Claude Mastermind: Building a Creative Strategy System That Powers 6K Ads Per Month

Today on Dead Sea Dio is we're doing a mastermind breaking down how we are using code right now across creative strategy and this is by far our most asked for mastermind. Now as we know the AI ecosystem has fundamentally shifted in the last six months and a lot of people haven't necessarily caught up. Platforms are very much racing to commoditize execution. Now what that essentially means is there needs to be more focus on strategic death. You know it's that ability to take a signal and turn it into a creative decision faster than ever. You know the ability to create operating systems, how to build processes to log context, how to best utilize data or or store it within a warehouse to be able to speak to it. The ability to actually use these tools to increase high-level strategic activity. So the problem is most people jump into using called code or they build pointless web apps that really don't have a business case right. They are just you know flashy and flashy features or flashy tools that aren't really actually great to use or don't have any utility. They also don't understand maybe the architecture, how to actually build or use called skills correctly, how to enrich brand or domain context. So today we are walking through the end-to-end process of all of this. So we're going to be looking at why context is the entire game right now and why your prompts matter less than you think. How to build code skills that compress your teams, best thinking into systems that can actually run and drive better results for your clients as well. So I'm going to be showing you exactly under the hood how we build our script writing skills as well as reporting and script QC skills as well. We're also going to talk about what unlocks when you stack skills and context fundamentals with called code. So that's the agentic layer where things stop being chat and start being real infrastructure that can compound at scale. We're going to look at how structuring AI inside of creative strategy department looks right now and also the part I'm really excited about is looking at the bespoke operating system that we've built for my brand and also that we are integrating into our agency and all of our brand accounts. This is a compounding self-learning intelligence layer that started as a junior strategist but through giving feedback to it and also implementing systems that allow us to easily log reasoning traces and strategic decisions. It now acts like a senior strategist or a head of performance creative. So we're going to break down the architecture of this and how it actually works and how you can utilize things like reasoning traces to compound AI's learnings with time. So whether you are new to your journey using ClawD you think you're advanced we will cover the end-to-end fundamentals in today's session to help you succeed in the next chapter of Econ. So without further ado let's get in. Now we create over 6,000 ads a month across high production, UGC and static. So utilizing these tools effectively has become even more important not only to ensure that we can kind of remain efficient and reduce our cost per creative but also to ensure that we are using these systems in a way that can actually improve the performance and the output of the stuff that we are doing and also as I like to describe it we break tasks now into points of leverage right so we've got low medium and high leverage. So maybe like static scripting once you've got your concepts or you know building the brief of a static ad is pretty low leverage now right now that isn't me saying static is not important but if you've provided the right inputs and the system has the right context to be able to do that as effectively as you that then means that you as a creative strategist have so much more time to focus on the high leverage stuff which is things like research strategy etc etc so that is why utilizing these tools effectively is not only important but it's essential because if you don't you will be left behind. Quick question how many tabs do you have open right now to manage your business expenses? There's your bank your expense tracker your other counts spreadsheet maybe even an AI companion to help you make sense of the chaos. If that sounds familiar in card has the solution in card is the financial platform built for how modern businesses actually operate multi currency accounts corporate cards and a fully connected banking experience all in one place it replaces chaos with one connected system and I mean actually connected Shopify zero matter all in one view all in one dashboard you earn cashback on the things that you're actually spending money on thinks sass travel ad spend the exact cost that eat into your margin every month and brands like azia beauty and alo Paris are already using it to scale smarter so check out in card through the link in the description below and back to the episode. Now we've gone through this kind of pretty huge shift over the last kind of six to 12 months and this shift is very much around the idea that AI is soon going to commoditize execution and that's kind of coming into the stuff that I just mentioned then and replacing those low leverage tasks with high leverage thinking is going to become more and more important now we run a service based business this is going to become less about providing a service to a client and you know in return for some you know for a retainer fee it's going to be more about outcome based pricing so people are going to be paying for an outcome and if they're paying for an outcome we need to be a strategic partner not an execution partner so that's why the the focus is going to become more and things like strategy, QC and prioritization. So what you'll know by the end why our context forms the basis of using Claude effectively how to create Claude skills and use them to accelerate your work and the capabilities of Claude code and Claude skills combined so we're going to actually run you through and you know what we are doing from an operating system standpoint once we have the foundations and basics of not only context but also skills and agents also how we're structuring AI within creative strategy so how we're thinking about this and then function builds that live inside the brain so some of those bills we're going to be going through as well with you. Now I think some of the mistakes made and I touched on this at the start is like diving into things like code and you know too quickly and without understanding the fundamentals because one if you're building architecture in the wrong way it is likely going to break but also style over substance what is the business case what is the KPI that we are attaching to this feature or this build that is going to either drive better performance or build efficiencies with our process so it's important that you are tracking improvements against the KPI or benchmark when you build something how we tracking the performance of that build. So before we get into any of this the absolute fundamentals or basics is context you know is context engineering now we went through this you know progression of starting with prompt engineering where it's essentially just you inputting a prompt then we move to context engineering which is actually more about speaking to the context and providing enough context of the output is strong enough and and it wide and that's why this is so so important right and the problem with prompt engineering that we see is that it's kind of like in the way I like to describe it is it's like a lecturer within a business school they've never been an entrepreneur themselves they've never actually run a business but they've just studied business now this is what I feel like prompt engineering is when you're just asking the LLM a question and expecting a response they're going to be tapping into their own context that they may have from you know a vast range of domain inputs through through the worldwide web now they don't actually understand the context the domain and business context that we have to be able to get the best possible outcome because ALI without you know you giving it the right inputs is blind to things like form data customer language internal insights nuances etc etc and the gap is you as the individual have the intelligence and it's your responsibility to provide AI with that they've got the processing power we've got the intelligence you can't use either individually we need to compound and bring those two together so the two types of context that you need to be thinking about and it is super important that you are because as AI you know becomes more and more important and integrated into what we do the it becomes even more important that everything