Latest / Selling on Giants: The eCommerce Marketplace Podcast / From Prompts to AI Agents: How eCommerce Teams Should Adopt AI | Edwin Choi

Transcript

From Prompts to AI Agents: How eCommerce Teams Should Adopt AI | Edwin Choi · Selling on Giants: The eCommerce Marketplace Podcast

• 0:11
  SPEAKER_00: Welcome to Selling On Giants.
• 0:14
  SPEAKER_00: Most e-commerce teams are still using AI at level one.
• 0:18
  SPEAKER_00: They open ChatGPT, they enter a prompt, and hope the answer is useful.
• 0:24
  SPEAKER_00: But the real opportunity begins when you document how your team thinks, turn that knowledge into repeatable skills, and eventually build agents that can run those workflows consistently.
• 0:38
  SPEAKER_00: Today I'm joined by Edwin Choi, founder and CEO of Jet Fuel Agency.
• 0:45
  SPEAKER_00: Before launching the agency, Edwin helped scale an e-commerce company from 5 million to 30 million in annual revenue while growing his marketing team from one person to 20 people.
• 0:58
  SPEAKER_00: He founded Jet Fuel Agency in 2017 and now supports CPG and e-commerce brands across paid social, search, email, SMS, conversion rate optimization, and SEO.
• 1:12
  SPEAKER_00: In this conversation, Edwin introduces a practical three-level framework for AI adoption.
• 1:19
  SPEAKER_00: You begin by learning how to communicate with AI.
• 1:22
  SPEAKER_00: You then turn your operating playbooks into repeatable skills.
• 1:26
  SPEAKER_00: Once those skills consistently produce the right output, you can schedule them, connect them, and develop agents.
• 1:34
  SPEAKER_00: We also examine the internal AI system behind the Jet Fuel Agency, including a competitor gap analysis that reviews advertising, product detail page, keywords, and customer prompts.
• 1:48
  SPEAKER_00: Edwin also shared how his team combines years of conversion testing and with Microsoft Clarity data to identify and build new website experiments.
• 2:00
  SPEAKER_00: And most importantly, Edwin explains what an e-commerce team should automate first, how to test whether a workflow genuinely is useful, and why human judgment and taste become more valuable as AI handles more of the repetitive work.
• 2:19
  SPEAKER_00: Let's get into it, guys.
• 2:21
  SPEAKER_00: Hey Edwin, welcome to the Selling on Giants podcast.
• 2:24
  SPEAKER_00: I'm excited to have you here to talk about AI and what AI is doing in terms of marketing ROI.
• 2:30
  SPEAKER_00: But before we jump in, Edwin, tell me a bit about yourself.
• 2:33
  SPEAKER_00: How did you get into the marketing space?
• 2:35
  SPEAKER_01: Yeah, absolutely.
• 2:36
  SPEAKER_01: Well, Will, it really started all the way back in elementary school.
• 2:39
  SPEAKER_01: So I didn't have an allowance like other kids.
• 2:42
  SPEAKER_01: If I wanted to buy something like gum or candy or chips, I had to figure out how to make my own dollars, right?
• 2:48
  SPEAKER_01: And what I found out is that if you buy something for a dollar and you sell it for two, all of a sudden you have an extra dollar in your pocket, right?
• 2:55
  SPEAKER_01: So I was selling collectible cars, I was making music CDs for some of the students and kind of selling my labor, making that, right?
• 3:03
  SPEAKER_01: $10,$15 a pop.
• 3:05
  SPEAKER_01: And that really helped fuel my love for selling stuff, right?
• 3:08
  SPEAKER_01: You know, until they took me aside and said, hey, you can't do that anymore.
• 3:12
  SPEAKER_01: We're gonna kick you out, right?
• 3:13
  SPEAKER_01: So fast forward till today, right?
• 3:16
  SPEAKER_01: I started my career in 2009 at a marketing agency.
• 3:20
  SPEAKER_01: I was there for two and a half years doing organic and paid search.
• 3:24
  SPEAKER_01: And I was getting into e-commerce.
• 3:26
  SPEAKER_01: Like e-commerce was a big deal back then, and that it was on the forefront, right?
• 3:31
  SPEAKER_01: And you're selling things and making a margin and it was slow safety, right?
• 3:35
  SPEAKER_01: And after that, I went to a cell phone accessory company, I was marketing employee number two, then marketing employee number one left right away.
• 3:43
  SPEAKER_01: So then I became the only marketing employee.
• 3:45
  SPEAKER_01: Uh and we scaled that company for five to 30 million a year in four years.
• 3:50
  SPEAKER_01: Went from 20 people to 110 people.
• 3:53
  SPEAKER_01: My team went from myself to 20 people on the marketing team.
• 3:56
  SPEAKER_01: So it was like a wild ride, right?
• 3:59
  SPEAKER_01: And we touched everything like we were meta, Google, affiliate, SEO, email, every single channel you can think of.
• 4:07
  SPEAKER_01: We we did it and we did it at a really high level.
• 4:09
  SPEAKER_01: And I left that company in 2017 to start Jetfield Agency because I just wanted something that was a personification of my love of marketing, not my love of teaching and mentoring others, right?
• 4:22
  SPEAKER_01: And that's how we came here to today.
• 4:24
  SPEAKER_01: So we've been around for nine plus years now, very fortunate to have worked with some amazing brands, especially in CPT, food and beverage, health and wellness, you know, in the pet industry.
• 4:35
  SPEAKER_01: Uh, so we do paid social, we do organic and paid search, we do retention, marketing, email, and SMS.
• 4:42
  SPEAKER_01: Uh, that kind of led us here to today.
• 4:44
  SPEAKER_00: That's pretty amazing.
• 4:45
  SPEAKER_00: So you build jet fuel around the idea of repeatable success.
• 4:49
  SPEAKER_00: What does that actually mean?
• 4:50
  SPEAKER_00: And why is it repeatable?
• 4:52
  SPEAKER_00: Uh, and why is it so important when implementing AI?
• 4:55
  SPEAKER_01: Yeah, well, absolutely.
• 4:57
  SPEAKER_01: Well, if you get successful once, you're you're lucky.
• 5:00
  SPEAKER_01: Okay.
• 5:00
  SPEAKER_01: If you get successful two, three, four, five times in a row, is the process, right?
• 5:05
  SPEAKER_01: So we want to make sure that everything is a process.
• 5:08
  SPEAKER_01: When brands come to us, we want to build things that are sustainable, great for the brand equity, great for your bottom line, and you want that to be repeatable so you control the variables, right?
• 5:19
  SPEAKER_02: It shows that you know what you do.
• 5:20
  SPEAKER_01: Now, how is this important for AI?
• 5:22
  SPEAKER_01: It's incredibly important for AI.
• 5:24
  SPEAKER_01: Because I don't know if anyone has used AI to like extremely high level, but one of the pillars of displaying AI fluency is making sure that if you talk to whether it be ChatGPT or SLOT or Gemini, if you talk to them on five different days and you have five different clients, then you get the exact same result every time.
• 5:46
  SPEAKER_01: And that is a skill and a language almost that you have to learn in order to get that.
• 5:51
  SPEAKER_01: And you really can't get them, you know, them being AI, to give you the output you need if you don't have an extremely clear idea of how you do things, like what is your process for marketing or running a channel or running Amazon, right?
• 6:08
  SPEAKER_01: Like all that has to be crystal clear, or else it's not going to be clear to AI and it's not gonna give you something repeated.
• 6:15
  SPEAKER_00: Agreed.
• 6:16
  SPEAKER_00: It's like uh a business too.
• 6:17
  SPEAKER_00: When you start it, you kind of solve your own problem before you start solving it for other people.
• 6:22
  SPEAKER_00: And I think that's something to be uh to consider as you're building uh solutions.
• 6:27
  SPEAKER_00: Um, but so let's talk about your AI experience before we jump into the main topic from AI experiments to marketing ROI.
