Your AI Strategy Is Really a Data Strategy

The biggest limitation with AI in marketing isn’t the intelligence of the models anymore. It isn’t the prompt, the wrapper, or even the general business context you give them.

It’s the data.

AI needs the same raw information a good consultant would request before making a serious recommendation: sales history, customer behavior, inventory, pricing, promotions, margins, operational constraints, and the relationships between all of those things.

Once I saw what happened when our team connected AI to that kind of data, I changed the way I think about AI strategy. The question is no longer, “How do we prompt this better?” It’s “What information can we responsibly make available so it can help us think?”

That shift has practical consequences for every marketing leader. If you want AI to make useful strategic recommendations, you’ll need to work with IT, consolidate your data, protect customer privacy, and bring your own judgment to the results.

AI can’t make strategic decisions from disconnected data

I work for Trim Healthy Mama, an e-commerce company that sells health and wellness supplements. E-commerce has a particular kind of complexity: you’re not just trying to generate demand. You’re trying to generate the right demand at the right time.

We have roughly 65 unique products that become around 400 SKUs when you account for different sizes, flavors, and packaging. Many products share ingredients. Some ingredients and finished products have expiration dates. Suppliers and co-manufacturers operate on different timelines. Inventory can be too high, too low, or headed toward expiration depending on the product.

That creates a difficult marketing question: what should we promote this week or this month?

The answer can’t simply be, “Promote the product with the highest margin,” or “Push the product that sold best last month.” We might need to sell down an overstocked item. We might need to accelerate sales on something approaching expiration. Or we might need to avoid promoting a top seller because we’re about to run out and won’t be able to replenish it quickly.

Promote the wrong product today, and you can create an inventory problem three months from now.

Other business models have different versions of the same problem. Enterprise software companies may not have inventory, but they have complex buying cycles, multiple sales touches, account-level data, and dozens of variables across the funnel. Lead generation businesses have their own set of channel, conversion, and customer-quality questions.

Every business has data complexity somewhere. The challenge is that the information usually isn’t in one clean place.

The real work is making your data usable

At our company, important information lived across WooCommerce, ShipStation, spreadsheets, and other systems. The data didn’t always line up. Pulling a simple answer could require a manual investigation across several tools.

Questions that sound easy in a meeting could become surprisingly difficult:

• How long does it typically take a customer to make a second purchase?
• What is the lifetime value of our 2024 customer cohort?
• Which products create the strongest repeat-purchase behavior?
• Which promotions have actually moved revenue?
• What should we promote now without creating an inventory problem later?

Before we had a connected data source, answering those questions took time and often involved educated guesses. We could look at individual reports, but strategic analysis requires connecting the reports.

Our CTO, Ron Brouwer, led the effort to create a database that synchronized the information from our different systems. Because he understood both the technical side and the operational side of the business, he knew what the data meant and how our teams actually used it.

That combination mattered. A technically elegant database that doesn’t reflect how the business works won’t help much. The goal wasn’t merely to store more information. It was to create a source that could compare the information and make it accessible to the people making decisions.

We also had to think about privacy. We didn’t want to send sensitive customer information directly into an AI tool. The solution was to anonymize the customer data so we could analyze patterns and cohorts without exposing personally identifying details.

That’s a useful principle: make the data available, but don’t make private information available unnecessarily.

What happened when we connected AI to the database

Once the database was ready, our team received read-only access through an MCP connection. That meant we could use tools such as Claude, ChatGPT, or Codex to ask questions about the business without giving the AI permission to alter the underlying data.

The first time I saw it analyze the data, it was both impressive and a little terrifying.

We were trying to figure out how to bring more revenue forward. Instead of starting with a collection of guesses, we asked Claude to review our data and propose a promotional plan.

It examined inventory levels, expiration dates, historical promotions, and product performance. We also gave it access to a Google Sheet containing three years of week-by-week promotions. That sheet didn’t contain all the corresponding revenue data, but it gave the model a useful record of what we had tried.

The resulting plan explained what to promote, when to promote it, and why. It identified patterns in our previous promotions and recommended several plays that had historically moved the needle. It also avoided simply pushing products that could create problems later.

I know how difficult that analysis would have been for me to complete manually. I would have needed to find the relevant reports, reconcile the numbers, check inventory, review past promotions, and then make the strategic decisions. Even with a full week, I might not have produced something as comprehensive.

AI didn’t replace the marketing work. It gave us a much better starting point.

This is the difference between asking AI to write a promotion and asking it to help decide which promotion is strategically sound. The first task needs creative context. The second needs business data.