within your brain is being downloaded in a way or formed in a way of broken into one of these two categories brand context or domain context now brand context is everything such as the products we sell who we buy for you know it's the you know the brand identity documents to really understand the tone of voices the usp's how we speak to our customers it's the reviews so the live reviews that we are seeing all the time on our website is the top-performing ads the data what is working right now that is all of the context from your brand that makes it bespoke domain context is more from a wider industry perspective perspective or skill point of view. So that could be industry best practices related to in this case creative strategy. It could be how to write incredible scripts. It could be best performing static ads. So that could be if you're an agent, you could have 50 static ads that have worked well across multiple accounts. If you're brand, it could be your top performing 30 static ads in a breakdown of why they weren't similarly with high product, UGC, etc. Now these aren't necessarily these are more kind of broad based pieces of domain context that uncover patterns in those best performance. So rather than like you know brand specific context, which is related to the data, these frameworks are more looking at came here are a lot of the things that are working right now within the market and let's hypothesize why those are working because that is great context that AI can use to enriching enhance its outputs. And the final one could be things like hooks so how to make great hooks if you give domain context to AI in terms of how to write great scripts, but maybe not hooks, then it's not going to really understand the fundamentals of how to really capture attention. So breaking that down and this is an exercise that I recommend you doing if you haven't done already we break context of the same to domain and brand and that it kind of breaks into from a domain perspective external learnings and insights and internal knowledge. So if I don't have the expertise around creative strategy to really train a model or provide a context document on what internal you know how to do that really really well now the good news is you can go to things like YouTube Twitter, LinkedIn, there are endless resources and articles that you can use that is going to help enrich your context on how to do things super super well. So great example is someone like Sarah Levenger on Twitter right she does these incredibly long 2000 word blog posts about hyper specialized parts of creative strategy so one could be how to increase your incremental reach as an example. Now if you take that article and asked ask your LLM to create a context document with that article then it's going to do a really really good job of doing that. Similarly if you take one of our common example actually is is a script we've got script writing context dog now script writing is something that again is very nuanced what works what doesn't now I did a whole mastermind on DC diaries will put the link down below. Inter which breaks down how to write great ads scripts or storytelling and script writing now you could literally take that video transcribe it or take the transcript put it into AI and ask it to turn that into a context document giving it very clear objectives what the idea outcome of that context document is now this is what you need to start doing in every skill or every part of. And you know the domain expertise required to do a job well because that is what is going to drive to better better outcomes now from a brand perspective that could be things like you know downloading your best performing ads and it could be internal s a piece data processes documents now that is the stuff that you already have is is quite easy to find. Now what we used to do is actually organize all of this context into something called a context library database so this is something that we have. We will talk about later we don't require anymore with our new setup in our operating system but I would recommend doing this is an absolute step one because if you've got a document like this that breaks down the different context that you require then what that means is when you're building skills you can cherry pick and tap into different things for different squirt skills so for an example if you're building a script writing skill you know that means we can use the storytelling and script writing 101 context dog. When you're doing a script writing 101 context document while losing scripts don't write script writing theory how to make hook great hooks or hooks 101 we can also pull in the high high prod best performers you do see best performers etc and it will be able to create skills quite easily because you've given it the context to do so the great thing with Claude when you build skills within Claude is that it can actually you can ask it to help you build the logic of what context to tap into. So that's the context to tap into at what stage of the process to avoid something called context blow. Now before you understood the AI needed context you essentially copy and paste brand docs etc etc but now you can productize your expertise and then you can deploy infinitely so that is why I'm super super bullish on logging every decision that I'm making the business the strategic decision because my operating system will start to learn how to make certain decisions which is super super exciting. But let's just take a step back now that we understand context and the importance of context and some of the things that we're going to be building today but how do we get here so it was very much this shift from kind of a chat to agent to some degree which meant that you know we there was accelerated reasoning which essentially is the the the some conscious reason why we make certain decisions coding capabilities now as well. And also a genetic tool calling which I think is really exciting and something that we have all been waiting for for a long long time now what has kind of happened over the last three to six months now as we can see we have introduced chord opus which is a top scoring and a them which on coding and debugging it has a huge context window and if you don't know like how we would describe a context window is like imagine a piece of paper you can only write a certain amount of words on that piece of paper if you write more words on it it won't be able to actually log or remember that so it's essentially the memory in which we can we can hold before we start to blow that memory or start having to compress so the bigger the context window. The more memory it the the LLM will be able to store and as a result the outputs will be much stronger so more cost effective and it has more in better and more intelligent reasoning as well then came called code which was a poorly named terminal based coding agent it understands and writes entire code basis so it can actually you know just by you speaking to it in latin terms it's able to fully manage a terminal based coding agent and as a result it will be able to create code that could translate into the light in the web app etc etc so it's super exciting some of the stuff that it can do and the great thing is it can do all of that while you sleep so if you give it a clear brief of what it needs to do it can go away and do that and buy you know within a few hours you can have a whole system bill which again is is really interesting and called code will tap into things like the context we discussed it will tap into things like skills dependent on what you are trying to get it to do then came called skills which essentially is a folder of repeatable knowledge so you are essentially training hard to be able to do one task really really well with a desired outcome of that task so that could literally be a reporting skills so you could train it to go in and you know actually create a report for you you could get it to write really great scripts and why it's great is because it's super super directional when it uses logic to be able to tap into context at the right time which is you know if you're thinking about like a chat GPT bot what that used you know a custom bot you were just upload loads of context documents but there was no logic in terms of when to tap into those text documents at what time so by doing it this way is much more directional and means that if it needs to tap into more domain context first about okay what is a creative strategist and then tap into how to write a great script then depending on what your input is so let's say it's a UGC script over a high static it won't need to touch any context related to high product static it can just go very focused and directional into UGC context so that's the kind of logic that it will