We Blew £30K on AI Workflows Trying to Scale Creative Ops

0:00Olly:We probably invested over 20-30 grand of Lucas's time and money into like AI workflows that became obsolete almost overnight because something new came out.

0:15Olly:Welcome back to another episode of D2C Diaries. I think we're on episode 47 now, so coming up to the half century. Big numbers. And today, joined by Sam, COO of Saw Group, our resident AI expert. First time on, no, second time on the pod. Second time, second time, this time in a new studio. Black Friday episode last year? It was a Black Friday episode, yeah. It was in the old studio that time. And it's a nicer setup. You can tell we've got sponsors. Yeah, we've come on a long way from then. both as a podcast and a business I think and like I said Sam's our internal AI expert or the driver of a lot of our process optimization at product level and a lot of the AI work we've been doing in the business and I think over the last sort of two months especially we've had a lot of guests on we've spoken a lot about AI in parts of episodes it felt right to do a full episode dedicated to the topic, not just the tools and not just the tactics, but more like the mental models, the frameworks, the way that we're thinking about AI operationally at a business level.

1:36Olly:And Sam is by far the best person in our business to talk through that. And yeah, I wanted to dedicate that episode to that today. So looking forward to diving in. Being a founder of an e-commerce business is a constant grind, as a lot of our listeners will know. You're the marketer, you're the accountant, you're juggling many, many things. And one of the biggest problems that we see in most e-com brands is cash flow. So let's say you've got an ad campaign that's absolutely crushing, but you don't have the capital to be able to pour more money into it. Or let's say you're about to run out your best-selling product and you needed to restock like yesterday.

2:12Olly:We know so many founders that have tried the traditional route. They've gone to banks and been told to come back in a couple of years when they've got more history. Or they go to traditional investors and they spend more time pitching than actually running the business. They thought there was no other option and that funding always meant giving up control. That's where Wayfly come in. They understand the speed and unique needs of e-commerce. They give you the funding you need based on your business's performance, not endless paperwork. And it's non-dilutive financing, which means you don't have to sell a single share of your company.

2:43Olly:No one wants to give away their equity and you get to keep all of it, which is something you've worked so hard for. Plus to give you more than just cash, to give you access to insights and analytics so you know how to best spend your newfound capital to drive growth. I think I can speak on behalf of Ollie and I here. So many of the brands we have worked with have used Wayflyer at inflection points within their business to help them scale to multiple eight or even nine figures in revenue. It's not just a loan, it's a partner in growth so if you're a founder looking to scale up without giving away equity it's time to check out wayflyer.com in the description below it's a pleasure to be back i see myself right now from a coo perspective almost like if you if you were to picture a venn diagram where on one side you've got cto and the other you've got coo it's like that intersection which i think is really important for a lot of operational leaders right now it's like how can you drive AI in a business through an operational lens and I think when you picture that Venn diagram it does a perfect job of visually telling you where you should sit within an org and that's where we've put a big emphasis so far this year so yeah excited to excited to jump in.

3:56Olly:Yeah I think that's a really interesting way of thinking about it and like visualizing it it's like we've talked a lot about that convergence of agency and SaaS for us and I think you almost can see that across other business models even product businesses I don't think it's going to be rare to see some of those functions that were maybe more say people or roles that were maybe more reserved for SaaS businesses, data, tech hires appear in some of these product businesses over the next 12 to 18 to 24 months like service system SaaS right and I feel like in a pre-AI agentic world it's really difficult to expect that you can take a leap from system to SaaS without significant investment in AI and with agentic AI I feel like we'll be seeing a lot more of that over the next 12-18 months where people can businesses can now take that leap to something that is more agentic because ultimately AI is mimicking human behavior and it's autonomous work done at a fraction of the cost which kind of ties into a few topics that we're going to going to speak about shortly 100 and I recently went into PureSport's office through invitation by Dan to kind of do a bit of a talk on on AI and D2C and I think I think we're entering this new era this like d2c i called it like dc 3.0 where it's like we know that costs are going up we know that competition is increasing we know that gross margin is more important than ever and it's like ai when utilized correctly in a world of lean opex is that enabler for margin expansion it's just like how do you achieve that outcome um yeah i wonder if there's going to be like a a different way to measure this in like the next six months like everyone was so focused on contribution profit but opex doesn't fit into contribution profit and with ai being such a high leverage part of the business to invest is there going to be a different way to to look at that and how that looks from a cost perspective yeah i wonder if it's like functional or at p &l level it'd be interesting to see how that how that looks yeah well in terms of topics um a bit of a guide to the episode we're going to dig into some of the mistakes we've made for with ai like we've been through a real journey over the last 18 months i'd still say we're early in the grand scheme of of where we're going to get to but we've already made some mistakes we've already thought about things wrong we've already invested money wasted thousands tens of thousands of pounds on projects.

6:35Olly:Then I'm going to talk about AI in the org chart, adoption rates, implementation, mental models for how to approach that we think we can apply to any businesses. And Sam's going to deep dive into a specific example of how to use AI context engineering to 10x your creative output with a few different tools and a few different workflows. And then just touch on a few other tools and tactics beyond that, that we think top brands and other businesses can use to drive those OPEX improvements that can unlock significant percentage gains in P &L once implemented correctly. But I wanted to start with those mistakes.

