Transcription
In this video, I'm going to show you the only four ways I'd use AI to make money in e-commerce right now. And I want to be really clear. I do not mean four AI tools. Tools are the trap everyone is falling into. I mean four places inside a real e-commerce brand where AI can really move revenue, margin, or improve speed without blowing something up.
Because over the past 12 months, I've tried putting AI into so many e-commerce brands, from brands just starting out all the way to my own brand that's doing over a hundred million dollars a year. I've tested things like AI ad accounts, email optimization, optimizing product pages, customer service, finance reports, 3PL invoices, creative, and so much more.
And the weird thing I learned is that the most profitable AI workflow is probably not an AI media buyer. It's also not some magic AI store builder. It's probably something so boring most founders will skip.
Now, here's the issue. Most people ask what AI tool should I use? Should I use Coda, ChatGPT, or Gemini? Or should I use something random on Twitter, a random app that someone says that they can automate an entire company? Almost always, that is the wrong question. The better question is, "Where is my e-commerce brand already leaking money, and how can AI help me fix that one step within that workflow?"
Because AI does not make money because you bought a tool. It makes money when you can connect it to a business metric such as cost per creative or email as a percentage of revenue. Then get AI to fix that instantly. That is the game.
Now, the first way I've used AI to make money in e-commerce is by optimizing email automations. And I'm being really specific here. I do not mean use AI to write a random campaign. Most brands already know they can use ChatGPT for a Father's Day email or a Black Friday email or whatever campaign they're about to set. That is fine, but that's not where I'd want to start.
The place I would want to start is automations, because for a lot of brands, that is where the majority of sends are actually happening. I'm talking about your welcome flow, abandoned cart, browse abandonment, post purchase, win back, replenishment, VIP, all of these flows that are sitting there sending every single day. And the crazy part is, a lot of brands barely optimize these automations. They build the flow once, maybe update it every now and then, and then it just sits there making a large percentage of revenue in the background, but it's not optimized. This is the perfect place for AI.
Now, quick disclaimer before I say any of this. In this whole video, if you're connecting something to AI, there is some level of risk. So, make sure you're comfortable with the level of access that you give it. But, for email, what I would do is this. Step one is I would use Chord Code or Codex and connect it through an MCP server or API into the email provider that you're actually using. If that's too technical for you, you can just use screenshots and actually paste everything into one of these tools, but I would much rather it be automatically able to pull any of this data that you actually want. I'd get it to look at every major automation and report back on revenue players, order rate, click rate, conversion rate, unsub rate, revenue per recipient, and where people are actually dropping off in the flow.
Then, step two, I'd get it to produce a report every single month. Not just some giant dashboard that nobody ever reads or not just a one-off that you don't keep coming back to. I want it to come back and basically say, "This month, you only tested two automations. Your abandoned cart flow is doing okay, but email two is getting opened and not click. Your welcome flow has not had a proper split test in months. And if I was going to test one thing, this is exactly what I would test first. This section or this subject line." That's the first job here for AI is analysis. It's almost always the first job for AI. Before it writes anything, it should tell you where the split test should actually go. Because this is where most teams waste time, and actually AI can often be better than the team member analyzing this. Often, team members will split test the email they feel like changing, not the email where the test has the highest chance of making more money. So, the AI should come back and say the abandoned cart second email is getting opened but not clicked. Try testing the hero section and the objection handling or even putting an offer in this email.
Now, the second part is where AI can actually help create things. Once it's chosen the best split test, then I would ask it to rewrite the section being tested. If it's a subject line test, it might give you 10 subject lines that you can choose from. If it's a hero section, it'll rewrite the hero and give you three of those options.
What's really cool that a lot of people don't understand is you can actually write these emails in HTML in something like Codex or Coral Code and paste that in your email provider and get beautifully designed emails that are very, very easy for the AI to split test. Now, this is a massive sticking point for brands that maybe don't have a designer or not doing email optimization because of the design element. They don't want to hire someone from Upwork. This is a great way to actually do it. You can actually just ask Coral Code or Codex to design beautiful HTML emails and you can split test those two things. But again, I really think humans should be reviewing all of this before it goes live. I don't want AI randomly changing live automations. I want AI to tell me where the money is every single Monday and then create the test for me and actually roll it out and actually then report back on what the test did.