• 6:35
  SPEAKER_00: When you talk about AI agents and workflows, what does your team actually build and put into regular use at JetFuel?
• 6:43
  SPEAKER_01: Yeah, absolutely.
• 6:44
  SPEAKER_01: So I'll we have let's call it a hub and spoke model, right?
• 6:48
  SPEAKER_01: So in the middle, which is the hub, we have what we call JetFuel HQ.
• 6:53
  SPEAKER_01: Right.
• 6:53
  SPEAKER_01: So that's that's the brain that runs everything.
• 6:56
  SPEAKER_01: So we believe that all of our skills and agents need a very strong organized content library and context library for them to do things, right?
• 7:07
  SPEAKER_01: So every meeting, every email, every Slack message, every experiment, every channel feeds into this AI in a darn near real-time basis and it organizes it per client per virtual, right?
• 7:19
  SPEAKER_01: So that's that's where the hub is.
• 7:21
  SPEAKER_01: Then we build skills that are the spokes coming out from the hub, right?
• 7:26
  SPEAKER_01: So our team, let's say we have a page search team, right?
• 7:29
  SPEAKER_01: We have a page search team, they will be uh joining the forces together to come up with a series of skills and agents that's gonna help them with their day-to-day work, right?
• 7:39
  SPEAKER_01: So we put our heads together, we spec it out, and then we connect it to the brain, which is in the middle.
• 7:45
  SPEAKER_01: So it's really easy for someone to get repeatable results because no matter what the client said, like last night a client said this morning, hey, we're raising budgets, we have some new initiatives, that brain already knows.
• 7:58
  SPEAKER_01: So later that afternoon, when they go run the skill or the agent, it already has that context.
• 8:03
  SPEAKER_01: It's getting sharper and sharper insights over time.
• 8:06
  SPEAKER_01: So what we're looking for is we're looking for compound uh learnings that really help make your work better and better, right?
• 8:14
  SPEAKER_01: So each individual team member, they have their own library of skills and agents that we we we obviously are in control of the version history and make sure that we're not just willy-nilly uploading new skills and agents in a repo.
• 8:28
  SPEAKER_01: Uh, but they have their own collection, right?
• 8:30
  SPEAKER_01: And then we divide that into buckets.
• 8:33
  SPEAKER_01: So we have one bucket called research, which where your skills or agents are good at pulling information, pulling competitor data.
• 8:40
  SPEAKER_01: We have one called monitor.
• 8:42
  SPEAKER_01: So collection of skills that's that basically look at everything 24-7 and they go, Hey, that looks weird.
• 8:49
  SPEAKER_01: You should look into it, right?
• 8:50
  SPEAKER_01: And then we have our experimentation skills, which says, hey, next month the client runs wants to run a big test to your four ideas based on four tests that you ran before, right?
• 9:02
  SPEAKER_01: And then last but not least, we have our reporting and dashboarding skills, which make important conclusions and insights.
• 9:08
  SPEAKER_01: And it really helps with, you know, let's say the client wants a big quarterly presentation.
• 9:13
  SPEAKER_01: So we use those skills to help pull the numbers together and then pull out patterns for you to look at, but it doesn't write the report.
• 9:20
  SPEAKER_01: I think that part is really uh very important to know is that we never really want them to write for us, but we want them to kind of give us the data on the silver platter, and then we will uh yeah.
• 9:33
  SPEAKER_00: That's pretty crazy.
• 9:35
  SPEAKER_00: And then uh, if I'm accessing that, is it a through a web portal?
• 9:38
  SPEAKER_00: Do you guys use Slack?
• 9:39
  SPEAKER_00: Like, how does your team uh access these bots?
• 9:42
  SPEAKER_01: Yeah, so there's three ways to access it.
• 9:44
  SPEAKER_01: There's our uh once a skill is hardened and durable, meaning that it doesn't crash, it could get really repeatable results, it's through our web portal, right?
• 9:53
  SPEAKER_01: So you don't necessarily need to be an expert in prompting to run the skill correctly.
• 9:59
  SPEAKER_01: It's almost like running a web tool, right?
• 10:01
  SPEAKER_01: So a good amount of our skills live there.
• 10:03
  SPEAKER_01: Now, if the skill is not ready for prime time, it'll live inside, you know, it has its own MD files and et cetera, that you could pull from the repo, right?
• 10:12
  SPEAKER_01: And then you could run it from your terminal, right?
• 10:14
  SPEAKER_01: So everyone has access to that and they're running that.
• 10:16
  SPEAKER_01: You know, maybe it only works 70% of the time.
• 10:18
  SPEAKER_01: So we we didn't take the we're still working on it and it's not fully upgraded yet.
• 10:24
  SPEAKER_01: And then last but not least, they we like to keep everything in Slack so you can talk to the agents on Slack like, hey, I love these keywords, just go ahead and upload, and then they'll go ahead and do it, right?
• 10:36
  SPEAKER_01: So our team can, you know, they don't have to switch windows, switch programs, they can't just move on with the day, say, hey, thumbs up, thumbs up, let's upload that.
• 10:43
  SPEAKER_01: We're good.
• 10:44
  SPEAKER_00: You built something really cool.
• 10:46
  SPEAKER_00: I'm excited to get into the weeds, and I do want to take it a step back for some of our more layman listeners.
• 10:51
  SPEAKER_00: But one last question before we jump into our main topic.
• 10:55
  SPEAKER_00: Uh, which A workflows have you produced that has the clearest and measurable business impact for your team?
• 11:02
  SPEAKER_00: Walk us through the before and then the the afterlife since implementing whatever AI solution in your company that's improved, you know, either time on accounts or uh profitability or some of the other metrics that I know we look at uh on the agency side of the business.
• 11:20
  SPEAKER_01: Yeah, so one of the skills we're really proud of is called a competitor gap analysis.
• 11:25
  SPEAKER_01: So this is very strategic, right?
• 11:28
  SPEAKER_01: And usually, you know, if you're let's say a paid search person, let's say you're an Amazon PPC person, a lot of times these individuals do no fault of their own so that they're really like laser targeted on their campaign.
• 11:40
  SPEAKER_01: Oh, now I need to build this campaign.
• 11:41
  SPEAKER_01: I gotta get the ROI out, right?
• 11:43
  SPEAKER_01: Which is a great focus to have.
• 11:44
  SPEAKER_01: But sometimes it it gives you so much value to zoom out of it, right?
• 11:50
  SPEAKER_01: And this is what this agent helps you do.
• 11:53
  SPEAKER_01: So you go to the web portal and you run what's called a competitor gap analysis, and you run it and it basically crawls every single surface that your brand is on.
• 12:04
  SPEAKER_01: So let's say, let's pretend I have a copy brand, like better for your author, and you know, there's like seven competitors.
• 12:10
  SPEAKER_01: So it'll label all those competitors for you, and it'll give you here's all the ad copy they're running, here's what their Amazon PDP says, here's what their keywords are, here's what all their uh Facebook ads look like, and then it'll do it for all seven.
• 12:25
  SPEAKER_01: And it'll say here here are all the things, you kind of combine all seven of them, your all the things are not sync.
• 12:32
  SPEAKER_01: Wow, like A, B, C, and then it'll connect to we have another agent called our LLM Prompt Researcher, and it'll connect to prompts.
• 12:42
  SPEAKER_01: So let's pretend that my coffee company is called Better For You Coffee, right?
• 12:47
  SPEAKER_01: Yeah, and then it'll say it'll give you prompts like, hey, based on all the prompt knowledge that I've gathered, a common question is what is a coffee brand that won't make me jittery?
• 12:58
  SPEAKER_01: And then you can take those prompts and your competitor gap analysis, merge it together, and then we hand that to our team.
• 13:05
  SPEAKER_01: And then all of a sudden they're their ad copy, their Amazon PDP description, the imagery, what they're running on meta, they're talking to these gaps.
• 13:15
  SPEAKER_01: You're saying something to the market that it hasn't seen before, you get better responses, right?