7:13Olly:Yeah, I guess rewinding the clock to early 2024, it'd be good just to chat through our journey with AI implementation in the SOAR group and maybe where we went wrong initially. Yeah. I think the thing to say here is like, nobody knows what they're doing. if someone tells you that they know what they're doing and they're a master at ai they're very wrong and that was very apparent for us in 2024 going into 2025 and some of the problems that we experienced and the things we got wrong are really going to set ourselves up for the coming topics that we're going to discuss but for us it was always like what are the operational levers like how can we drive more output per unit of time which is always the right way to think about improving a system deploying something ai related into a business it's thinking about what are the constraints in the business and how can i solve those with ai which has always been the right approach to to think through these things but there's like a big missing piece to that and it's a lack of appreciation of the things that ai can do at a model level you've got to understand how models work before you start trying to replace human behavior with something like a model or an LLM they're more than capable but it's there's a massive intelligence part missing and we didn't appreciate that enough at the start of our journey, it was very much here's the constraint and like most operations leader will find a constraint and then deploy a process to improve that constraint and that process might typically look like wireframing out what are the problems that we need to solve in this process but right now you as the person deploying that process has the most context but when speaking in an ai language ai doesn't have the context you do so you can't expect to deploy an agentic system without the context and i didn't appreciate that enough when we were first getting started and various consultants as well didn't appreciate that either it was actually i'd say anti what what we should be doing is what we were getting advised at the at the back end of 2024 which we'll we'll touch on but yeah would you would you agree with that what are your thoughts yeah i think um i think we started through that lens with the right like approach to thinking of where do we start to invest where do we start to look at and for us it was all around that creative strategy role right that's our biggest bottleneck and i imagine that's the biggest bottleneck for a lot of for a lot of brands or agencies that's the the individual in our org chart who's the most time constrained who has the most like things on their plate and even if we try and brief in like briefing's a good example i think we've done quite a lot of work there recently around how do we accelerate process there but it's so specialist like you say it's so like um it's very technical a skill set and try to transfer that knowledge and context

10:24Olly:context at a business level or a skill set level of like DR principles, psychology etc but also then at a brand level is a lot and understanding the limitations of the LLM within that and yeah we didn't appreciate that enough. I also think we tried to, even rewinding before that, we tried to like build tools and workflows that were sort of built for the state of AI today rather than the state of where AI was going like we probably invested over over 20 30 grand of Lucas's time and money into like AI workflows that became obsolete almost overnight because something new came out yeah so I think even when you're building with these tools it's like trying to start with fundamentals and building for the rate of progress so you can just plug these these models in to that context to that foundation of data to that foundation of yeah the amount of understanding the amount of people including us back then that forget about fundamentals like fundamentals are called fundamentals for a reason yeah it's so easy to lose sight of them when you've got all these shiny objects in an ai world you've got these you've got tools launching like every single day you've got people on x linkedin launching ai workflows that probably just not doing anything to drive impact i know there's there's certain ones that are that i've seen but the majority aren't and we shared that perspective and it really led to like stripping back and looking at the fundamentals of our business and it's like i know we've got notes on this but it's like business level versus process level and i think when you actually break down the meaning behind that it's like business level is there to create repeatable outcomes at scale it's like what are the processes that we follow to deliver a service to clients to sell product whereas process level i see it more as like operational memory and i think if you had to look at how that applies to human it's like intuition it's like you know from like like failing all the time you know from testing things what what frameworks works and that creates intuition where you know when you test something you've got an understanding of if it's going to do well or not almost like an opportunity score as it relates to a test and it's separation of those two which i think is really important and if you can tie that into agentic ai you're on the right tracks but at the yeah at the start of the year we just didn't appreciate that enough um but yeah for sure like creative strategist role was was a big focus and it's it's it's really funny it's like you you invest to fix the constraint but by improving what you thought was the constraint you create another constraint yeah it's like you could produce 10 times more scripts but can you produce 10 times more ads yeah yeah it's like you fix pre-production but you fix post-production whereas post-production was the original bottleneck and we'll touch more on that shortly it's just that evolving yeah it just all just moves somewhere else it's like you push it's like pushing down the lump it just pops out in another place i think what you said about intuition is is true from a from like that's the hardest thing to train it's the hardest thing to train in individually strategy right it's like learned experience guides that that more than anything like it's what separates like a really experienced growth strategist from someone who's new into the industry it's just seeing it's the it's like seeing more examples knowing how to like look at a business based on say it's a supplement brand based on another 15 you've seen and knowing what leave it to pull next the ai will never know that super hard to do that right now um and that's where a lot of the the i think the the real value when building these tools come from is can you codify that down because that's what makes our business or your business for the people watching like unique is that that that context that that strategy that approach to to growth yeah so it reminds me of what you just said that reminds I was listening to a podcast with the OpenAI chief product officer.

14:36It was on Lenny's podcast and Lenny asked him a question, which was, if you were building an AI right now, what would you be building? And it's pretty much exactly what you just said. Build with the thing that AI will never have access to, which is your business intelligence, it's your data, it's like intuition somehow stored. AI is never going to have access to that. And it feels like a lot of people right now are trying to like design the system without actually understanding the fundamentals of the system. And I think in an AI world without getting philosophical or too deep, it's like you've actually got to understand how the human brain works versus the robot.

15:18Yeah. Like weird as it sounds.

15:21Olly:We're going to come on to that in a second as well. Yeah. I think two final call outs in terms of just mistakes. I think, and I think it's important for anyone to appreciate this is like, a lot of the building of these tools starts top down because obviously you have to make decisions to resource it. You have to make decisions to implement, et cetera, et cetera. But I think you really need to see adoption as a bottom-up exercise and we're going to come on to a little bit about that. And then when we're looking at tools, and I think a lot of brands and businesses are just implementing so many tools, so many process changes.

