once once it's come to fruition.

Pauline Bertry 44:19

I love it. I mean now it kind of leads me I think to the last to the last question. and I mean I know you already started talking about it, but I want to kind of develop a bit the thinking there. So

Callum 44:33

Course.

Pauline Bertry 44:33

then what's next? Like how will the future for agencies leveraging AI look like? How will an agency employer look like in in the future? What's what's your take on that?

Callum 44:49

I think that this is almost the most important goal for everyone that I work with, and rightfully so. You know, people want to be prepared for what's next. The

Pauline Bertry 44:56

Yeah, of course.

Callum 44:59

honest answer is that we don't know how quickly AI systems are going to evolve. We don't know where the kind of filter is, the great filter on the technological development. There are, of course, ways to prepare for this, though, which I'll discuss. Quite an interesting and morbid viewpoint about AI is that currently it's A it's a form of narrow intelligence. We have to actually give it the impulse and configure the systems for it to think and work. We have a general intelligence as people where we can theoretically learn anything on our own will. We can go and I could

Pauline Bertry 45:36

Yeah.

Callum 45:37

pick up a guitar and just as if I play it for long enough, then I'll figure it out. Watch enough videos and whatnot. But AI doesn't have that autonomy to go and learn stuff itself if it wants to yet. When it does, if it does, hopefully not, because there's this thing called technological singularity, where if AI

Pauline Bertry 45:54

Yeah.

Callum 45:55

has that autonomy, it could develop better AIs. And then the better AI is better at making itself smarter. And that's that kind of loop, we don't know how quick that would happen. It could be overnight because it's it's like an exponential growth of intelligence that Perhaps we couldn't comprehend. And not to try to scare anyone there, but that's almost to frame my point here: is that to future-proof your business and yourself with the future of AI, it goes back entirely to context. If you've saved these context files and workflows, honestly, if the more AI knows about you as a person and your business, the better. Because whenever these new tools develop and Perhaps Claude, because it seems to be dominating everything right now, especially my day to day tasks because it's so amazing. They could r rise their prices, raise them to one thousand pounds a month minimum.

Pauline Bertry 46:58

Yeah.

Callum 46:59

And that's where it gets scary because they've got, you know, they're integrated in that many businesses that a lot of them would just have to take the hit for that. I use an account called GitHub, which is a very popular place to store code and repositories and whatnot. All of the files that I use in my cowork and my Claude code systems are saved in GitHub and this is all the context like what

Pauline Bertry 47:23

Yeah.

Callum 47:23

are my frameworks? You know, who do I work with? Certain facts about my clients to help me with with their workflows, my brand voice. There's about a hundred of these files and and different skills. Say if Claude right raised their prices or they they just weren't as good anymore and Gemini started to take over and there was a new feature on Gemini, the context is the value. Not the system.

Pauline Bertry 47:44

Hm.

Callum 47:45

In the same way that the brain isn't necessarily the value of the person, it's the the skills and the memories within that brain that is the value. And so the

Pauline Bertry 47:58

Yeah.

Callum 47:59

AI, the hardware of AI isn't the value, it's the data and the context. Like I said before, that is that's the way to future-proof your business and yourself. Capture as much context as you can and Don't capture context where it's regulated data, of course, but in general, meeting transcripts, you know, this podcast perhaps you re digest the transcript and go save that somewhere because it might be useful in the

Pauline Bertry 48:24

Yeah.

Callum 48:24

future. You know, there might be something that we can pull out of this conversation as as of course there is because it's been a great conversation. But

Pauline Bertry 48:30

Yeah, definitely.

Callum 48:31

it's a as you go on, that's the main thing, you know, to for agencies to survive, capture as much context as possible. I have a client who

Pauline Bertry 48:39

I love it.

Callum 48:40

Has a subscription to a transcription tool for meeting transcriptions, but the team don't use it. And to me, that's I'm like shocked because I'm thinking you need to start doing that. But people

Pauline Bertry 48:52

Yeah.

Callum 48:53

in their team, they don't know how AI works. So they they couldn't possibly understand that context is the value. But as soon as people realise this, they will go, okay, the more of a digital kind of vault I have of of my knowledge. the better, 'cause you it's for future proof. And as as well as that, I think purely a growth mindset to change. I think ti times change. AI might be old news in ten years. There could be some new hot topic that everyone's using to accelerate their personal lives and business lives. Digital was the big thing in the two thousands where people

Pauline Bertry 49:29

Yeah.

Callum 49:30

were going to laptops, for example, and everyone thought Google would make people dumber because the information's all readily available and they don't have to think and There's implications about that of AI, which of course some of them are true, but it's about understanding that AI, when used correctly, can just greatly help your life in general. To me, it's a way of Enjoying your work more, simply. I think the the

Pauline Bertry 49:54

No.

Callum 49:54

parts of yourself that are naturally competent and you're s you're well suited for certain types of work, which means you do better work, you get the satisfaction from progression, you naturally enjoy that more because you're good at it. Keep that to yourself. Keep keep that to your own inputs and and work. Use AI to speed it up perfectly fine, but Delegate the tasks you don't want to do to AI. I think that's the main message. You know, I'm I'm

Pauline Bertry 50:23

Yeah, I love it.

Callum 50:24

I'm an analytical thinker. I love solving problems. I don't like sitting there and designing things. I'm not a graphic designer by any means. I I actually hate it. And some people some people love it and that's completely okay. Some people might delegate analysis work to AI while they do the thing that they love, which is graphic design. In my situation I'm letting AI design for me and I'm doing the analytical work. You know, that that's the kind of that's the balance, I think. And e even the stuff you're naturally good at and your skill, like we've discussed here, it's it's you can make AI systems effective because it's your knowledge. And so

Pauline Bertry 51:04

Yeah.

Callum 51:05

it's never going to replace you. It's just going to be a tool that helps you do more of the work that you're already good at and avoid the work that you don't want to do.

Pauline Bertry 51:14

Yeah, and I I love I mean I think it's a it's a perfect closing framing, right? Like use AI to be able to to do more work that you love and less things that you don't. I love it. Thank you so much.

Callum 51:29

Yeah, well thank you very much.

Pauline Bertry 51:30

Th Thank you, Kalam. It was great talking to you.

Callum 51:33

Yeah, it's been a great conversation. Yeah, as always. Hope everyone's enjoyed listening if you got this far.

Pauline Bertry 51:39

Yes, yes. Thank you so much. Thank you everyone. Bye bye.

Callum 51:43

Right, sir. Thanks for listening.