• 13:22
  SPEAKER_01: So that has been tremendous because before this process, we're doing that, right?
• 13:28
  SPEAKER_01: We're doing that, but we tend to only do it on onboarding, right?
• 13:31
  SPEAKER_01: And then we're like, kind of go about our business and we semi-forget, right?
• 13:35
  SPEAKER_01: But now, afterward, we're running it monthly, we're running it quarterly, we're we're evolving this, and that brand equity messaging and the efficacy of what we're doing.
• 13:46
  SPEAKER_01: All of a sudden, our ads are cheaper, all of a sudden our confronters are all of a sudden people are lingering on site or on Amazon or on Walmart or wherever you know our clients are, right?
• 13:57
  SPEAKER_01: And it's and it's a little bit of a hidden underlayer where we're saying all these things that no one else is saying, but that's what people care about.
• 14:04
  SPEAKER_01: So hopefully that makes sense.
• 14:06
  SPEAKER_00: No, it makes perfect sense.
• 14:07
  SPEAKER_00: And uh what I like is like the competitive Intel is definitely a way just to stay up on the competition to catch trends early.
• 14:14
  SPEAKER_00: Uh, and to your point, before uh AI being what it is today, you know, having pulling doing competitive research, it was you know, a lot a lot of it was a lot of manpower.
• 14:26
  SPEAKER_00: Um so the fact AI can do it monthly and have it delivered to you at eight o'clock on the first of every month is a game changer.
• 14:35
  SPEAKER_00: And whether you're selling products or services, that's uh yeah, but wouldn't that be valuable?
• 14:40
  SPEAKER_00: You'll have to share it with me uh after the show.
• 14:44
  SPEAKER_00: Um so let's dive into it now.
• 14:47
  SPEAKER_00: I'm excited.
• 14:47
  SPEAKER_00: So from A from AI experiments to marketing ROI, uh let's talk about AI agents and workflow first.
• 14:55
  SPEAKER_00: So, in plain English, for those who maybe uh so far this conversation has been in Spanish.
• 15:00
  SPEAKER_00: Uh, the difference between AI tools and AI workflows and AI agents, like what yeah, what are we talking about here?
• 15:08
  SPEAKER_01: Yeah, absolutely.
• 15:09
  SPEAKER_01: So let's let's think of it like a pyramid, right?
• 15:11
  SPEAKER_01: So you kind of build a pyramid in the bottom layer, let's say it's level one, right?
• 15:16
  SPEAKER_01: So low barrier to entry, but also your results are probably not mega stellar, but it's a good place to start, right?
• 15:22
  SPEAKER_01: So on the bottom would be just chatting to your LL, right?
• 15:26
  SPEAKER_01: So you open up a chat GPT, you open your plot, whatever your tool is of choice, you write a prompt and you get something better, right?
• 15:34
  SPEAKER_01: So I think a lot of people are not this level one, and level one is important, right?
• 15:39
  SPEAKER_01: Like for our team, when we onboard new employees, we start them off at level one because you need to learn certain things.
• 15:44
  SPEAKER_01: Like you, you know, you probably talk to uh let's call it chat GPT, and you type in a prompt and it gives you something like that.
• 15:51
  SPEAKER_01: Is not at all like that is not at all what it wanted, right?
• 15:54
  SPEAKER_01: So try again, yeah, yeah.
• 15:56
  SPEAKER_01: Try again.
• 15:57
  SPEAKER_01: All right.
• 15:58
  SPEAKER_01: Um, so you have to learn what you know what we call prompt engineering.
• 16:01
  SPEAKER_01: So we have some courses that we have internally for basically saying, here's how LLMs think, here's how humans think.
• 16:08
  SPEAKER_01: Like you gotta make sure that those languages match up.
• 16:11
  SPEAKER_01: You gotta be specific about certain things, you gotta make sure to give appropriate context, you gotta guide it, right?
• 16:17
  SPEAKER_01: And then let's say they get really good at level one, okay?
• 16:20
  SPEAKER_01: So let's say level one is good, they're solid, they're gonna move up the pyramid, they're gonna go to level two, right?
• 16:26
  SPEAKER_01: Once you go to level two, you have your skills.
• 16:28
  SPEAKER_01: So, what is a skill?
• 16:30
  SPEAKER_01: Well, a skill is let's say you're uh you're you're you're running Amazon PPC and you do something every single week.
• 16:37
  SPEAKER_01: Let's call it keyword research, right?
• 16:39
  SPEAKER_01: And you're you're thinking to yourself, oh, holy smokes, you know, I spend two hours a week doing keyword research, it's a very manual process, right?
• 16:47
  SPEAKER_01: How can I automate it?
• 16:48
  SPEAKER_01: Well, skill is something that you can say, here's my mindset.
• 16:52
  SPEAKER_01: When I sit down on my computer, here are the steps I take to do keyword research, right?
• 16:57
  SPEAKER_01: And you train and build a skill to do the same steps, right?
• 17:01
  SPEAKER_01: And you call that let's call that skill PPC research.
• 17:04
  SPEAKER_01: Then you go to your terminal or your like flawed code or codecs or whatnot, and you say, every time I want to do this thing, I'm gonna run the skill.
• 17:15
  SPEAKER_01: So instead of me going through an interface and building stuff myself, I'm I'm gonna run the skill, right?
• 17:20
  SPEAKER_01: And then I'm gonna run it manually, it's gonna work for me, and then in 10 minutes, like five to ten minutes, I get my result, right?
• 17:27
  SPEAKER_01: And then you can train your skill based on output to get the result you want.
• 17:31
  SPEAKER_01: So that's level two.
• 17:32
  SPEAKER_01: So level two is a whole other ballgame.
• 17:34
  SPEAKER_01: Like you have to think about how do you train a skill, how do you write it out and scope it out, right?
• 17:39
  SPEAKER_01: In a way that you know Claude or Codex understands how do you make it repeatable?
• 17:43
  SPEAKER_01: How do you make it so it doesn't break, right?
• 17:45
  SPEAKER_01: So all of this is level two.
• 17:48
  SPEAKER_01: Now, let's say you do level two and you're like, man, I am so comfortable.
• 17:52
  SPEAKER_01: My skill is always doing what I want.
• 17:54
  SPEAKER_01: Let's go to level three.
• 17:55
  SPEAKER_01: Let's make it an agent, right?
• 17:57
  SPEAKER_01: Agent is a skill that is either running automatically or maybe it's triggered by another skill.
• 18:04
  SPEAKER_01: Uh, it has a lot some level of autonomy, right?
• 18:08
  SPEAKER_01: So you might say, hey, this skill, like let's say the the simplest thing to do, I want to run every Monday at 8 a.m.
• 18:15
  SPEAKER_01: Right?
• 18:16
  SPEAKER_01: Very, very simple agent.
• 18:17
  SPEAKER_01: It's almost like a scheduled skill.
• 18:18
  SPEAKER_01: Oh, let's take it a step further.
• 18:20
  SPEAKER_01: Uh, let's say I have a skill, I'm gonna run it at 8 a.m.
• 18:24
  SPEAKER_01: on Monday, and then on Tuesday, I have another skill that evaluates the quality of Monday's run.
• 18:33
  SPEAKER_01: And it has its own logic to say, I'm gonna key way it, anything that's wrong, I'm gonna fix it, and then I'm gonna go back and improve that skill.
• 18:43
  SPEAKER_01: And you run that Monday, Tuesday.
• 18:45
  SPEAKER_01: Now it becomes more autonomous, right?
• 18:48
  SPEAKER_01: Yeah, and then let's take another step further.
• 18:50
  SPEAKER_01: Let's say you built something called a supervisor agent that just looks at these other skills and you say, Hey, supervisor agent, my coffee company right now is doing half a million in revenue a year.
• 19:01
  SPEAKER_01: I want to get to a million in revenue a year.
• 19:03
  SPEAKER_01: Your job is to figure out the gap, like how do you get that incremental half a million, right?