15:58Olly:I think Alex Amozzi always says this as well but it's like whenever you implement a change to a process or a change in business you have to expect a massive decrease in efficiency for a period of time whilst that gets adopted I love that and that's why we've kind of really focused into this year on doing less better and using less tools really extending and utilizing the most really maximizing the capabilities of that tool before and a lot of it is just using the llms the core llms in different ways rather than try to use every shiny new thing that comes off the shelf because it just creates like a drag on team efficiency over time yeah yeah to homo's point i think the way he articulates it is assume any change will create a 20 decrease so the change that you want to implement needs to have at least the 50 upside chance so and i think this really applies to an ai world like if you want to invest yours and the team's time in creating a tool you think big like this tool that you implement has got to increase throughput drive operational efficiency up yeah opex down nobody thinks through the lens of both operational efficiency and throughput everyone's thinking right now of one of the two yeah but if you drop opex like your throughput's probably not going to increase you need to think through both and that's exactly what homos is trying to say by saying what is the 50 upside like yes opex could go down but can throughput truly yeah get driven up as well or is it constrained somewhere else yeah yeah yeah exactly perfect um well yeah i wanted to chat through off the spring, like piggybacking off that piece on AI adoption.

17:53Olly:I think we've gone on a journey into this year of like really focusing on getting 100 % adoption rates, obviously spearheaded by yourself. And I wanted to just chat about how we've done that, because I think the way we've approached this, I think any business can apply it. I think it's fundamentally quite simple. It just requires an owner and a driver of yourself. I think a lot of this, obviously again has to start top down in terms of like education in terms of leadership and those at the top of the business having the right mindset towards ai and almost like implementing it in sort of hiring processes vetting for people having the right mindset towards it utilizing it understanding it um which obviously we've we've we've implemented but wanted to chat about how we've within the team like really driven that adoption rate up over the last six months six to twelve months yeah this was a massive massive part we didn't appreciate enough last year was any new initiative needs to come from the people that will be practicing that initiative so bottom up adoption has been like tremendous for us once we actually implemented that and you got to understand why like that's important in the first place and the reason why we wanted to do that is like everyone knew what ai was everyone was probably experimenting with it already but it was probably a bit fragmented like everyone was using it but everyone was using it slightly differently and when you've got 50 plus people operating in silos using tools differently you're probably just going to get 50 plus dead ends so bringing that together was super important to us because that's something that i suspected was happening and you like what the way we did this was through firstly talking about it and like almost embedding it within one of our cultural pillars it's like people a cultural pillar is like a it's a behavior it's a shared system of behaviors if you want ai to become a shared system of behaviors talk about it like it's a baseline expectation like toby does with shopify yeah and it creates this forcing function when business objectives are aligned to how you want people to use tools it's like bring people together to understand how a tool works and what we're actually trying to achieve by using the tools and that that is that force function where everyone's doing the same thing and the first session that we did when we actually decided to go with an ai committee and get a group of people together that really have a passion for ai and improving the group through these tools and the first session wasn't even about ai it was literally about theory of constraints and making sure that everyone's aligned on how we should be thinking about this and really solving a problem through ai not trying to create more problems for ourselves like back to the whole mosey quote so that was like a massive part for me personally that I wanted to get across to this get across to the team and another like really like massive benefit of doing something like an AI committee within a business is it's just like compounding knowledge over time like every single session we do knowledge is compounding people are coming to the table with new experiments that they've done as it relates to like the initial business objectives we set people are coming forward people have built some like crazy things which is like so cool to see it's crazy some of the stuff that we've got designers offshore designers coming with like crazy innovations in like design workflows we've got production team building like apps on their phone to mark to building like mobile apps on their phone yeah to mark good shoots so that we can tick it faster in creative strategy it's like like i think getting that innovation bottom up is almost better because they deeply understand the problem in their workflow.

21:54Olly:It's like any process feedback is better coming from the person working it than the person who built it. And it's just been amazing to see that momentum. It's been crazy. Yeah, I think it's just super powerful. Yeah, I like to view something like this as give people a platform where they can explore AI tools, LLMs, at a company level. So we're building knowledge and data that can eventually create operational logic. So by them using the tools, you're almost building a context foundation. It's like, this is what a really good prompt looks like for me. This is what really good system instructions look like.

22:33This is the output that was the best I've ever got. All of that will compound and almost act as that knowledge layer, which is essentially context. That's then going to feed that next evolution of AI implementation within an agency when it gets more towards agentic AI. um so yeah that was that was a big one um a couple of examples josh head of search he's next level when if he's watching this like josh mate you're doing the right things he um last week he ran through so he's got cursor set up on his desktop and he in half an hour he said what should i build for the next ii committee just because he was experimenting with cursor.

23:20And I tasked him with, what was it now? It was basically like a chatbot embedded on a landing page where you could upload a photo and AI would analyze the photo and tell you which product to recommend on an advertorial with the intention to sell the product that we were trying to sell through the advertorial. and he did it in like probably less than half, and I was like 20 minutes, and then he shared that on the committee and did a breakdown of how he did that, which was amazing to see, and that's just one example of many. Like you said, the design team, what's it called now?