build as a result of the inputs that you provide it. The next is called code code code code work so essentially called code work is a the same as code but it doesn't require necessarily a terminal experience so it is much more user friendly way of actually building something but the difference is it's far less postponed so within called code you can hook it up to something like VS code which will allow you to actually build clear hierarchy and commands that ensures that we are building a fully fledged operating system that can be better utilized externally there's less back and forth though with code work and it is kind of like instructor instructing someone and it can tap into multiple different agents or sub agents at a given time. Now this you know when we go past all of these we are then at a point of a bespoke a gentick OS system or you know, operating system. And what this means is we can start actually building custom solutions for our business that truly, truly learns how best to deliver on an outcome. So that could entail self-learning intelligence. So as a result, you could instruct the system how to do something better as it goes or how to actually, you know, log some log piece of reasoning. So here's what we made a certain strategic decision. And it will be able to log that context. And then as a result, get better as a result, get better after that point. Now, if you were just doing that in code code work, it won't, it won't quite be the same. Now, obviously, you could go back into the same co-work chart and it will have memory, but that memory will eventually diminish over time. And as a result, it won't remember those things that we changed. A one of these systems means that we can actually build a back end data warehouse that logs data, live data all times, context decisions, and it is all centralized within one brain. The agents there and all the skills live on top of that brain. So it can actually tap into that live context at a given point and all of that compounds over time. This isn't an individual skill that does a singular task. It is a holistic system that essentially can, you know, replace doing these things individually across lots of different platforms. So just to get started on called skills before we get into that other stuff, now I'm going to talk through how we actually build our script writing skill as a clear example. Now, a skill is broken into two parts. You have the skill MD, which essentially is like the brain. It's the routing logic. So for example, as I used the, using example earlier, if our input is, you know, we want to create a UGC ad targeting this persona, it will know using routing logic that it should yes, look at fundamental context across creative strategy and script writing, but then it will only tap into context relevant to UGC after that point. So for example, our UGC framework, and then if we said to use the specific persona, it will only pull relevant persona information for that one versus all of our personas because it doesn't need it. And there, you know, if it was tapping into all of our persona context, then it would likely bloat in terms of its output. There's also format detection. So as I say, as it is, this UGC or high broad, so it's very directional in terms of what it uses. There's clear rules and guard rails that we put in with a skill. So for example, you could tell it to, if, you know, when I ask to deploy the skill, it will ask me for three key inputs. And I have to provide it with those inputs. One of those could be a link to a notion document. So we use a notion creative strategy operating system for each brand. So for example, it knows on the back end that as soon as I provide the notion link, it will go into that notion page. It knows where to find the brand DNA for the brand in question. It knows where to look at the latest report. And it does all of that before actually driving its output. So again, it's finding up to date context and it knows where to find it. So we are setting those rules within the skill. In terms of it, the second part of the skill is the reference files, the knowledge. That's the context that we have been discussing and building together. So that's the framework, the templates, the best practices. When you build skills and you've already got these context documents ready to go, you can, we can actually ask for to have a conversation with Lord and say, okay, what, what based on the outcome or the objective of this skill, what context documents or reference files do you need to give the best possible output? And you can actually have a conversation with it to actually build those things properly. So that is super exciting that, you know, even if you don't have that experience or expertise building a skill, it really helps you through it. Like it's super, super user friendly. Now, your stack for, you know, that we have been using as a V1 for a lot of our skills. Now, as I said, we now have an operating system that we use across all of our client accounts, which is a fully bespoke custom intelligence system that is self learning over time. Now, you don't need to do that as a V1. Like a V1 very much should be how do I, how can I start using these tools in a way that's going to drive better outcomes for performance, but then also efficiencies and building a fully fledged all bells and whistle system is not necessary from day one. We've only just started doing that. We've been using skills for over six months and actually it isn't really required until you reach that kind of next level. But the first one I would recommend from a creative strategy perspective is having a, within notion or wherever you can build these things is having a CSOS. So essentially, this is one dashboard on notion that is your single source of truth. So ours looks something like this essentially has right now what the primary growth objectives are. It has up to date blended KPI. So what is the benchmark? How are we performing against that benchmark? That is the north star that whenever we are tapping into these skills, it can go into the CSOS relating to that brand and it already has that context that it requires. We also have all of the strategic reports that we build for our brands. Now we've got a skill that automatically uploads these into this database for each bar brand so that again is automated to some degree, but you can do it manually. But I'd recommend having a database within your CSOS on notion that has those reports baked into it. Because as I said, you can instruct things like the script writing skill to actually go in and read the latest reports that has all of the latest insights. And as a result, in my tweak is output as a result of that. We also have our persona banks. So obviously if you haven't go back and watch our persona mask of mine, but this will include all of our personas and then the micro personas. So if I clicked open on that, it would have a massive breakdown of who that persona is, what their fears, their desires, their motivations are. So again, if we're building a script writing skill and I say let's use P1.1, it just goes into the CSOS, it reads the primary growth objectives, the strategic thesis, it pulls that persona information and as a result can speak on behalf of that persona. It also has things like awareness levels, splits, personas, splits just relating to our creative strategy. But this is the single source of truth. And if you have a notion that keeps this all updated, you can quite easily build skills in a way that can you can set rules that tap into different things when you require them. And that is the single source of truth. Now we're going to provide an action template for this that you can take away or use if you want already building, say operating systems. Now I definitely recommend doing this if you are running creative strategy for yourself or someone else. The other one is framework database. So now we've also got this built into our kind of web app or operating system that automatically pulls frameworks. But if not, one you can build an automation to do this quite easily where it pulls best performers into notion directly or you can just do it manually using a V.A. or something like that. So we did this manually for a long time before it was possible doing it without manual inputs. But essentially this maximise your unfair advantages as a brand or as an agency. You have best performers, you have lots of frameworks that you need to be logging and logging hypothesis of why they weren't. And you need to be logging them by ad type, funnel position, framework and then obviously brand if that's relevant as well. Now why this is super, super important is because you know as soon as for example with a script writing skill, if you've got a whole framework bank of high prod UGC and statics, let's