• 19:10
  SPEAKER_01: And then it I will allow you to talk to the other agents to build your way up there, right?
• 19:17
  SPEAKER_01: And then the supervising agent says, Oh, I'm gonna, you know, instead of running, you know, I determined that instead of running this skill every Monday, I'm gonna run it Monday, Wednesday, Friday.
• 19:25
  SPEAKER_01: And it's gonna make that take three, right?
• 19:28
  SPEAKER_01: Yeah, and then if if you want to take it even a step further, right?
• 19:31
  SPEAKER_01: So let's say you have that autonomously going.
• 19:33
  SPEAKER_01: Once you get to here, you have what's called the it iteration or self-learning boost, right?
• 19:38
  SPEAKER_01: Where you have another agent that says, you know what, my only job is to see how this whole system is running and figure out my own experiments to improve it to get this whole system to a version too.
• 19:52
  SPEAKER_01: That's true.
• 19:53
  SPEAKER_00: That's yeah, that's that's where the matrix lives.
• 19:57
  SPEAKER_01: That's where the matrix lives, right?
• 19:58
  SPEAKER_01: That's where the matrix lives.
• 20:00
  SPEAKER_01: So it's it's almost like you you kind of build the skills from intern to specialist and manager to director to CMO.
• 20:07
  SPEAKER_00: That's kind of how I it's it's it's well put too.
• 20:10
  SPEAKER_00: And what I what I'm hearing too, so anybody who's new to this, it's like everybody starts off the same way.
• 20:16
  SPEAKER_00: And I I as you were talking through your journey, I'm you know, not on the uh CMO level yet, but like you start off talking to the chat bot, then you build some skills out, and they've made it, you know, like Claude's made it really easy being able to record your screen, and then uh I have uh agents that have roles, uh, orchestrator, executor, uh quality control, and the three of them work together, and then they work within a system.
• 20:43
  SPEAKER_00: So you can hear how you know uh how sophisticated these can get, but the level of problem solving and time it could save you is uh is is pretty crazy.
• 20:55
  SPEAKER_00: This is the future of business, and it's exciting.
• 20:58
  SPEAKER_00: Um, so I wanted to summarize that, but now real quick before we kind of get deeper into this, you're let's keep it on the layman level.
• 21:07
  SPEAKER_00: Uh, I have just been using ChatGPT to reformat my emails.
• 21:12
  SPEAKER_00: I I can I guarantee you there's probably 30% of our listeners who are like, it can do what?
• 21:18
  SPEAKER_00: Um what's the first thing you would do?
• 21:21
  SPEAKER_00: So if you are uh a CMO or you're the CEO of an e-commerce business um and you're hearing this, how can I identify my first opportunity for AI uh implementation?
• 21:35
  SPEAKER_00: Like where I and how I should think about the time investment to get it there and what it's gonna save me.
• 21:41
  SPEAKER_01: Yeah, heck yeah.
• 21:42
  SPEAKER_01: So the absolute first thing I would do is I would sit down and write a I would write a specific playbook for the things you do on a daily basis.
• 21:52
  SPEAKER_01: So let's say for me, let's uh let's kind of go back in time, right?
• 21:56
  SPEAKER_01: Hop in the time machine.
• 21:57
  SPEAKER_01: I go back to you know when I was in the house, right?
• 21:59
  SPEAKER_01: So let's Say I was doing paid search and I was doing organic SEO, right?
• 22:04
  SPEAKER_01: And then I have tasks I have to do in order to hit goals, right?
• 22:07
  SPEAKER_01: For the company or for the client.
• 22:09
  SPEAKER_01: So I'll write down these tasks.
• 22:11
  SPEAKER_01: But I do these tasks in a certain way.
• 22:13
  SPEAKER_01: Like there's certain data points I look at, there's certain judgment calls I make.
• 22:18
  SPEAKER_01: You gotta write that all down.
• 22:20
  SPEAKER_01: Every single person, you know, you, yourself, me, we have our own internal playbook for how we think strategically about how to tackle, let's call it growth problems, right?
• 22:31
  SPEAKER_01: And it's unique per person, right?
• 22:33
  SPEAKER_01: What my playbook's probably not the same as yours.
• 22:35
  SPEAKER_01: Yours is probably not the same as mine, right?
• 22:37
  SPEAKER_01: And they're both good, but they have different flavors.
• 22:39
  SPEAKER_01: That's our DNA, right?
• 22:41
  SPEAKER_01: You need to write that down.
• 22:42
  SPEAKER_01: So it needs to be written down in a very structured format.
• 22:46
  SPEAKER_01: You can Google Docs, Notepad, whatever.
• 22:48
  SPEAKER_01: You can write it by hand.
• 22:49
  SPEAKER_01: Uh and it basically tells you here's my North Star, here are the things that I do to hit that North Star, here are judgment calls that I make, and here are some outliers that come up frequently, right?
• 23:00
  SPEAKER_01: Write that down.
• 23:02
  SPEAKER_01: Once you have that down for all the various buckets of channels or companies or verticals that you participate in, you have something organized.
• 23:11
  SPEAKER_01: So let's say you have a collection of seven playbooks, right?
• 23:14
  SPEAKER_01: Um and if you're really big into sports, then you know that you know playbooks are a big part of like football, right?
• 23:20
  SPEAKER_01: You have your coach philosophy, you have your coaching philosophy, how do you approach defense, how you you know, attack, right?
• 23:25
  SPEAKER_01: Like all those things are in these playbooks, right?
• 23:27
  SPEAKER_01: Same thing for channels or businesses, right?
• 23:30
  SPEAKER_01: Once you have that down, then you go look at okay, what is something that's not gonna replace me, but it will replace a lot of time and it'll exponentially accelerate my human superpower, right?
• 23:46
  SPEAKER_01: Everyone has different human superpower.
• 23:47
  SPEAKER_01: Some people are like incredible tastemakers, right?
• 23:50
  SPEAKER_01: Like I have people on my creative team, they can predict what type of TikTok video is gonna go viral next, right?
• 23:56
  SPEAKER_01: They're not looking back, they're looking forward.
• 23:57
  SPEAKER_01: That's their superpower, right?
• 23:59
  SPEAKER_01: So I might have something for them like, okay, why don't I build you a skill where you can look at the top 20 viral TikToks of last seven days for the vertical of your client.
• 24:10
  SPEAKER_01: I just hand it to you on silver platter.
• 24:12
  SPEAKER_01: You used to have to do that and take you know 45 minutes a day, right?
• 24:17
  SPEAKER_01: Or I could just basically give it to you in four seconds.
• 24:20
  SPEAKER_02: Yeah, right.
• 24:20
  SPEAKER_01: So you kind of have these tasks, but the tasks are reading the playbooks, so they kind of have their DNA in it, so they don't go you know willy-nilly.
• 24:31
  SPEAKER_01: They they understand like this is how Edwin thinks, this is what he cares about, these are things that he don't doesn't care about.
• 24:36
  SPEAKER_01: These these this is what he thinks the bar quality should be, right?
• 24:40
  SPEAKER_01: And then I would go forth that way.
• 24:42
  SPEAKER_01: So that I think that has been the best way for us.
• 24:45
  SPEAKER_00: I like that too.
• 24:46
  SPEAKER_00: And uh some tips to add on to it.
• 24:48
  SPEAKER_00: Uh, for those who like to talk, I use whisper flow a lot.
• 24:52
  SPEAKER_00: I just hold the mic button on my keyboard and think out loud, and it's helped me solve problems without actually having to type it out.
• 25:00
  SPEAKER_00: Um, and I think that can be a great thing to use as you're thinking through some of these ideas.
• 25:06
  SPEAKER_00: And you know, your playbooks to your pointer, uh, I think they're the memory files pretty much.
• 25:11
  SPEAKER_00: They'll always refer back to them.
• 25:13
  SPEAKER_00: And you could always ask AI to ask you the questions you need to make an effective playbook, essentially, and then take your time and answer it as honestly as possible.