24:00It's like N-A-N, but for design. Gumloop, and then there's the other one. I can't remember what it's called now, but yeah, they basically built out a flow that allowed us to iterate on top-performing backdrops of ads and things like that so if something's working really well maybe it's like a a bottle positioned on this table we might change the table to something more engaging and it spat out like 10 variations of that um and this is i i don't think we'd have been in this position without

24:30Olly:that alignment yeah i think that's that's the interesting thing that you said i think it's like aligning it to a business objective codifying it within culture and then and then repeatedly driving communication through something like an AI committee gets the right like people like bought into it and um gets momentum under that I also think like baseline education and people like understanding I think a lot of people a lot of teams probably haven't like understood fully like what an LLM is and like the actual con the constraints like there's a guy called Andre CarpathianXApenAI guy I'll put the link in the description he does like an hour and two hour breakdown of like an LLM I think every team should watch every leader, every manager should circulate that across the team and make sure like the team understands deeply what an LLM model is, its constraints how to utilise it properly and then I think you can really get better output and then I think the AI committee's been a game changer for us, I would recommend that everybody implements that or something similar within their business to drive that.

25:40It's just like if you've got a performance-focused culture and you're always talking about performance on, I don't know, business reviews that you might do every month, it's like treat AI the same because when everyone's communicating about the same thing, it signals that this is a fundamental part of our culture and this is where we're moving. It's like aligning people on the expectation that if they're not willing to change with AI, it's like this is probably not the right business for them because this business in three years is probably going to look vastly different to what it does now. If people can't go on that journey, they might not be right for the group.

26:15They might not be right for you. I think that's important as well. And touching on hiring as well. It's like, I always try to reserve five to 10 minutes for every single person that we have an interview with just to talk about AI. It's like, how are they using it in their workflow right now? I always like to ask the question, what do they think AI will replace them? Yeah, I see what they say. Yeah, just their perspective on that is really interesting. You can kind of understand how deep someone is in AI from asking that question. So I would say it's become part of our hiring criteria as well.

26:52Olly:It's been codified in our way of bringing people into the business for sure. Yeah. Obviously, we're talking about it at service delivery level. We're also, I think, even on our marketing output, so many different areas to utilize it. and I think that memo from Toby is the perfect springboard for like a communication session around it I saw him repost something that someone had built internally at Shopify off the back of that memo I'm going to butcher the explanation but it was like an analysis tool that someone had just coded and it had just, yeah, I can't remember but have a look on his Twitter feed it was crazy that someone had just built this in the spare time as part of their daily work but similar to obviously our use cases are a lot like smaller maybe not quite as impactful as what that individual built but we've seen that behavior from going on a similar journey yeah there's compounds over time coding's a really interesting one like the amount that's been disrupted is super super interesting to me and you look at tools like lovable which have gone from zero to probably a billion in valuation in the space of one of the fastest growing companies yeah yeah in like the space of 12 months could that sustain it is it just a fancy wrapper with yeah a really good prompts built in we don't know but yeah that's that's really interesting um but yeah i think i think any business with like over five people should get together it doesn't maybe need to become in the form of a committee but just at least dedicate time per week to talking about it and really aligning it with business objectives like if you can get people bought into the theory of constraints like what is a constraint and how we should be using ai to solve that i feel like it's just an absolute no-brainer to do right now 100 um and consistency on top of that is yeah don't let that fall by the wayside because it'll stack up and it'll compound over time yeah kind of wanted to switch gears so we've got we've got a really solid adoption i wanted to chat a bit about how we're thinking about AI within our business to set ourselves up for success longer term and the learnings we took from those previous mistakes.