say if you give the input within your skill, okay let's use persona P1.1. We're going to use problem aware in the funnel and I want you to actually go and pick a concept or a framework and give me a hypothesis of why you picked that. It will then go through its flow or its logic flow where it will actually be able to directly pull going to your CSOS. It will be able to read the brand objectives. It will be able to look at the recent report. It will then based on that report go into your framework database and pick a framework that it thinks makes most sense based on where your data is right now. And it will obviously you've given it a funnel position as well so you've given it some constraints. So again really great thing to do if you want to start executing a scale and by your skill utilizing data, it can actually start picking ads that are working. Now just a quick hack to be able to like easily do this, just take your best performer or if you don't have your own best performers and go to ads library or meta and you can sort by highest impression videos. They are usually the best performing videos for your competitors if you go on and impress highest impressions. I've tested this usually is pretty accurate. So that's what you want to to do and then download them. You want to then upload it to Gemini. So Jim and I, can you create the a framework breakdown of this exact script? I want the script breakdown, the visual description, the shot type, audio, etc. I also want you to give me a hypothesis of why this worked and it will give you a hypothesis as well. Now, I also have a gem on Gemini, which has a bit of context on how to best hypothesize. So again, it has some context there on like what to pull, what information to pull, etc. But again, this is a super, super important part of your process. And once you do it this way, which I guess is the more manual approach, is very easy down the lines and then build this into an operating system. Because we had the approach already built out previously, it meant when we then built this into an actual web app, it was quite easy to do that. So what that workflow looks like from a skill building perspective, the skill MD as I said, it's the orchestrator. It defines the faced workflows, the decision points, the input outputs, etc. You've got your reference files. That's the knowledge base, the context. It taps into those at the right time, depending on how we build that skill or what the, the ideal outcome of that skill is. We have the notion integration. So empty piece, you can build these in essentially like API connectors into different tools. It can, you know, we can do the same with ads manager. But you know, you can actually integrate notion and set up a connector in Claude. So just with the notion link, it can go into notion and find what it needs to with the right rules. Then we have quality gates. So we set these for every brand, which essentially is like, you know, we have a QA step, which has a checklist that it kind of, that runs itself through to make sure the output is in line with the standards that we have set the skill. So that's something I would recommend also including in your workflow, flow builder. Cool. Now let's run through the script writing skill and the exact flow of logic that we follow. And also the rules that we set to make this skill really, really strong. So what's too exciting about this? We go into multiple layers, right? We aren't just looking at this from the perspective of, oh, create me a script for this type of person with this angle. And we are have the ability now to go much more in depth and look at psychological briefing as well. So we don't just look at final position, persona, framework, but also looking at things like valence zone and self concept anchor. So as we know, valence zone, zone one to four, they are all different emotional states or psychological states in which we speak. And the reason why we want to diverse by these emotional states. So let's say angry versus scared or disappointed or, you know, empowered. And the reason we want to speak in these different states is because this is another layer that is going to really help drive incremental reach in the add account. Sarah Levenger does a great article on this that we've spoken about in the podcast before. And actually, the issue a lot of the time isn't that you don't have enough volume within the add account, but it's just that you're getting in front of the same people because you're speaking to you're speaking in the same way within your ad creative. And as a result, net is andromeda isn't able to diversify. It isn't able to, you know, pick up on the triggers that you put within your scripts that allows it to put in front of put itself in front of different people. Now, as we know, and dromada is like the smartest librarian in the room. So it doesn't just know every book in the library also knows what is on every page and every book and every word on every single page. So that means it is incredibly talented at being able to get in front of the right person through the likes of, you know, and what we've described previously as soulmate theory, which means there's perfect add out there for every single person. Now, if we're speaking to everyone in the same way, then we speak into everyone you're not speaking to anyone. So that's why layering in things that they don't know is also really important within your scripts. Now, what we did here is Sarah Levenger has a video, sorry, a blog article where she talks about valence zones and also self concept anchors and language intensity as well and the importance of these now. This is a really in depth article and this is a great example of how we've used external domain context and then translated it into a context document for our our own process. So we actually are able to for the AI to go into our reports, understand what's working and not pick up on those signals and then pick a valence zone that it wants to put against a certain ad that we are producing. Now, we also do another thing called valence gap analysis within our reporting, which again is part of the skill right in utilizing matters API where it can actually pull every single ad creative that is live in the ad account that's spent over 500 pounds. It can do an analysis of maybe where the gaps are from a valence perspective, what emotional states we aren't speaking in. And then as a result, it can ensure that we are balancing out our creative strategy when we get into concepting so our scripts are also speaking to those different valence zones, so just wanted to give that I guess ballpark understanding of how that works before we get into I guess the meat of the actual system. Now the outcomes of this is increased blended ad spend so we are seeing better and spend through our creators within the ad account as well as better, you know, heightened hit rate because it is able to be much more dialed in ensuring that every decision that makes is tied back to a hypothesis was tied back to data. So that's really important it also allows our crypt strategist to be much more efficient, so I think there about 20 to 30% more efficient now by using this skill and also advanced psychological death, so I just ran through an example of that in terms of how we can tap into valence zones and speak to different people in different ways because it is able to pick up on those signals and adapt our messaging accordingly, which is super super important. Quick question how many tabs do you have open right now to manage your business expenses there's your bank your expense tracker your other counts spreadsheet maybe even an AI companions help you make sense of the chaos if that sounds familiar in card has the solution in card is the financial platform built for how modern businesses actually operate multi currency accounts corporate cards and a fully connected banking experience all in one place. It replaces chaos with one connected system and I mean actually connected Shopify zero matter all in one view all in one dashboard you earn cash back on the things that you're actually spending money on think sass travel and spend the exact cost that eat into your margin every month and brands like azia beauty and alo Paris are already using it to scale smarter so check out in card through the link in the description below and back to the episode. Cool now that we've gone through that and i'm going to actually just give you a very very top level view of actually what the eight phases are of this strategic report skills you can go. Sorry not really reports script writing skills you can see the process in which we go for a little bit under the hood that you can go away and take some of these elements and fundamentals and build them into your own skills so as I said. The thing you'll need