• 25:24
  SPEAKER_00: Um, beautiful.
• 25:25
  SPEAKER_00: So we we we understand some of the fundamentals now of AI agents and workflows and how they work together.
• 25:31
  SPEAKER_00: Let's talk about um let's talk about using AI for marketing and e-commerce.
• 25:37
  SPEAKER_00: So walk us through how you take a manual marketing process and turn it into an AI workflow.
• 25:44
  SPEAKER_00: What give me the steps from identifying the problem to testing the final output?
• 25:49
  SPEAKER_01: Yeah, yeah.
• 25:49
  SPEAKER_01: Let's let's let's do step one, right?
• 25:52
  SPEAKER_01: So this let's say we we take a channel, uh try to just pretend it's Amazon PPC.
• 25:56
  SPEAKER_01: And then you have you have a a high value laborative task, right?
• 26:02
  SPEAKER_01: So let's say it's uh you know, keyword research, keyword line, right?
• 26:06
  SPEAKER_01: And then you and then you think to yourself, okay, how I'm gonna sit down and write the playbook first.
• 26:10
  SPEAKER_01: If I was to do this myself, or like you said, you can screen record and just record, like pretend you're teaching an intern.
• 26:17
  SPEAKER_01: This is how I do it, these are the stats I look at.
• 26:19
  SPEAKER_01: I I do these date ranges, and this is what's important to me.
• 26:23
  SPEAKER_01: Get that all out on paper, have it come down here, right?
• 26:27
  SPEAKER_01: Um, I think for a lot of the listeners who may be uh a little more on the level one side, I would recommend going and researching skill building skills, which I know sounds interesting, but there's um there's a couple skill building skills I can send them to you afterwards so the the listeners can benefit.
• 26:47
  SPEAKER_00: I'll put them inside the show notes.
• 26:48
  SPEAKER_00: Thanks, Edwin.
• 26:49
  SPEAKER_01: Yeah, yeah.
• 26:50
  SPEAKER_01: So you would take these skills and helps organize your thoughts, right?
• 26:53
  SPEAKER_01: Like say, hey, I have this recording, and then the skill will ask you some questions like, oh, what about this?
• 26:58
  SPEAKER_01: What about that?
• 26:59
  SPEAKER_01: What about this?
• 27:00
  SPEAKER_01: Like, do you want the output to be X or Y?
• 27:02
  SPEAKER_01: So basically it does sum up the thinking for you, it interviews you.
• 27:06
  SPEAKER_01: Then it builds a skill, right?
• 27:08
  SPEAKER_01: Then your job is to make sure the skill has repeatable output for its inputs, right?
• 27:15
  SPEAKER_01: And at this point, you can get pretty fancy, right?
• 27:17
  SPEAKER_01: So I think in the beginning, what I would recommend to people is you just drop a file and see how it processes it, right?
• 27:26
  SPEAKER_01: So on Amazon, you export your you know keyword data in CSV, you give it to your skill, and you see what the output is, right?
• 27:32
  SPEAKER_01: Yeah, and then once you feel really happy with it, you're like, oh, holy smokes, like this is looking really solid.
• 27:37
  SPEAKER_01: Um, and you're kind of training it, you're giving feedback.
• 27:40
  SPEAKER_01: You said, I don't like this, I like that.
• 27:42
  SPEAKER_01: Then you can make it more automated.
• 27:43
  SPEAKER_01: So you can use you know MCPs.
• 27:46
  SPEAKER_01: Uh so MCP is just to kind of give Lehman's term.
• 27:50
  SPEAKER_01: Think of it as a uh data pike between two platforms, right?
• 27:55
  SPEAKER_01: So I think you mentioned email early on in the conversation.
• 27:58
  SPEAKER_01: So let's say it's Flavio.
• 27:59
  SPEAKER_01: Flavio has an MCP, so it's basically like a like a port, like think of it like a virtual port that has consistent data passing through and it connects to your you know chat GBT or your your motor spot.
• 28:15
  SPEAKER_01: And basically it's a very um standardized way for them to talk to each other, right?
• 28:19
  SPEAKER_01: Uh or if you use APIs.
• 28:22
  SPEAKER_01: And then you can say, hey, instead of me giving you a file, I want you to pull it yourself.
• 28:28
  SPEAKER_01: That data pipe is connected, it's always open, right?
• 28:31
  SPEAKER_01: And then once you get that going, you sit and evaluate, you run the skill for a couple weeks.
• 28:36
  SPEAKER_01: Like, is it really helping?
• 28:38
  SPEAKER_01: Is it really helping, or is it just giving you more work, right?
• 28:41
  SPEAKER_01: Because sometimes you you you build a skill and it's not going well, it actually doubles or triples your work a lot.
• 28:48
  SPEAKER_01: Okay, so you have to really make that evaluation.
• 28:49
  SPEAKER_01: Like, do I need to keep refining it, or is something else a better priority?
• 28:55
  SPEAKER_01: Yeah, once that is done really well, then you can go into scheduling, right?
• 29:00
  SPEAKER_01: Say, hey, I'm so happy with this skill.
• 29:02
  SPEAKER_01: I ran it 10 times in a row.
• 29:04
  SPEAKER_01: All 10 times I'm happy, I'm gonna schedule it.
• 29:07
  SPEAKER_01: Drop it right here, right?
• 29:08
  SPEAKER_01: And then you can, you know, then go on so forth.
• 29:10
  SPEAKER_01: You can then automate it, you do self-improvement moves.
• 29:14
  SPEAKER_01: Uh you can have, you know, we we don't typically do this, right?
• 29:17
  SPEAKER_01: But you can have it start to work on the account autonomously.
• 29:20
  SPEAKER_01: We like to always have the final say.
• 29:23
  SPEAKER_01: So he'll they'll like maybe do something in draft, but we'd like to have a human always for you, you trust it.
• 29:30
  SPEAKER_01: So yeah, hopefully that makes sense.
• 29:32
  SPEAKER_00: That makes a ton of sense.
• 29:33
  SPEAKER_00: And I'd say for the layman listening, uh, the easiest way to start is kind of reporting.
• 29:38
  SPEAKER_00: You kind of hinted at it, Edwin.
• 29:40
  SPEAKER_00: I think you're right.
• 29:41
  SPEAKER_00: It's like uh it's quick to verify.
• 29:44
  SPEAKER_00: Um, you're really not gonna break anything if you're just getting, you know, connected to an API and pulling to your point, Amazon PPC data or or Clavio data and building dashboards is a great way to kind of move from that level one up to that level two and start getting comfortable with agents.
• 30:02
  SPEAKER_00: Because to your point, when you have multiple layers working together, and I've been there where I've worked on a project and went way too deep where I should have pulled out and went back to the drawing boards, and uh, you know, it'll mislead you if you're not prepared.
• 30:18
  SPEAKER_00: Um, so that is super helpful.
• 30:21
  SPEAKER_00: Um, can you give a specific e-commerce marketing example and walk through kind of how you how you got put it together, what the triggers were and anything along those lines?
• 30:32
  SPEAKER_01: Yeah, absolutely.
• 30:33
  SPEAKER_01: Uh I'll give you something that we ran very recently.
• 30:35
  SPEAKER_01: We're pretty happy with it.
• 30:37
  SPEAKER_01: Uh, and it's a little technical, but so I'm I'm gonna try to keep word it in a way that that makes sense.
• 30:43
  SPEAKER_01: So um I think something that we really like to do is if we get an e-commerce client, we really like to have the conversion rate be higher on the site, right?
• 30:53
  SPEAKER_01: So let's take the coffee examples.
• 30:54
  SPEAKER_01: Coffee client comes to us and says hey, I have a whole suite of channels, and then we want you to run them, and we run them.
• 31:02
  SPEAKER_01: We're like, oh, you know what?
• 31:03
  SPEAKER_01: I really wish instead of the site converting at 3%, I want to convert at 6%.
• 31:07
  SPEAKER_02: Yeah, right.