29:06Olly:And I love the quote of like, I think this is all anchored around that concept of build for the slope, not the intercept. And I think that's like the fundamental quote that we have kind of sort of anchored ourselves to in terms of our approach to building AI systems, processes and tools longer term building not for where we are today so chat gpt5 but building for where that's going to be by the time you even finish that project because it's easy to lucas's research phase they knitted together with a lot of nann workflows that then became obsolete overnight when they rolled out chat gpt3 is a perfect example of how you can invest thousands of pounds of time and money and just it all be pointless um if you're not careful um so yeah i wanted to to to get your view on and on how how how are you approaching this within within the org today and where do you think where do you think brands should start or businesses should start yeah i love intercept versus slow thinking i think to like really crystallize this for viewers like intercept thinking it's like chasing quick wins it's like that initial bump and momentum that you'll see from implementing a tool but that decline that you'll see three months later from something that gets released and short-term increases don't compound like a slope does which is sustained growth and thinking with the slope is always about investing in infrastructure and the context layer that ai will never have access to so as models get better and better how can you best set yourself up to give models that data for them to perform better and better it's like it's like if you were to task a new person that joins the business with creating a winning ad they would really struggle because they don't know anything about the brand yeah and that's what intercept thinking is it's like tasking ai with something that it knows how to do because it's intelligent it's like got the same intelligence level as someone from harder with a load of degrees but it's like do they understand the context specialization isn't it i guess yeah specialization business intelligence intuition like intuition like can you can you systemize the collection of intuition i think is a really interesting thought exercise and that will lead you down a path of not really even using ai like if you if you want to build for the slope you don't actually need to use ai and you don't need to use ai to the level that you think you do it's like if you're thinking with the intercept you are using ai to build this flashy tool that's going to create a small bump but when you build for the slope it's like infrastructure context so you create a foundation that you can then build on top of it's like it's like if you were to implement a training program in your business but without any training it's exactly like that you're just not going to get very far so yeah I think that's like probably the fundamental aspect like the top line of how businesses should be thinking with AI it all starts with that data warehousing right data is the foundation that data layer and vector database data pipelines and how you approach warehousing that context yeah what's your thoughts on the approach to that it links to what we touched on earlier which is understanding the context layers and data layers of a business so you've got like macro level and local level or you've got business level and process level like you said macro level i see as creating like a spine of context that ai could query to do a successful retrieval so i see macro level as repeatable processes what drives outcomes how do we conduct business what are our what is our process to conduct an employee performance review for example whereas local level like process level is that operational memory layer which is that intuitive like intuitive layer where it's like this is what good looks like this is what bad looks like here are all the frameworks that we're testing here are all the prompts that we've tested here are all the good outputs that we've got so it's not driving repeatable outcomes but it's storing memory yeah and i think it's important to differentiate between the two and before you even create a data warehouse you don't actually need a data warehouse you first need to create a system to collect context and data so right now like uh on the from a creative strategy perspective we're doing a really good job here which is storing like all winning ads and a breakdown that uh gemini does um look at studio does sorry google ai studio does um breaking down like why that ad worked what was its framework what the hooks what were the visuals and you're pairing like cold hard readable language that an AI model will pick up with the actual visual yeah that can be queried by like computer vision models further in the past and that's that operational memory whereas macro level SOP level how do we conduct business most businesses should have a lot of that already set up already so it's like how do you make it readable to an AI like if you were to upload that in a project in chat gbt would it be able to recite exactly what you need to do if so yeah you've got a good you've got a good macro level like foundation there and i think that's the thought exercise for a lot of businesses now we've got a lot of our processes down but it's like almost code codifying like your philosophy like i say philosophy but i don't know if that's the right word like your like ideology around like how you produce the outcome that you say that your mission like how do we produce the outcomes for our clients is like kind of like stored in our head still it's like that's the next phase is like really codifying our manifesto for for for delivery and like and yeah so i think like that business level piece i think is really interesting i think you can even like call notes direct report note direct report notes employee reviews like all of that stuff is valuable valuable data as you compound it over time it's just like setting up a process for collecting it which I just don't think a lot of businesses are thinking about it just sounds so basic doesn't it just make sure you actually store these things have a google drive set up with like clear folder structures that you can eventually get all of it out and then ingest to AI and it's ridiculously simple and this is why fundamentals just aren't sexy.

36:19This is why nothing you see on LinkedIn or X is about what I've just said. It's just not sexy, but that's the way to build with AI.

36:27Olly:100%. And that creative strategy process, I think, is super interesting. I think that's been... What we're doing there is we've got an automated process where we find a winner in an account, it pings into Gemini or Google AI Studio. We produce a hypothesis and a framework off that top spender that we store in a database. which is codified by niche, vertical, persona, gender, age and that's just instrumental in time to value for clients because we can onboard someone and be like right they're in this category, they're in this demographic here's 15 frameworks that we've found to work over the last six months Yeah, and you can see where it's going like if you've got this structured database if a human can query it with logic and some form of relational understanding of how a concept or a framework links to the brand, an AI can do that too.

37:23So it's setting yourself up for that eventual querying of a database. And that's just one aspect of that. What if you had a database for creators, for scripting, for copywriting styles, for visuals?

37:39Olly:It's like specialization within that approach. Yeah. Yeah. Yeah, I think that's what an agentic future looks like. It's like a corpus of business intelligence that multiple AI agents can query, bring them all together, and then produce the outcome that you want, which makes me really excited. Being a founder of an e-commerce business is a constant grind, as a lot of our listeners will know. You're the marketer, you're the accountant, you're juggling many, many things. And one of the biggest problems that we see in most e-com brands is cash flow. So let's say you've got an ad campaign that's absolutely crushing, but you don't have the capital to be able to pour more money into it.

38:21Olly:Or let's say you're about to run out your best-selling product and you needed to restock like yesterday. We know so many founders that have tried the traditional route. They've gone to banks and been told to come back in a couple of years when they've got more history. Or they go to traditional investors and they spend more time pitching than actually running the business. They thought there was no other option and that funding always meant giving up control. That's where Wayfly come in. They understand the speed and unique needs of e-commerce. They give you the funding you need based on your business's performance, not endless paperwork.

38:51Olly:And it's non-dilutive financing, which means you don't have to sell a single share of your company. No one wants to give away their equity, and you get to keep all of it, which is something you've worked so hard for. Plus, they give you more than just cash. They give you access to insights and analytics, so you know how to best spend your newfound capital to drive growth. I think I can speak on behalf of Ollie and I here. so many of the brands we have worked with have used Wayflyer at inflection points within their business to help them scale to multiple eight or even nine figures in revenue. It's not just a loan, it's a partner in growth.