for this is your notion CSOS now as I say that's if you don't necessarily have a custom bill operating system yet and but you know it does exactly the same job essentially as long as you give it the right rules of what to tap into what point reference files so. It will actually that's your context document so it knows when to based on logic when to lean into those reference files at different points inputs so what are the inputs that it requires to know what to go and do because obviously based on certain parts of logic it will only do certain things if you tell it to so for example. If you're creating a u gc you need to tell it you are because otherwise it might it's not going to pull the right context because there's some very directional context that are related to just u gc ads or u gc best performers. Which it will tap into if you tell it that versus looking into high point u gc and static and then as a result of that it bloating its context with a lot of information that it doesn't require the only information it needs should be surface essentially. The outputs so that's what it will send back to us at the end the decision points so there are certain points where it might stop and after decisions we've got a few modes that we're going to go through with you that we set up within this skill we have quick mode and calculated mode quick mode is it just goes through the whole process and when and gives us a script calculated it acts more like a crit strategist maybe you know your. Your assistant right and you're having to sign off at each stage of the process make sure you're happy with what it's doing there are also q get q a gates that happen at the end of the skill which means it is essentially pulling itself for a checklist to make sure the script that comes out of the back end is good enough and as it needs to be to be able to align with our standards right. So the things that it will we will provide it and our final position so is it unaware to most aware, where does it sit persona code so that is you know we just codify our personas because then it will know through the rules built into. skill MD that if I say persona 2.1 then it just goes straight to the CSOS it pulls persona 2.1 it pulls the fears the desires the motivators the breakdown of that persona how we speak to it and it will be able to extract that context to build into the script to ensure that it is really really hyper focused on that because as we know with Andromeda the more specific we are with persona the better that's how we're going to get in front of the right people and as well as I said creative strategy created type so is it UGC high-profile static because again we want to make sure it's getting in front of the right getting the right context from the right back here we also ask the input of sending the CSOS link so then AI knows exactly what brand we're working with and it can go in and find that the relevant information there's also optional inputs like angle so again if we wanted to be more specific about the angle that we wanted it to create a script for we can tell it same with framework is there an exact framework that we wanted to use we can either link it from the sit the framework database or we can give it a concept description of what you're trying to achieve and then obviously the execution mode I mentioned already which is obviously quick or calculated mode here so what happens in phase one once we've actually kind of given the the input so it reads the brand's creative strategy operating system from notion essentially and extracts everything that it requires to build great scripts that are on brand and aligns the data at that given point so what it's going to do is go firstly into the CSOS and the first piece of brand context it requires is who is this brand so we've got brand DNA document so it knows to go into that brand DNA document it extracts who are the brand what they usp's what they're doing and don'ts what's the tone of voice how do we speak to their customers you know the benefits of the product everything it needs to know to get a clear understanding of how to translate that into a script so the first thing it will do is go into that brand DNA document and read that information the second thing it will do is actually go into the strategic report sorry the next thing it will do is actually go into the persona bank so that's why we've told it to persona p1.2 then it just knows it can go directly into the persona bank and pull all the information it needs for p1.2 it can pull the right call fears and motivators etc and the next thing it will do is go into the strategic report database within the CSOS and it will pull the latest report that we built for that brand now why this is great is because it's going to mean the data is super super you know up today and it's pulling the right insights and translating that into our scripts so it's actually using that data to hypothesize why it makes certain decisions with the script and it will tie every decision it makes back to that data now this is a great example of you know why you need context with different parts of the process and we've had to provide reference file that tells the skill how to understand data and benchmarks so understand CTR, CVR, TSR, Holdray, CPAs etc etc. It knows the benchmarks for those obviously the soft spectrum and venture benchmarks it knows if you've got load TSR what the the reasons might be it knows what you can do to enhance that TSR off the back of it so again it not only is reading the data but understanding and articulating what that means in the form of a hypothesis so now it's gone and pulled all of those things the brand DNA document it's pulled the persona code and the breakdown of that persona and then it's also pulled the the strategic report so it's done all of those things and we've provided the brief input to allow it to do so we've decided on a quick or calculated mode so it actually does this one and then obviously pulls those relevant bits accordingly and then it once it's got all of that information and it's it's followed the right parts of logic because it only needs certain pieces of context at certain times to ensure that one it's own narrative and understanding is correct like if you think about yourself as a as a you know creative strategy let's say you try to understand things in different layers in your own head so if you're looking at a a report as an example you're going to look firstly at like the wider branding objectives which is the North Star and then kind of trickle down into the bits that fall below that which are essentially different layers but if you did that in different orders then it would confuse you and it wouldn't make sense so that's why it's important the flow of logic is is key so the face there is at loads and domain context so there's some always loaded reference files that it will go through that are super important for a script writer which is the creative strategy MDs this is our philosophy to script writing how to make great scripts sorry how to you know the fundamentals of creative strategy storytelling fundamentals how how to make great storytelling within ad creative how to craft the right pacing you know how we tap into core human desires we then have funnel positions so it will read all of those funnel positions and what the new answers of those funnel positions are but it will only tap into the the context relevant to the funnel position it needs so if we pick problem aware it will go into that reference file but only read the information it needs from problem aware because otherwise its context is going to blow and it might potentially confuse the output and these calibrate because then it means that you know it's ensuring that it is some of those kind of fundamental domain things that we would always do with our scripts as someone who's a talented script writer it then knows how to do those things at a really good level so we've got past the the phase three which is domain context loading phase for a psychological targeting so valence zones is something that we use as part of our scripting now and approach Sarah Levenger has a really good blog post on this on her twitter and and you know this is a great example of like someone else has written something that's truly valuable that you want to utilize or implement or you know enrich your skills you can just you can go and do that so we did that exactly with this article that Sarah Levenger wrote about valence zone self concept anchors and language intensity so we just pulled that blog post wrote a little bit more context about how it relates to our business and then it created a reference file for that will context document that it could then use in this process so that is like the valence zone is just to give you a very