• 31:08
  SPEAKER_01: So what we did is and this is very interesting, right?
• 31:14
  SPEAKER_01: So I have run CRO tests, conversion rate optimization tests since 2011, right?
• 31:20
  SPEAKER_01: So I have hundreds of these test documents that I hand wrote, right?
• 31:25
  SPEAKER_01: That I hand wrote, you know, back in the day.
• 31:27
  SPEAKER_01: But basically it's very it's it's almost like uh scientific paper.
• 31:30
  SPEAKER_01: Here's my hypothesis, here's the data that I used to build it, here's how the test ran, and then here is uh the final result, right?
• 31:38
  SPEAKER_00: Here's the lean analytics fundamentals going on.
• 31:40
  SPEAKER_01: Yeah, exactly, exactly.
• 31:42
  SPEAKER_01: So I put that all in one bucket, then I have a second bucket is all the conversion material I've collected.
• 31:48
  SPEAKER_01: Some I paid for some I paid a lot more, like 500 bucks per PDF.
• 31:53
  SPEAKER_01: So I've this whole library of conversion rate optimization books and principles that formulated my mindset for being a strong conversion rate optimizer.
• 32:03
  SPEAKER_01: So I put that in a second bucket.
• 32:05
  SPEAKER_01: Then I have a third bucket is my data connectors.
• 32:08
  SPEAKER_01: So we actually built a connector that goes to Microsoft Clarity.
• 32:13
  SPEAKER_01: So for the listeners who don't know, Microsoft Clarity is brand is a free tool and it's amazing.
• 32:18
  SPEAKER_01: You install it on your site, and it basically records user sessions of people on your pages.
• 32:25
  SPEAKER_01: So you can see where they're dropping off, you can see where they're hovering, you can see where they're frustrated with certain areas, um, and basically it has these recordings.
• 32:33
  SPEAKER_01: And you know what I found out is that these recordings how they make the recordings in a series of numbers, right?
• 32:40
  SPEAKER_01: It's always like paint, like have you ever done paint by numbers?
• 32:43
  SPEAKER_02: Yeah.
• 32:44
  SPEAKER_01: So you have it it basically is like that.
• 32:46
  SPEAKER_01: So it's like a file that says, you know, six, eight, nine.
• 32:49
  SPEAKER_01: And I guess that means that the mouse went here or here and clicked this, right?
• 32:53
  SPEAKER_01: So we tapped into that, right?
• 32:56
  SPEAKER_01: So we built this skill that that combines all three.
• 32:59
  SPEAKER_01: It's trained with all the tests I've ever run, all the material I've ever read, and how people are navigating, we're dropping off on the site.
• 33:08
  SPEAKER_02: That's amazing.
• 33:09
  SPEAKER_01: Then it goes, yeah, yeah, like that, right?
• 33:12
  SPEAKER_01: And then it goes, and this is so cool.
• 33:14
  SPEAKER_01: Like we're we're like we're nervous about this, right?
• 33:17
  SPEAKER_01: It'll make a mock, let's say it's a product page.
• 33:19
  SPEAKER_01: It'll make a mock of the product page and it'll it'll make the test for you.
• 33:23
  SPEAKER_01: It'll literally have like a one.
• 33:25
  SPEAKER_01: You highlight the one, and it's like, oh, that adds a part button to be up here, and it will say two, and it's like, oh, based on the competitive gap analysis, you need these value props in section two, and then we'll run the test.
• 33:39
  SPEAKER_01: Oh, so that thing used to be me or my team are sitting down, you know, combing through the recordings and like gathering all this stuff, and now at your fingertips, like, oh, I want eight of these tests.
• 33:52
  SPEAKER_01: Boom, boom, boom, eight, done.
• 33:54
  SPEAKER_01: Yeah, then you just review right and it size everything.
• 33:56
  SPEAKER_01: It's like, oh, the funnel pass through is lower than your benchmark.
• 33:59
  SPEAKER_01: So we think this pays a good pace test.
• 34:02
  SPEAKER_01: So yeah, hopefully that gives you an idea.
• 34:05
  SPEAKER_00: So much more information now.
• 34:06
  SPEAKER_00: It's just it's crazy what you could do with it.
• 34:09
  SPEAKER_00: And you know, what I love what you mentioned there, uh using tools like Microsoft Clarity, but understanding the user flow, like you know, CRO basics and where drop like to your point, this information would take tons of sessions and some tons of money.
• 34:23
  SPEAKER_00: Uh, and now it's available pretty easily.
• 34:26
  SPEAKER_00: Um, let's go a little deeper.
• 34:28
  SPEAKER_00: So, before launching an AI workflow, what baseline should a business establish?
• 34:33
  SPEAKER_00: Like, what metrics are you looking at to know you know, is is it all just time?
• 34:38
  SPEAKER_00: Is there something else I should be looking at?
• 34:41
  SPEAKER_01: I think is time is a factor of it, right?
• 34:44
  SPEAKER_01: Time is a factor of it.
• 34:45
  SPEAKER_01: If you can, you know, as an agency, we we track hours, we track tasks, we track output, right?
• 34:51
  SPEAKER_01: So if you went from six tasks a week to 12 tasks a week and you're still working a great 40 hour week, you're not putting OT, everyone's happy, right?
• 35:00
  SPEAKER_01: Yeah, that's certainly one thing that we track.
• 35:02
  SPEAKER_01: However, we found that we don't want to be hyper-focused on it.
• 35:06
  SPEAKER_01: We don't want to be hyper-focused on.
• 35:08
  SPEAKER_01: So the second thing that we track is okay, let's say before you did six tasks a week.
• 35:13
  SPEAKER_01: Right now you do static tasks.
• 35:14
  SPEAKER_01: Okay, so you your your output only went up like a little bit.
• 35:18
  SPEAKER_01: But we we call it a smell test, right?
• 35:21
  SPEAKER_01: But is your output ten times better than it was before?
• 35:26
  SPEAKER_01: Right?
• 35:26
  SPEAKER_01: If when you you know when you cook and you smell like, oh, that's dish smells delicious, or like, oh, that dish smells off.
• 35:31
  SPEAKER_01: If I look at their seven tasks and I and I do my smell test, you're like, holy smokes, your seven tasks are ten times better in terms of quality than the the previous six, I'm a happy camper.
• 35:44
  SPEAKER_01: Like, I don't want to like push the team so like we do want more output, but yeah, if you're 10 times better, you're gonna get you hopefully, and I think we're seeing it, 10 times more detention as an agents, your attention's everything.
• 35:57
  SPEAKER_02: Everything, yeah.
• 35:59
  SPEAKER_01: So that's kind of how we look at it is like, yes, we do like more time being saved.
• 36:06
  SPEAKER_01: Yeah, but are we hyper focused on that?
• 36:08
  SPEAKER_01: No, I I don't think so.
• 36:09
  SPEAKER_01: Like, if we can get the quality up for the same time invested, can we get 10x full?
• 36:14
  SPEAKER_01: Yeah, we're we're happy capitalist.
• 36:16
  SPEAKER_00: Not like hey, you tie it into like a major business KPI, it's to retention.
• 36:21
  SPEAKER_00: So, you know, when you're working through an AI challenge or something you're trying to solve with AI, you can always tie it back to like, is this gonna 10x the value we're adding to the brands or um you know a teammate or whatever else?
• 36:35
  SPEAKER_00: So that's a really smart way of doing it.
• 36:38
  SPEAKER_00: Uh, last question: if an e-commerce brand has no dedicated AI team, which you know we see a lot, what 30-day plan would you give them to identify, build, test, and measure one useful workflow?
• 36:53
  SPEAKER_01: Yeah, absolutely.
• 36:54
  SPEAKER_01: First thing I would do is I would go on Google, search engineer choice, whatever, type in Prompt Engineering 101.
• 37:01
  SPEAKER_01: Prompt Engineering 101.
• 37:03
  SPEAKER_01: Uh, Anthropic actually just came up with a really cool free course on how prompts work and how LLM works.