39:24Olly:So if you're a founder looking to scale up without giving away equity, it's time to check out wayflyer.com in the description below. I wanted to, I think this leads nicely into this talk you did, which I wanted to recap on here with he obviously delivered it with Triple Whale a couple of weeks back with the title of how to 10x your creative strategy with AI context engineering and had loads of feedback, three or four people on WhatsApp messaging me saying you dropped too much sauce on this one so wanted to go through like a bit of a just chat through I guess we're not going to have time to go through the full deck maybe we can give that away below maybe comment common ai and we'll send you the deck yeah um but yeah it'd be good to just chat through at a high level like the the thought process that's that sort of flowed through that session yeah yeah so again linking back to like a lot of things we've been speaking about already which is context engineering this is why nan workflows that you see on x linkedin don't produce a really good output is because of context so i try to approach this through the lens of well already knowing that ai can do 90 of pre-production work scripting etc but everyone is that a lot of people are at 50 realization of that because they don't understand how to pull in context in a scalable way that's another key part of this like how can we create a workflow that is automated but can also pull in context like i think very very quickly i think that's the understanding the limitations of where ai can go wrong as well isn't it because if you provide an ai model with too much information it finds it really hard to query it in a way in a structured manner yeah yeah so like what what is context engineering it's it's essentially like structuring information tools and data to give an llm an exact understanding of how to perform a task to a high standard that's what context engineering is and people get that mixed up with like prompt engineering which is still important but a 10 out of 10 prompt without the context is still like a 5 out of 10 response so yeah that was like the core like the principle behind creating this this workflow um that the team are now using that we're getting some the outputs from this are like really really impressive i even showed it to lucas and he was like i was pretty bearish on this but then he saw the outcome and he was like wow that's pretty good um so yeah i wanted to create briefs uh ugc briefs specifically in lightning speed using hundreds of thousands of data points pulled from various sources first party data like ad account data and then third party data like reddit trust pilot and i know we've gone into this before in previous episodes on the podcast but we heavily rely on things like trust pilot so i think the way to look at this is like what are the customer's saying and what are potential customers saying yeah potential customers being people suffering from pain points on platforms like reddit whereas your customers they're probably already expressing how your product has fixed their pain point so really high quality data sets there and what a human would do there is look on reddit probably spend like an hour and a half looking through all of these communities threads comments etc same with trust pilot which is going to take a lot of time so i was really really i really really wanted to automate that part and the other aspect of that is like creative analytics so what are all the data points associated with creative and why is this performed and when you look at so all of this came from looking at how a human does that work first breaking how a human would do that down and when a human analyzes how creative is performed you look at the analytics but you pair that with you pair analytics with the visual it's like you are watching a video you are comparing the first three seconds that you saw to the hook rate how can you mimic that in an ai flow and you send it to ai studio you break that down exactly second by second the visual what was said with the data you pair that together in a workflow which we use triple well ai agents for which is like so easy to set up the hardest part of that is just getting the prompt right and asking it understanding what you want to do but before you even approach something like this understand how the human does it first because it's like i was saying before it's like ai mimics humans so get it to understand how a human performs that work first i think you also when you're doing that you should you should like take it to as granular as possible it's like take it one layer deeper than you would if you were building a process for for people to follow yeah so it's like what you said there about taking it to like a second by second play it's like don't just say you analyze a hook you analyze it like frame like just frame by frame yeah go one go one level deeper than maybe you would think you need to because i think it helps you produce a better output yeah i think if you also look at the creative workflow for a lot of strategists it's it's like what's that thing that sparks inspiration and that typically might come from competitor research going through platforms like foreplay and seeing what competitors are doing so you might have a really good understanding of what what great looks like what does that framework look like that we can draw inspiration from so that was another key part of this workflow as well so it was marrying the inspiration with i like to view it as like the driver and the vehicle so it's like the format and the winning message yeah the winning message ai has collected because you've asked it to break it down frame by frame marry that with the data so you know why something works can you pair that with a new format a new vehicle to continuously i net new impressions is like the goal so it's like get that winning message paired with something that also has high likelihood of being able to get pushed out to new audiences and that was done in in triple whale and that was pretty easy to set up we've actually done a breakdown of this which we'll give away if you comment ai and you'll see how all of that's structured but that's only that that's where most people stop yeah it's just data and inspiration where we wanted to take it to that next level was reddit and trust pilot and the easiest way to do this in a conversational way is through claude called desktop it's like one of the only, well it's a really powerful open source model.

46:13When I say open source it allows you to customize the functionality of it essentially and when you, if anyone knows about MCP they will know that MCP allows you to communicate with tools, third-party tools. It can conduct essentially like API calls through natural language which is super cool. I actually think that's like the future of

46:38Olly:I think that's like the thing that's it's getting more popular but it's it's it's it's been spoken about more but I think it's arguably one of the most powerful things if if you utilize it well yeah it's so powerful and um what we do with Claude is essentially feed it the concept that triple whale produced get it to analyze the concept and understand what are those like emotional anchors that we're really trying to tap into here what are the pain points and Claude will detect what that is based on that produce keywords that it can then scrape reddit with and it will scrape like probably equivalent to a book of data pull that into the workflow analyze all of that and then create the script off the back of the reddit data it will then cross-reference it with trust pilot data to make sure that you're saying the right that reddit is saying the right things that we know will resonate with the customer and the output is that script that is essentially a combination of hundreds of thousands of data points the equivalent of a buck of high quality data pulled from people that are talking about a problem yeah and it took seven minutes so and it cost less than like 60p so that would have taken me personally probably about two weeks yeah to do the full thing yeah and it was seven minutes and yeah the output was like so good and it's easy with that mcp to see how you can start them plugging in other tools and you extend that cross-reference you extend that process you just build off that process to it becomes more refined you may be plugging more brand relevant context like historical but top performers by persona angle framework concepts etc the limitation with claude right now is the context window it's terrible it's like you can't get a contact window longer than 500 000 tokens which is like not that much really yeah it's probably like what 25 30 pages of a book so if you implement this you'll have to do a lot of trial and error with the context window and make sure that what claude is retrieving doesn't actually break that so that's a key limitation in ai right now and i think once any ai provider finds a solution to that like will be like there'll be a massive step change in potential through mcp and that's super exciting um but yeah you can do so much with with mcp right now i've got it hooked up to my watch my garmin and this tells me all of my health data it'll tell me what my heart rate is right now everything and i also have it hooked up to my calendar slack messages gmail and i just just experimented one day like when am i most stressed throughout the week based on my stress data broken down by the hour compare that to the slack messages i was sending and the events that i had on that day and it just produced this like report stress analysis report when you're on certain calls yeah yeah leadership leadership calls um yeah it's like that that i think mc i agree with you i think mcp is the future of like just extending what's possible and making those workflow so much more dynamic being able to pull in so many different tools different um yeah and i think context the context piece is that i think that goes back to that that that why i recommended that people share that andre video around is because i i think understanding context windows that that theory of it being like a piece of paper and like that's what you can retrieve and anything off anything that falls off that is is unretrievable and unusable yeah i think that is the key one of the one of the core limitations in any in any process build right now it's just having it's not even like the volume of context it's also being able to retrieve from that context effectively based on what was entered like it has a bit of a recency bias i think yeah yeah and etc yeah i think there's still work to do around i know there's like some good advancements with like relational databases like neural networks and how they're kind of plugged into retrieval which is super interesting i don't think we're quite there yet but that's exciting and it kind of links to what we said earlier if you want to build with ai you probably shouldn't be hiring an engineer that will create some form of product that allows ai to communicate with tools because as we saw six months later mc pre protocol got released and that allowed you to do that so yeah again it just comes back down to not building against ai but setting up the the foundations to eventually allow AI to query effectively, call tools effectively.