top level view is like there's four kind of core zones and and that those zones tap into different emotional states so actually like you could be running a thousand ads and your incremental reaches declining yet that might not be because you have an ad volume problem it might be because you do have a diversity problem a psychological diversity problem because as you know meta say and dromiter is the smartest librarian in the world which means it doesn't just know every book in the library it knows every page in that book and every word on those pages it comes back to that idea of soulmate theory if we aren't diversifying the emotional targeting within our creative the meta won't have the signals to put that in front of the right individual so it's ensure it's super super important that we are diversifying our valence zones as a result of that so one thing that we also do just so you are clear as well is we actually run reports that are just valence zone analysis and it's the idea of it is a gap analysis where we use meta's API so download all of the ads it will basically identify and analyze those ads in germany to see a key where are the gaps where does this ad place right now in valence and then it will say you've only got three of your 500 ads that sit in zone three negative lower-ausal and then as a result we know that's an opportunity that's a gap that we need to be tapping into with our scripts so this is the process that it will take it through to identify what the psychological targeting of that script is so just to give you like a top level view as I say we've given it the inputs it's gone to the CSOS it's pulled the brand DNA document the personas the strategic report it's then gone through its domain context references to ensure that it understands storytelling and creating strategy and it then goes into its psychological targeting it then goes into the framework database and picks the framework that it think makes the most sense based on what the data is telling us right so that is why you know having a framework database or some form of database that can be tapped into is really important because now it will pick one of our framework databases that aligns most to the brand in which sorry to the ad in which we are trying to create and it will give us a hypothesis of why it picked that framework as well so you know it will match to persona psychology it will also ensure that it is there's the links back to data within the latest strategic report it will then execute on that format so for example if it's a UGC video the last thing it will do is then look at UGC frameworks because it will look at pattern recognition of those so if we have then you know reference individual reference files for UGC you see Hyde Broad and static was breakdown how to make great UTC ads with examples and why what made those ads gray because again that context is the final piece of the puzzle to then execute on the script that we need. So it's got all of that as well and also a, this is where it would tap into hook mastery. So it will actually, you know, understand how to make incredible hooks as well. And then we've also got one that is for losing useful scripts. So you can see there's a lot of logic here and a lot that it goes through before it actually creates the script. And then finally script writing so it will build it out for us in the table as you see here below and then do a quality check off the back of that. So it has five key gates and validation points that link to link to different checks, right? So I won't go too much into this today. But again, really bullish on QA with AI and, you know, linking a certain level of standard that you want to see and ensuring that it is checking itself. Because if it sees something wrong like for example, if it checks itself on hooks goes back to our hook reference files and realizes that hooks aren't strong enough, then essentially it will change that hook accordingly if it doesn't check enough of our hook checklist. Cool. The next one is our reporting skill. So just to go through this, this is what I was talking to you before about how it will weave and I'm not going to go too much into this one, but I want to show you how else you can be building skills without the need of building full end-to-end bespoke operating systems is just by linking matters API. Make sure you're careful in the way that you do this and you get approved for it because I've seen ad accounts get banned if they do this in the wrong way. But essentially what we are able to do just by connecting matters API and creating a plugin which essentially is like a selection of skills that lives within co-work or agents that also have different things that you can do within it and different links to APIs if we require to go into or connect to different tools. Just by us giving a few inputs, so what we do here is we the inputs that for this skill that we need to give it when we call it within co-work is say what the ad account ideas, a link to the CSOS and any relevant context that it might need that like might be specific to the last month. Now you don't have to do that part but we found it's quite valuable to say to the skill yeah this is something we've been testing over the last 30 days because if that is the case then it might start looking for those things in it which I think is always interesting. So it is then able to just from giving it those small inputs and the connection of matters API it goes into the ad account it pulls the top 30 ads within the ad account at that given time it uploads them to Gemini it analyzes all of those creatives to see what's working what's not and hypothesizing that and essentially puts it into our strategic report template. It takes about 30 minutes to end to end and we are going through the process of iterating and improving that right now but it's really really good and it saves huge amounts of time for the creative strategist. This is the example of the psychological heat map that we mentioned where it will actually tell us as you can see this example here there's a lot of ads that aren't being you know there's a lot of parts that aren't being covered within that matrix. So for example zone 3 and zone 4 that's going to tell the creative strategist that there's huge amounts of opportunity in zone 3 and 4 we need to speak to zone 3 and 4 within our ad creative in the next campaign. And the outcome to this we're saving about two days a month per CS which is huge that that could go towards advanced research and strategy which is massively high leverage. Advanced reporting depth so as the time to do things like the gap analysis that we personally wouldn't have time to do and also advanced psychological depth as well. So another example of how you know these skills can be used and the value of these skills. Now the third one is our grading skill so the this again is like I've touched on this briefly already but we essentially are able to grade every script that we do. So for example where I see real value in this one is if you have new creative strategist but join your business or you're trying to get maybe more junior people to start scripting in your business you might have a skill that you've provided them but they don't know what good looks like right they don't know how to QC that properly. So we've created a skill that allows them to QC their own work and why that is great is because it increases the impact of those scripts so it also you know reduces management QC so if you're you know senior strategist or a leader in your business you don't have time to be reviewing every single script so it massively reduces time spent there from a management point of view and I've personally seen that this has really helped improve strategist and upskill them because it teaches them things maybe they didn't see. For example you know we talk about valentzones and speaking to different psychological states. AI is really good when you tell it to to be able to know if you are jumping between different valentzones or speaking in different psychological states so you know it's a great way of you actually not maybe being able to see that as a strategist and it's telling you to and you've learned something off the back of that so it's a great way of really high-tening not only the quality but also ensuring that the strategist is learning something from that. So I'm not going to stay on this for long but just to show you then how that works from the perspective of the grader and it will firstly load all of the context the relevant stuff that that it requires so the script obviously we need to give it the script that it's reviewing where it sits in the final persona code the created type as well we can also share optionally if we want to things like the CSOS so if it wants to get you know a little