Now, obviously that's just a very quick overview of how it can work with the email, but let's not forget this. If you're ever stuck with the process that I just described, all you need to do is actually ask the AI itself, give it the objective of what I just described of creating that report every single week, creating the emails and let it work backwards. And whenever you get stuck, just screenshot it and ask the AI what's the next step. For example, if you're confused about what CLI is, MCP is, APIs are, then you just need to ask AI and say what would you recommend to get this level of access. You can even ask it what's the security risks of doing this. I don't want you to listen to any of the ideas that I have today and say I'm not technical enough to actually do this without actually going into the AI and actually trying.
The second way I would use AI to make money in e-commerce right now is ad creative intelligence. And I want to remove one idea straight away for beginner. I do not think the most interesting use of AI in Facebook ads right now is letting it make budget decisions. There are a lot of people right now trying to use AI to say increase the budget and turn this ad set off, decrease the budget if these things are actually happening in the ad account. And I get why that feels exciting. But truthfully, Facebook has had rules and actually automated rules and campaign budget optimization from a long, long time ago that will probably satisfy a lot of this criteria. The big and simple opportunity with using AI in Facebook is to understand what creative is actually working and then making it way easier to get more of that creative into the account.
Think about the average ad account. There might be hundreds of ads, different naming conventions, UGC, founder videos, statics, product demos, before and afters, all of this kind of stuff and it's all messy. And most teams are still basically guessing. They know the average CPA and ROAS, but they don't really know what creative pattern is driving the account because the learning is buried inside random naming conventions, spreadsheets, screenshots, Slack messages, and someone saying I feel like this worked, let's do a little bit more.
So, the first thing I would do is fix the naming convention enough that AI can actually read the account. I would then pull the last 30, 60, or 90 days with spend, purchases, ROAS, even the hook type, the format, and the avatar, as well as the asset type. And if the naming convention is messy, that is actually a great first job for the AI to actually go in and fix all of that. Then you go in and ask it to help classify the account into something useful. So, you might want to understand if it did founder-led product demos really work versus UGC ads. Is the before and ad comparison really working? I want it to come back and say the ad account is spending most efficiently on these types of formats. Statics are getting spent but not converting. And the last three winners all use this problem solution hook in the first 3 seconds. That is actually useful.
Then step two where AI is really good with Facebook is creative volume. You should get the AI to look at how many new creatives were introduced over a period of time versus the amount of spend going through the account or the results the account is getting. Because if you're spending a lot and barely introducing new creatives, you're probably just waiting for the fatigue to actually hit you in the face. So, I want the report to say you spent this much in the past 30 days, introduced this many creatives, and based on your current spend, you probably need to introduce this many more tests based on the rest of the market or competitors introducing way way more creatives. That is a kind of analysis a founder can actually use.
And then step three is competitor and adjacent niche research. But this is where I'd be careful. The Facebook Ads Library is really useful, and there are tools that can help you look at ads that have been running for a long time and appear to get a lot of reach because you can now sort by impressions on the Facebook ad account. But Meta is strict around automated data collection. So, I would not connect random scraping tools to my business manager or build some aggressive scraper and pretend there is no risk to my actual account. A safer way of doing this, but there is still some risk, is actually using approved APIs where they actually apply, use reputable tools, keep permissions as narrow as possible, and reduce the platform risk.
This is why I was so excited about the tool Mana because Facebook seemed to buy that and it was an agent that could go in and analyze ad accounts, and I knew that they were connected. But that's kind of up in the air whether or not that transaction is happening. So, you could just make sure that you connect nothing of your Facebook ad account. Use something like Codex or Core Code and just use URLs where you don't have to log in, scrape that not aggressively, and then connect it through the Gemini API to actually analyze the creative and get really good reports about what the industries are doing.
The fourth step with Facebook that AI can really help with is the actual operations of it. A lot of people would be using Google Sheets or project management tools like monday.com or ClickUp to actually manage their creative inputs. I've seen a lot of people set up systems where all you need to do is upload the creative to Google Drive and simply click approved in some of these tools. The AI will actually automatically upload it as a draft not spending or it just up into the ad account so that media buyers can quickly go in and grab the creative rather than someone having to manually upload it.