51:40So again, if you're creating a process and it's clearly detailed in the process, this is what tool we go to do X, Y, and Z. Here's where we store that. Eventually AI will have access to do that through MCP.

51:52Olly:Just be able to plug into an ocean and just go wild. Yeah, it will be able to. It's almost there to be fair. Watch all our Loom videos and everything. Yeah, yeah. I wanted to, anything else to add on that process? the voice switch? No, I don't think so. I would just really encourage everyone to test with MCP because it's yeah, I just can see it being a massive part of our futures. I was going to say, I just kind of wanted to just touch on a couple of bits to close. Firstly, where do we think for us as an agency, I think first what does the next 6 to 12 months look like or like how are we thinking about that roadmap and how do we how do you think other other service-based businesses should be thinking about that roadmap for sure context collecting yeah like the next i would like you can do this over three to six months go on go a little bit crazy with making sure you can somehow systemize data and context collecting which is like fairly easy through automation nowadays and store that somewhere to then eventually set yourself up for a world where ai can query every single thing that you've got access to and be able to conduct an action yeah off the back of that get everybody on the same subscriptions yeah yeah like same gpt same um record meeting recorder a lot of that is like high impact people just don't even think about it it's like everyone's using 15 different fireflies and it's like maybe centralizing even that context is so valuable yeah yeah it's it's like the basics isn't it and i think three month projects do that and then at the end of that three months i i can see ai looking completely different anyway it's like what what's going to be the next gpt what is the next going to be what's the next agentic claude looking like and you will be in a position where that context window is probably expanded claude and mcp is probably a standardized protocol you can upload that sop to a large language model and actually have it conduct something that you would otherwise do and would have taken you hours to do but i think it's important not to rush to that because it can't do it yet and it especially can't do it unless you are in a position where you've stored all of that context um other things as well like like make sure you've got bottom-up adoption through something tactical like an ai committee yeah like get everyone talking the same language it'll signal that this is a priority incentives as well we obviously did that vo3 day where everybody in the creative strategy team made a vo3 ad we uploaded it to an ad account we've tested it there's a prize off the back of that yeah doing small things like that goes a long way if you want to monitor usage as well you can do the OpenAI Enterprise Plan where it will literally give you an analytics dashboard of how people use sorry I'm dying over here how people and when people are using AI what models they're using what project people have got set up and you'll eventually be in a position where you can share that amongst the team and actually start having like you would track KPIs you will eventually be able to track AI metrics as well AI usage metrics Perfect, yeah and I think for us it's then I think we've talked through a few the way we're thinking about it evolving those processes to make fine tuning the outputs to make it more impactful on I guess output per unit of time, we come back to that as the key measure right how do we drive that up in a in a service-based business um yeah is key just like really understanding what the constraint is as well like i know we've talked about enough today but i just think it's such an important part of like understanding how to use ai to your advantage um theory of constraints the goal that book is amazing to read if you want to start approaching AI implementation within a business because you'll often find that your initial assumption as to what is the thing that we need to fix is not the thing that you need to fix the thing that you need to fix is probably not going to be fixed through AI, it's going to be fixed through something that you and your team get together and put the time into fixing and then you will find that AI can improve a certain metric and you will just improve both throughput and operational efficiency OPEX etc production team is the perfect example of that to be fair it's like we can make as many podcast ad scripts as we want but we're going to have to shoot them somewhere we've got so many people who can do that on every day and all that has done has made your short term OPEX rise because now you're investing in additional production studios people for those production studios so you've got loads of scripts but no ads disaster if you're in an e-com brand I think I'd do a lot of what we've already discussed.

57:10Olly:I think for an e-com brand, you've obviously got slightly different functions in terms of CX, ops, creative, marketing. It's interesting to think where would we start there. I think the fundamentals we've mentioned is key. Any further thoughts on... Nothing on the fundamentals. I think AI is amazing at doing what a human can do at times in it by 100, so analysis. It's amazing. I do very little analysis now, to be honest. I don't know if that's a bad thing, but AI does it way better than I would be able to do, and it picks up on things that I can't see, which is why I think analysis should be a key part of that.