• 37:11
  SPEAKER_01: You need to understand the plumbing, right?
• 37:13
  SPEAKER_01: Like if you need to understand how they think and how they look at data and how they respond to you, that's step one.
• 37:19
  SPEAKER_01: Once you have that done, then you go into I need to be really good, you know.
• 37:24
  SPEAKER_01: We're talking about level one in the pyramid.
• 37:26
  SPEAKER_01: I need to be really good at writing prompts.
• 37:28
  SPEAKER_01: Like the prompts need to behave the way I think they should be, right?
• 37:32
  SPEAKER_01: And like, let's call that you're on day like eight, right?
• 37:35
  SPEAKER_01: So let's say you took a week to learn, you took a week to practice prompts.
• 37:39
  SPEAKER_01: Then uh then days 14 and 21, look at what you're doing, write down your playbooks, right?
• 37:45
  SPEAKER_01: Write down your playbooks.
• 37:46
  SPEAKER_01: Don't just go willy-nilly start building skills, right?
• 37:49
  SPEAKER_01: The skills need your DNA, or else they're gonna be doing who knows what, right?
• 37:54
  SPEAKER_01: Like what whatever they think is good, they're gonna do it, right?
• 37:57
  SPEAKER_01: No, it has to be what you think is good.
• 37:58
  SPEAKER_01: So they need to follow the playbooks, right?
• 38:00
  SPEAKER_01: It's kind of like again, having a football team.
• 38:02
  SPEAKER_01: Gotta follow the coach's philosophy in order for the team to work, right?
• 38:06
  SPEAKER_01: And then once you get to the last week of the month, right?
• 38:10
  SPEAKER_01: Make those skills scheduled.
• 38:11
  SPEAKER_01: Make it a goal to get one skill so good that you can run it every Monday, 8 a.m., and the results always the same.
• 38:19
  SPEAKER_01: And once you do that, you're you're well ahead of the general population and you're you're well on your way.
• 38:25
  SPEAKER_00: Yeah.
• 38:25
  SPEAKER_00: I know uh we're we talk about it a lot.
• 38:27
  SPEAKER_00: I'm sure Edwin, like myself, you take in a lot of media.
• 38:31
  SPEAKER_00: What do you you're probably 10% of the population, this deep?
• 38:34
  SPEAKER_00: I find uh in a lot of conversations I have with brands and other agency owners, uh, they're not as deep into it yet, but I think everybody is moving in that direction.
• 38:44
  SPEAKER_00: So consider yourself uh an early mover.
• 38:47
  SPEAKER_00: So let's talk about jet fuel agency real quick.
• 38:50
  SPEAKER_00: So I love having other agency owners online because the challenges of like serving brands and to your point, instead of being efficient with your time, you know, you're really worried about quality output and retention.
• 39:02
  SPEAKER_00: That's the lifeblood uh of any agency.
• 39:05
  SPEAKER_00: So let's pretend I'm uh I'm an e-commerce, I'm a beauty brand.
• 39:09
  SPEAKER_00: Who doesn't love a good beauty brand?
• 39:11
  SPEAKER_00: Uh what what's it what's it like working together?
• 39:14
  SPEAKER_00: How do we go from discovery to to the first 90 days if if we're a good fit and we decide to work with with your agency?
• 39:22
  SPEAKER_01: Yeah, absolutely.
• 39:23
  SPEAKER_01: So I mean when it when when we onboard, we have a 30-day onboarding process and we go deep, right?
• 39:29
  SPEAKER_01: We go really deep.
• 39:30
  SPEAKER_01: So we're we're saying, okay, 2026, what's your goals for the next five years?
• 39:34
  SPEAKER_01: What what do those numbers look like?
• 39:36
  SPEAKER_01: Do you want to go five, 10, 15, 20 million, right?
• 39:39
  SPEAKER_01: That's if that's the pace of growth.
• 39:41
  SPEAKER_01: Now what how is the business plan going to support that?
• 39:44
  SPEAKER_01: We're just one piece of it.
• 39:46
  SPEAKER_01: Like, yeah, we were just one piece of it.
• 39:47
  SPEAKER_01: We need to make sure that all the pieces in the puzzle make a coherent picture, right?
• 39:52
  SPEAKER_01: Once we have that down, it's like, okay, we have our goalpost, right?
• 39:56
  SPEAKER_01: Then we go into all the research that we're talking about.
• 39:59
  SPEAKER_01: We do audience research, we do personas, creative research, competitive gap, where and then we start to figure out the levers, right?
• 40:06
  SPEAKER_01: Like, well, is the biggest lever to pull right now?
• 40:08
  SPEAKER_01: Your email program.
• 40:10
  SPEAKER_01: Maybe you have two million emails and you're just really getting against benchmark or against market, half of what you're supposed to be getting.
• 40:18
  SPEAKER_02: Right?
• 40:18
  SPEAKER_01: Or maybe your site converts at 2%, but we know beauty converts at 8%, right?
• 40:22
  SPEAKER_01: So we we find these big gaps and we do a very comprehensive audit.
• 40:27
  SPEAKER_01: Once we do the audit, we need to clean up the data, right?
• 40:30
  SPEAKER_01: Like if we're gonna be working and we're being held responsible to the results we're driving, we need to make sure that the analytics, tracking pixels, the fees, dashboards, how you measure incrementality, all that is aligned, all that is clean, all that makes sense to you.
• 40:47
  SPEAKER_01: We agree you you know, with you, you agree with us.
• 40:50
  SPEAKER_01: So making sure that all that's put together.
• 40:52
  SPEAKER_01: Once we once we get there, then we you know build our channel campaign strategy.
• 40:58
  SPEAKER_01: We say based on everything we've seen, based on where the goalposts are, this is your gap, and this is how our channels are going to cover it.
• 41:05
  SPEAKER_01: And this is how the next 30, 60, 90 days for those channels, here's the playbook for.
• 41:10
  SPEAKER_01: So that's like a 40,000-foot view of how we operate.
• 41:13
  SPEAKER_00: I like that though.
• 41:14
  SPEAKER_00: And I hear like too, it's like you go in and you get an idea of the landscape, and then you you're kind of going in at the same time and like you know, getting goals and helping to align on expectations, and then you're and then with that information, you're using that to put together a strategy, which makes sense because you can't come in on day one and uh obviously just start pulling levers.
• 41:36
  SPEAKER_00: You have to know what what and where to invest your time.
• 41:39
  SPEAKER_00: Um, so anyway, you're your team, you guys work across a lot of verticals.
• 41:44
  SPEAKER_00: You guys do meta, Google, TikTok, email, sms, cr o and sco.
• 41:49
  SPEAKER_00: How do you recreate systems and processes without rem while still maintaining the strategic differences between the brands and the channels?
• 42:00
  SPEAKER_01: Yeah, so I think the easy answer to your question, the short one, is everyone needs to be strategic, right?
• 42:07
  SPEAKER_01: In this in this day and age of AI, everyone needs to be able to be a PL.
• 42:12
  SPEAKER_01: Everyone needs to be able to figure out when and how a business grows just from a macro example, right?
• 42:18
  SPEAKER_01: Uh so that's number one, right?
• 42:20
  SPEAKER_01: Then number two, once we get that part down, you have like a really good baseline, we need to train AI fluency.
• 42:26
  SPEAKER_01: And like I said, there's different levels to that, right?
• 42:29
  SPEAKER_01: Everyone's got a different level, right?
• 42:30
  SPEAKER_01: I would say that a lot of our team is at one level one, two, three right now, right?
• 42:35
  SPEAKER_01: They're not yet at you know three, four, right?
• 42:37
  SPEAKER_01: So making sure that you have AI fluency.
• 42:40
  SPEAKER_01: And then after that, what's the AI culture?
• 42:44
  SPEAKER_01: What is acceptable AI use and what's unacceptable AI use?
• 42:48
  SPEAKER_02: Interesting, right?