bit more of an understanding of the brand it can do that but it will the good news is because we're giving it the CSOS if you give it the persona code it's actually going to go in and find the persona code naturally you know review whether we've spoken to that script in the way that the persona in which we're looking to target would so it would identify things like that it would then load the reference files the final positions three dimensions which is the psychological states losing scripts hook master in storytelling fundamentals and then it goes into this weighted funnel essentially where it will have a criteria on 10 different areas and it will score those out of 10 so look at thumbstock power persona recognition as I said it will link it back to the original persona are we super super clear who the persona is here because one thing I've noticed is the scripts or the as that do the best of the ones that very clearly are targeting one persona that comes back to the sole mate theory I spoke about earlier valence zone consistency so always staying consistent to the valence zone storytelling and narrative narrative arc curiosity loop architecture so while we using curiosity loops in our ads language authenticity emotional death funnel alignment so are we depending on the funnel that we've said was we're creating this ad for is that actually true does that translate trust signals so are there enough trust signals within the creative that are going to drive things like CTR and then CTA and conversion architecture so is there clear risk reversals and are we you know super super clear about what the the viewer has to do next to go and buy the product so as you can see it will actually score each of these and give it the hard floor is 8 out of 10 so we it will not be a pass if it's under an 8 out of 10 and but really we want to be looking for a weighted average of over 9 out of 10 to be actually be able to go and say okay yeah go go run this ad now and so that's just something to keep in mind as well so it will actually and score each of those give it a blended score overall and then the great news is it will actually then give you a report and say this is what you need to go and do to improve on this script so like diagnose diagnose prioritize it could rewrite it for you or say to you do you want to rewrite it it will then rescore it for you compare it against the original and then if it needs to repeat it it can so it almost has this perfect feedback loop of based on those inputs it will know where to prioritize changes and why and then it will re run it for you and give you the output as a result so the final output is essentially the improved script and giving clarity of why it made those certain changes so you can learn as well it's not just doing it and not giving you that context so with our army of skills and I guess you know context and understanding of how the fundamentals work that's where you can start looking at building real operating infrastructure in your mix and you know we're doing this right now with you know the with source so we've created a control room which essentially is a it's is where we store all of our data where we warehouse all of our data across our client accounts it is the centralized anchor in which we work from and build infrastructure on top of to be able to achieve our client's goals. So, you know, we can actually build forecasts within this. We can actually build reports. And it has agents that do different things, all because it is layered towards that back-end data warehouse. And the outcome is really positioning us as a core strategic partner. And this is a sticky product that is going to compound with value over time. The other one is I've created a CSOS for a club neuro, which is my brand. So essentially, there is this does lots of different things, this CSOS. It does exactly the same thing as the control room that I mentioned to you before. So it has a fully back-end data warehouse. And you know, the why this is so valuable is because from an architecture standpoint, this is an intelligence engine, right? It becomes my co-pilot. It not only stores data as we go, but it compounds reasoning and self-learning intelligence at scale. So it becomes an incredibly strong creative strategist. And I'll walk you through maybe what that looks like. And so it has a co-pilot built into it, like this, as you can see, so it can speak directly to the data at all times. So I could say to it, what give me a comparison report of Persona 1 and Persona 2? What are we, which of these two we're seeing driving the highest AOV as an example? Or give me some insights in terms of where we're seeing strongest blended soft metrics for each Persona breakdown and what are the opportunities? Where, what were our NCPAs over the last 30 days? I could speak directly to the data and it will be able to consolidate and give me that information back. We've then also got a straight creative strategy tab here. So we've got an overview, which essentially breaks down blended soft metrics. We've also got a creative strategy tab, which essentially is like where we set our creative strategy. So where we set funnel positions for every month, you know, creative typesplits, Persona splits, anglesplits, and the great thing about it is there's a back end reasoning trace log that is attached to it, which means every single decision I make gets logged on the back end. And as a result, it enriches its learning and context as a result of that, which I'll get into in a moment. We've also, as you can see, built the best performance database into this. So whenever a creative hits a thousand pounds and spend minimum, it automatically downloads the ad, it uploads it into the best performance database here. And it will hypothesize why the ad works and it will store the framework. So then similarly, if I'm in co-pilot and want to go into scripting, it will be able to use the scripting agent and it would be it will have all of that context in terms of the frameworks that are working. And also, you know, it will consolidate all of the hypotheses that are at the seas from all the ads that we've created and look for those patterns. And as a result, that is going to enrich the context and drive better outputs on the co-pilot here as well. We also have that for hook. So every hook that exceeds 30% TSR, it will pull that hook, turn it, you know, hypothesize what it works, it will break down what we saw in the visuals, what the hook actually said. We also have a competitors tab, which is an always on spying tool that looks at any new ads that come into a competitors ad library that hit certain level benchmark of impressions. If it does that, then it will pull that framework again into the competitors tab and hypothesize why it works and turn it into a framework. We also have an audience research tab, which does the similar thing where essentially it will be able to scrape reddit all times. I've given it certain keywords to target. And if anything new pops up in reddit, that are opportunities that maybe we have missed that have linked to messaging, angle, persona, after-stay, any of these kind of elements that make up an ad creative, then it will identify that as an opportunity and turn it into a strategic opportunity for me. Reports, which essentially is like a one click strategic report build. So as I've said in the reporting skill, I built that reporting skill, so it was so easy to then build it into this operating system infrastructure. But I guess before we look at any of those, let's just look at the layers that make this up. So essentially a four layer brain that sits in a vector database and super base, so each layer feeds into the next. The copilot is the surface and it doesn't store anything strategic. It queries the warehouse and reads brand context from the file system essentially. So in terms of what that looks like on the bottom there, we've actually got the raw and structure data. So it's connected to things like Shopify, Fabio, Meta ads. It also has actual raw data in terms of context that is related to the brand, the personas, etc. The layer two is the reasoning traces. So essentially we have, and I'll get onto it in a moment, we have different buckets that fit into our reasoning that means that depending on what I'm logging, it will make changes accordingly and as a result in rich, it's understanding of what that needs to do and that will become clear in a moment. Then that feeds into the actual intelligence itself. So okay, reasoning traces, I guess the why, but then what does that translate to in terms of how it does its job better moving forward and that's the compounded learnings. And then finally the function, so that's the features in which everything else runs through. So as I said, like a very top level view, we've