57:57There's things like triple whale AI agents that you can use that have helped me personally a lot um research i think is like a massive one if you haven't had ai do like a massive deep research pulling in various data points from reddit trust pilot amazon reviews all of these data points and producing like a 10 page report on your brand like do that because you'll get so many insights from that that within that might find your next winning ad that could like scale your business exponentially um those are like two immediates that whole

58:34Olly:customer understanding piece is just so valuable it uncovers so so much yeah understanding your customer like post-purchase surveys as well if you can somehow like systemize the collection of that data maybe get that data into a sheet have ai analyze that every week by doing something simple like uploading it to an llm or something more sophisticated built out in a workflow again producing insights from that every single week i think is where i would be focusing my time if i was a if i was a brand right now like analysis and insights at scale to give you the time back to do what you're gonna yeah automated through i also think that keeps on top of like sentiment changes if you're a big brand like getting ahead of those for sure yeah another one that you can do this through Claude MCP but if you can have AI scrape and watch what is going on on TikTok organic every week changes in top viewed videos week on week there's so many golden nuggets in that that you'll just take away and produce a winning ad off the back of one of our top spending ads for our biggest client was that exact process there's so much value in that um we've got that automation setup where every time there's a viral like pretty much every time there's a viral video in certain categories it just gets pushed into slack which is just an easy yeah when wanted to do just to finish because we're nearly at time um alex cooper recently released on twitter like this uh tiered list for ai tools um his was very creative focused so it was it's that classic tiers of maybe we could put it up on screen here um no pressure editors uh it's like s tier a tier b tier c tier d tier of like ai tools um don't want to go through it all but i wanted to focus on that s tier in today's landscape like and i think it's interesting to even look at this personally and professionally what are we putting in that in that S tier maybe we can go back to back I'll kick off because I'm a big he has this in there as well but I think Reddit Answers is just a huge huge one I love a bit of Reddit Answers myself for all sorts of reasons just makes it so easy to do Reddit research probably start simple with my so my favorite I've got two favorite like two favorite LLMs like Claude is amazing at creative writing and agentic tool calling whereas GPT is like your everyday LLM that's pretty much like super versatile I'll give two there but GPT 5 hasn't been that great yet I think it's been limited though I think they're rolling out updates at the moment I think that they wanted to do that to save well they needed to save costs on usage so GPT will call a model based on how effectively it's going to respond to a prompt whereas some people like me i just constantly use o3 which is like the most powerful model just like just probably doing sam altman's heading yeah i'm a big cloud advocate as well that would be that would be top of up there with mine i think um gemini studio um and vo vo3 they've just recently they've just updated that over the last seven days to a to a new model i'm sure lucas will want to be talking about that on one of the next episodes, but it's scary how impactful that's going to be over the next 12 months.

1:02:11Yeah, I would echo Google AI Studio just from a computer vision perspective that can break down a video. So I've got into the habit recently of just uploading Looms to it and it's creating a process off the back of that. This is a really easy way to create the context layer as well. Just record yourself step-by-step going through everything you do and send that off to AI to eventually create that documentation for it.

1:02:36Olly:Yeah. Bit of a different one I like, Fixer, which is the, it's like a personal productivity email tool growing at a crazy rate right now. UK based, London based, I believe. I don't find that super useful for just inbox management and general productivity. Real big fan. I haven't used it yet, you know. I need to get on it. I need to get on it. um i was going to say oh sorry yeah nann i would say is probably s tier just purely from an ops perspective if you need something to be done repeatable and like a conversational llm like you can create a workflow for that with a trigger have it do it every day every seven days the example with that was like the tiktok scraper get it to pull insights in from tiktok that's not really possible without something like NAN.

1:03:26Olly:Yeah, I think NAN is also interesting because you see everyone posting these mad workflows and it's like a lot of the benefits of NAN are like the two three-step processes that are just done on autopilot. It's like how you'd use Zapier but with a little bit more like, a little bit more of like a human touch, I guess. Yeah, and just the ability to like embed AI within that if you want to. you so like our example as to why it's probably s tier for me is client onboarding yeah how before a client has even had the kickoff call with us we've already scraped the full website we understand all the product range we understand what the customers are saying what reddit are saying according to our pain points the competitors on amazon before we've even taken the kickoff call and that's not really possible without yeah without any and that's like a really simple flow it might sound complex but that's not many nodes yeah whisper flow would be my next one alex cooper put me on that one um and he said it would change my life and it is it is a game changer like if you're just the ability to use natural language to add context and prompt if you're doing long prompts in a way that's like super it's just it just it's just very accurate and it allows you to correct yourself etc it's so valuable yeah yeah i agree um with that can because i've just started using it can you say to it like put something in brackets or yeah you can you can say like bracket and you can also say actually didn't like you can kind of correct yourself and it'll remove that delete the last thing i said yeah yeah it's just like really really precise and really really strong I would say apart from that I'm probably finding things to add to the list like it's really really simple you've obviously got your core tool stack like notion the means to collect and store the data which is important for us but aside from that AAA Agents is the last one Triple Whale would be the last one for me obviously we do a lot of marketing with Triple Whale everyone knows we're big fans of Triple Whale but the reason for me is not like just the dashboard views the analytics like they're great and i know this is where their business is going as well it's their context window for a brand it's like they've focused on the fundamentals like we've discussed here at brand level through becoming the data warehouse for a d2c brand and it's like the ability to then use to have so much availability of data and be able to build ai agents off that and be able to call so many different inputs.