• 42:49
  SPEAKER_01: I think that's really big because uh a lot of my team members, because they've gone through like LLM training, they know what hallucinations are.
• 42:57
  SPEAKER_01: They know when it's making stuff up, right?
• 42:59
  SPEAKER_01: Just tell us that you look at it, and then they have to be able to look at any output and and you know, like again, smell test that smells them off, right?
• 43:08
  SPEAKER_01: Because if they ever just take output from LLMs and plop it to me or a client without that review, that's stripe.
• 43:17
  SPEAKER_02: Yeah, right?
• 43:18
  SPEAKER_01: That's a strike.
• 43:19
  SPEAKER_01: It's very, very dangerous, and you can only have a couple of those, right?
• 43:22
  SPEAKER_01: Yeah, so you you put the AI culture in in place, then after that, you know, you you get really organized.
• 43:30
  SPEAKER_01: So let's say you have someone who is doing SEO and meta apps, right?
• 43:34
  SPEAKER_01: Just making meta.
• 43:35
  SPEAKER_01: Then you know, for their skills, like what they have access to, it would be SEO and meta, right?
• 43:40
  SPEAKER_01: And then ideally, all of it through our JetFuel HQ, meaning that they're not, they can run it in there, right?
• 43:46
  SPEAKER_01: So they don't have to be like a master at prompting to get the same output because the JetFuel HQ, you know, our we have our own MCP, they can just plug into that and it makes it consistent for you, right?
• 43:59
  SPEAKER_01: You know, obviously.
• 44:00
  SPEAKER_01: We want people to be fluent enough where they understand what's happening, uh, but we don't necessarily need or want everyone to be like level 100 master at prompt engineer, yeah.
• 44:11
  SPEAKER_00: Right, yeah, yeah.
• 44:12
  SPEAKER_00: It's it's hard.
• 44:13
  SPEAKER_00: Uh, some skill sets don't align.
• 44:15
  SPEAKER_00: Somebody who's like uh charismatic and maybe a good account manager is probably not gonna have the same skills as somebody who's hands on keyboard doing PPC maintenance.
• 44:25
  SPEAKER_00: You know what I mean?
• 44:26
  SPEAKER_00: Like I totally understand how you'd want to lean as much on skills and consistency with AI.
• 44:33
  SPEAKER_00: So that's a really good answer.
• 44:34
  SPEAKER_00: Thanks for sharing that.
• 44:35
  SPEAKER_00: Yeah, um, awesome.
• 44:37
  SPEAKER_00: Um, thanks for coming on today.
• 44:40
  SPEAKER_00: We covered a whole bunch about AI and it was a really great topic.
• 44:43
  SPEAKER_00: Uh, we have a tradition here.
• 44:45
  SPEAKER_00: I'm gonna ask you to make a prediction about the future, even though I feel like this whole episode was future future-facing.
• 44:52
  SPEAKER_00: But over the next 12 to 24 months, which marketing skill becomes more valuable as AI takes over more repetitive tasks and execution?
• 45:03
  SPEAKER_01: It would be tastemaking.
• 45:05
  SPEAKER_01: It would be tastemaking, tastemaking because AI cannot taste make.
• 45:08
  SPEAKER_01: AI cannot predict the next trend, it can next can cannot predict the next viral video, it cannot say, hey, your coffee brand needs to stand out in the market by doing XYZ.
• 45:20
  SPEAKER_02: Right?
• 45:21
  SPEAKER_01: AI trains on data and it looks backwards.
• 45:24
  SPEAKER_01: Not really good at establishing what the path is forward.
• 45:28
  SPEAKER_01: So as marketers, we need to be really good at okay, yes, AI would take the repetitiveness out.
• 45:34
  SPEAKER_01: I don't have to do so many data polls, and it gives me a lot of stuff on the silver platter.
• 45:40
  SPEAKER_01: But I need to look at the silver platter and determine okay, well, for us to really be memorable to a customer for the next five years, we need to try XYZ.
• 45:51
  SPEAKER_02: Yep.
• 45:52
  SPEAKER_01: Yeah, so that is still a very human superpower that we all need together.
• 45:57
  SPEAKER_00: Yeah, I like that.
• 45:58
  SPEAKER_00: Taste agent almost.
• 45:59
  SPEAKER_00: And it's uh, and I think there's always gonna be there's always the human element as AI continues to build out.
• 46:05
  SPEAKER_00: Edwin, thank you for joining me and sharing your experience.
• 46:08
  SPEAKER_00: Before we wrap up, where can our listeners find you?
• 46:11
  SPEAKER_01: Yeah, you could go to our site at jetfield.agency or go to LinkedIn, type in Edwin Choi.
• 46:17
  SPEAKER_01: I should be the first one, I hope.
• 46:19
  SPEAKER_00: Yeah, I bet you are, and all links will be in the show notes wherever you're tuning in, Apple, YouTube, Spotify.
• 46:25
  SPEAKER_00: Uh, you can find Edwin's information inside the show notes.
• 46:28
  SPEAKER_00: Edwin, thanks for joining us.
• 46:30
  SPEAKER_01: Absolutely.
• 46:31
  SPEAKER_01: Thanks for having me, all.
• 46:32
  SPEAKER_00: That wraps up today's episode of Selling on Giants with Edwin Choi.
• 46:37
  SPEAKER_00: The biggest takeaway from this conversation is that you should not begin by building an AI agent.
• 46:44
  SPEAKER_00: You should begin by documenting how you already do the work.
• 46:49
  SPEAKER_00: Write down your objective, the data you review, the steps you follow, the judgment calls you make, the common exceptions, and what a high-quality result looks like.
• 47:01
  SPEAKER_00: That playbook contains the experience and perspective that Edwin describes as your business's DNA.
• 47:10
  SPEAKER_00: From there, Edwin recommends moving through three levels.
• 47:14
  SPEAKER_00: First, learn how to communicate effectively with tools like ChatGPT or Claude.
• 47:22
  SPEAKER_00: The second, turn a recurring process into a skill that produces consistent output.
• 47:28
  SPEAKER_00: And third, once the skill works reliably, schedule it or connect it with other skills to create an agent.
• 47:36
  SPEAKER_00: The key is not to automate too early.
• 47:39
  SPEAKER_00: Edwin recommends running a skill repeatedly, refining the instructions, and confirming that it produces dependable results before allowing it to operate automatically.
• 47:50
  SPEAKER_00: His examples show where this becomes valuable for e-commerce teams.
• 47:56
  SPEAKER_00: AI can continuously monitor competitors, identify gaps in advertising, product pages, analyze customer behavior, and surface conversion testing opportunities that once required hours of manual research.
• 48:12
  SPEAKER_00: But the human still owns the final decision.
• 48:15
  SPEAKER_00: Edwin's team uses AI to place place the research and data on a silver platter.
• 48:21
  SPEAKER_00: The strategist determines the story, decides what matters, and protects the quality of the final work.
• 48:27
  SPEAKER_00: That also changes how businesses should measure AI.
• 48:31
  SPEAKER_00: Time saved and additional output matter, but so do quality, rework, employee adoption, client retention, and value created from the same amount of effort.
• 48:43
  SPEAKER_00: Edwin's prediction is that tastemaking becomes one of the most valuable marketing skills.
• 48:50
  SPEAKER_00: AI is excellent at studying what you or what has already happened, but people still need to determine what a brand should do next.
• 49:00
  SPEAKER_00: To learn more about Edwin and his work, visit jetfuel.
• 49:07
  SPEAKER_00: Both links are available in the show notes.
• 49:10
  SPEAKER_00: If you found this episode helpful, please follow, subscribe, like, leave a review for selling on giants.
• 49:16
  SPEAKER_00: It helps more e commerce operators discover this show.
• 49:19
  SPEAKER_00: Thanks again for tuning in, and we'll see you next time on Selling on Giants.
• 49:25
  SPEAKER_00: This is Mr.
• 49:26
  SPEAKER_00: Will signing off.