got the framework and hook database, the reporting agent, the audience insight agent, the scripting agent and the copilot. All of this is speaking directly to that vector database and super base. And in terms of the reasoning traces, which I think is the most valuable part of this whole thing. And if there's one thing you're going to learn from the kind of operating system part of this talk, which obviously I'm not going to go into like the depth of how to build this on the back end. Comment below if you would want me to, but it would take way too long and I think it would go over a lot of people's heads because it's just very nitpicky essentially. But essentially, there's one thing you're going to take away. It's this idea of reasoning traces. So reasoning traces, as I said, that's the, that's almost the thought that connects itself to a trigger that is a reason why we did something. So subconsciously, you make decisions every day, but actually if you think about the reason why you made that decision, that is going to link to some form of trigger. Maybe it's a prior experience that you had that you know, maybe that's the right thing to do. Maybe it's a piece of data. You know, there's lots of signals in which a reasoning trace can link back to. And the strategic dashboard is essentially where as I say, follow positions, etc. sit, but every time we make a change in that strategic dashboard, then we have to log the reason why. So it pops up with a little box. So let's say we're on that page and it says problem where solution aware, etc. And it is at like 25% each. And I want to increase solution aware. And the reason I want to increase solution aware is because we've had a signal in the out account that suggests there's a lot of people comparing our product to another product. That would tell me that we need to do a better job of comparing us to other brands and why we are better. And so as a result, I'm increasing solution aware in the mix. And it looks says, why do you want to log this reasoning trace? Why do so why do you want to make this change? So it's asking for a decision so it can log that decision. And I give it the reason it then takes that and buckets that into strategic decisions that then uses to enrich its own context and learning over time. So what this is is like training right? Like what we don't do enough of with AI is feedback and training it to become better. Treat it like an employee. 360 feedback loops are really important to enrich the system and its ability to understand how and why we make certain decisions. It's not going to have the nuances of knowledge and context that we might have until we train it to. You wouldn't have a new person join a business and not give them feedback on what they need to do better. Reasoning buckets. So essentially we bucket our reasoning into three different type strategic decisions, hypothesis edits and context rules. So we touched on strategic decisions briefly. I'm setting the strategic dashboard of splits for the next month. It will ask me why I will give it its reasoning. Whereas hypothesis edits, this again is really, really valuable. So as you know, we automate framework. So every time a framework hits a thousand pounds in spend in the add account, it will, you know, log it as a top spender and it will download the add, upload it and turn it into a framework and a hypothesis as to why it worked. Now, when you first start doing that, even if you give it lots of context and information on how to analyze creative, it will actually not do it that well. So what we I've done to enrich that context and make it better what it does is actually read it is actually then I can make edits into that framework. So for example, and I'm just going to show you one now. So here's an example. Let's say this add has just been pulled through. And here's a breakdown of why it works. Here's the script framework. Here's the performance breakdown. Now if I read this performance breakdown, I thought it was a bit generic, it's not gray. I would click edit here, make certain changes. Let's just certainly do that, 'cause it's in it will require reasoning. And then it will actually say before, after, it will say, why are you making these changes? And the reason then we do that is because it will then come back and log it in the hypothesis edits. And then it will be able to hypothesize better next time. So it will enrich its understanding of how to actually read and add and learn what's working and why, what the new months is of that add are. And then it creates this compounding effect. It's like an A player employee when it's got the right guidance and context and it will compound over time. So I didn't go into that load obviously, but I just wanted to give you a little view of how with viewing this and using it in our own way. More will come on our control room, which is even more advanced than my club newer operating system. I've done that just for creative strategy. But if there's one thing you're gonna take from this final part that we've gone through, it is reasoning traces. Because even when you use, there's models like, you've got obviously open claw and these more agentic models that can go off and do their own thing. And you know, there must be a context windows and they're able to restore context and do lots of different things that, that even a skill wouldn't necessarily be able to go and do. The difference is like it still doesn't have the context or understanding of the nuances of you and your life. So this idea of reasoning traces and logging decisions to make the AI better at what it does is super important. You never just create something and it's the end state. And that's the same with skills. The skills are the exact same thing. You need to test those skills and improve them and tweak them and iterate on them just like you do with your ads. So that's the one thing that I would say off the back of that. And then the final one that we are doing as well, which is working really well is a static generator. So again, this feeds would feed into the same operating system it will feed into the same back in data warehouse. And essentially it's able to pick a framework and it's able to go and create those ads accordingly. So it's this system to must produce static ads at scale. And the inputs that this will need is framework. So what is the framework that we are using? The persona because it needs to know specifics about that persona and how to speak to it. We can obviously add any copy if we want to or we can keep it loose. And then essentially on the back end, it goes away. It pulls the relevant context of the persona and the brand DNA. So it knows, okay, here are the fonts, here's the copy, here's what I need to use for this ad. It has the framework. So a template of how to execute that static framework well. It then will take it to Claude, give it the framework template, give it the brand DNA information of your branding question. So it knows, you know, it has the colors, the fonts, et cetera, et cetera. And then it will essentially fuse the brand DNA with the framework. And then you've got a prompt that's ready to go into nano-banana that again, it does on the back end. It goes to nano-banana with the prompt that you give it. It will actually take the product image of your brand. It has your colors, your fonts, your guidelines. It has the copy that it built itself in Claude. And then it will produce that ad for you. So the main thing, the other thing I didn't mention is that, you know, you can pick multiple different frameworks and 10 variations per. So you could just do one with one click, 50 static ads. Now that will take probably 30 to 40 minutes to produce. But then obviously it actually loads it within this board section here, which essentially is just folders of whatever campaign number we're in. So if we're in CM2 for a brand, campaign2, I've told it what campaign we're in, it will load them all within that system. Again, we've got reasoning traces logged into this. So if an ad is really off, I will give it that context of why it's off and it's going to improve itself for next time. So that's just again, I wanted to just give you a very quick rundown of how when you've got the fundamentals in check, how you can then turn that into wider builds and operating systems that can enrich your whole way of working as a business or in this case, creative strategy. So yeah, I hope you enjoyed that today. If you've got any questions, feel free to drop them down below. I know we went through a lot and hopefully I didn't speak too quickly. But yeah, comment down below, Claude, and we will send you those reference files that we mentioned that's going to help you get started using a similar workflow to what we have done with the skills provided here. [MUSIC PLAYING]