The third way I would use AI to make money in e-commerce right now is the creative creator and influencer content. If you're in the growth stage, maybe doing 100,000, 200,000 dollars a year, this might be the most practical tip today. Because at that stage you need more content, you need more creators, you need more people talking about your product and you probably do not have a huge team to manage all of it. But the part I think most brands get wrong is what they're actually trying to measure. They measure things like how many creators they message. They measure how many people actually reply. Maybe they measure how many videos came back. In the early stages of reaching out to creators, I want to actually know who is excited and who is pumped to get their first product and who is super transactional. This might be indicated by them just sending back a rate card before they've even touched the product or even entertained the idea of a collaboration. You'll often get this with people that are over 20,000, 30,000 followers. Because I don't really want my creator pipeline filled with people who don't care about my product.
So this is where AI is really useful. Not to spam 1,000 creators with the same fake personalized message. This is how you ruin your brand. What we actually want is a pipeline where AI helps one smart team member manage the whole conversation properly. It can find the creator handles, summarize where they fit, draft the first outreach message, maybe even with a voice message from the founder. And then once they reply it can start scoring the conversation. Not just positive or negative. I want it to tell me this creator seems genuinely excited about the opportunity. Then I would have it all feed into a dashboard into monday.com or ClickUp or Airtable, whatever you're actually using. It'll have the creator, the sentiment score, the latest conversation summary, and the next recommended action all sitting in one place.
And once the team is happy, all they should have to do is change the status to approved, then AI can trigger the next step using Shopify's AI toolkit or API. It creates a 100% off discount code, one use only, capped properly so nothing weird can happen, and sends the creator the email with that code. And this is important, using a discount code means that you don't actually need to collect their address manually. It's so much more streamlined. They can then go through the normal store like a customer, choose their product, and actually everything gets tracked properly. Then you can also create a second code for their followers, so they get their one-time product code, and then they get their public code that they can share with their audience.
Now the AI can also track what actually happened, whether they posted, whether the content came back, whether the code drove revenue, and whether the content actually turned into a powerful ad that performed. And this is where AI becomes really good, because influencer managers are often terrible at circling back. The AI should be circling back automatically. Not in a creepy way, just normal follow-ups. Like, "Hey, just checking whether the product actually arrived." Or, "I would love to see did it fit?" Or something like that. You could even ask, "If you posted already, can I see the link?" The key here is also to have a human touch in this process. You don't want the AI handling everything, you want to build a really strong relationship with 10 to 20 creators. But by automating the rest of the process, you can really apply your focus to these key creators.
The fourth way I would make money using AI right now is to actually build an AI finance and operations control system. And this is where I'd put a few things together. I'd put profit leak audits, inventory management, customer service cost, fulfillment, sales commissions, agency invoices, finance reports. All of these fit into one category, in my opinion. It is anything in the business where a cost changes based on what actually happened, like a consistent percentage cost. If there's a variable cost tied to real activity, AI can usually help check whether the numbers that you've actually been charged actually make sense. And honestly, this is probably one of the biggest opportunities for most e-com brands because nobody likes to talk about 3PL invoices, but the P&L does not care what is sexy. The P&L cares about where the money is going.
Now, big disclaimer is this is something where like Open Claw might actually make sense, like an open agent that's scanning everything, but I would not recommend this especially for beginners. It is not simple, and if you do not understand AI, APIs, permissions, all of that kind of stuff, something could break and something catastrophic could happen. Personally, I would start much simpler. I would use something like Claw Code or Code X, probably a cron job, and build a very narrow agent or skill that runs the same audit every week or every month.
The first thing I would probably build is the AI CFO layer simply for 3PL audits because that is a huge cost for most businesses. So, with fulfillment, it pulls your Shopify order data, total orders, approximate weight, delivery zones, all of that kind of stuff, and it will actually compare it to the 3PL invoice that you've actually got. It might say something like these pick fees look higher than expected. This weight band does not make sense. These delivery rates look worse than last month. Then a human can go through and check it. You're not letting your AI email your 3PL and start a fight. You're using it as a point of reference so that you don't have to go through line by line in spreadsheets.
We've actually built this full process in Daily Mentor. You get an AI-powered report card every single month so you can see all of these key focuses that you should do and analyze a ton of financials. So, if you're interested in that, make sure you click the link and apply, and we'll help you out with our mentors.