Use AI as an analyst, not an unquestioned decision-maker

The recommendation still required human judgment. The model didn’t know everything I knew about our audience because it hadn’t talked to our customers. It didn’t understand every emotional or cultural nuance behind the products. I have relationships with the audience, and I’m close enough to the customer that I can recognize when a technically sensible idea won’t land the way the data suggests.

So I reviewed the recommendation, challenged parts of it, and added context. I could say, “This week doesn’t work because of a factor you don’t know about,” and continue the conversation from there.

That human contribution isn’t a temporary inconvenience. It’s part of the job.

AI can quantify patterns and analyze more variables than most teams can handle manually. Humans still supply values, convictions, relationships, taste, responsibility, and lived experience. We decide which goals are worth pursuing and which tradeoffs we’re willing to accept.

That’s why I don’t think a less experienced person automatically gets the same results just because they have access to the same model. Someone with years of experience understands the history behind the numbers. They know which metrics are misleading, which customer objections matter, and which recommendations need more scrutiny.

The strongest workflow is collaborative:

1. Give AI access to trustworthy, relevant data.
2. Ask it to find patterns and explain its reasoning.
3. Check the inputs and assumptions.
4. Add the context the data can’t capture.
5. Make the final decision as the accountable human.

This fits with the broader paradox of AI: it can be extraordinarily capable in some areas while remaining dependent on human direction in others.

Marketing leaders need an IT relationship

If you’re a marketing leader, you can’t treat data access as someone else’s problem. You don’t need to personally build the database, but you do need to help define what the business needs from it.

Start talking with your CTO, IT team, operations team, finance team, and anyone else who owns a meaningful piece of the customer or revenue data. Explain the questions you need to answer. Bring real marketing decisions into the conversation.

IT may initially think in terms of one system or one type of data. Marketing often needs the connections between systems. For example, sales data alone might not tell you much about customer lifetime value. You may need customer cohorts, product history, purchase intervals, subscription status, and promotion history together.

I had to go back to our CTO and point out that we had sales data but not enough customer data to answer the questions I cared about. That led to a better solution: anonymized customer information that could be analyzed without creating an unnecessary privacy risk.

These conversations work better when marketing brings specific questions instead of vague requests for “more data.” Ask for the information required to make a decision:

• What should we promote?
• Which customer group should we prioritize?
• Where is the bottleneck in the funnel?
• Which product has the strongest repeat-purchase potential?
• Where are we underperforming against relevant benchmarks?

When operations, procurement, marketing, and IT can all ask questions against the same source, the quality of the discussion improves. The team moves away from “this feels like a good idea” and toward “here’s what the data says, here’s what it doesn’t say, and here’s the judgment we need to apply.”

That’s much healthier than copying a competitor’s tactic without knowing whether it worked for them or whether it fits your business.

Small and medium-sized companies have a real advantage

One unexpected advantage here belongs to small and medium-sized companies. Large companies may have more data, but they often have more silos, more approval layers, and more systems that don’t communicate well.

A smaller company can sometimes move from scattered data to a useful shared system much faster. If it has been operating for five or ten years, it may also have enough history to make the analysis meaningful.

A startup with almost no customer or sales history has fewer options. It may need to rely on public data, benchmarks, and assumptions. An established company with real traction can analyze what has worked, what hasn’t, and how different customer groups behave over time.

That combination—enough data and the ability to move quickly—can become a significant strategic advantage.

The companies that build these connections earlier will be able to make better decisions sooner. This isn’t just about saving time on reporting. It’s about increasing the speed and quality of strategic learning.

A practical starting checklist

You don’t need to begin with an enormous data project. Start with one recurring decision that currently requires too much manual work or too much guesswork.

1. Choose a high-value question. Pick something tied to revenue, retention, inventory, or conversion—not a vanity metric.
2. Map where the data lives. List the platforms, spreadsheets, databases, and teams that own the relevant information.
3. Identify the gaps. Decide which missing fields or disconnected systems prevent a reliable answer.
4. Define privacy boundaries. Use anonymization and read-only access where appropriate. Don’t expose sensitive information simply because you can.
5. Build a shared source. Work with IT to synchronize and standardize the data needed for the selected question.
6. Test the analysis. Ask AI to show its assumptions, cite the underlying data, and identify uncertainty.
7. Apply human judgment. Add customer knowledge, brand context, and business constraints before acting.

Once that workflow works for one decision, expand it. You might move from promotion planning to cohort analysis, inventory forecasting, customer retention, or channel allocation.

This is also why I’ve become less interested in treating AI as merely a content-generation tool. The bigger opportunity is strategic analysis. If you’re thinking about the limits of automating marketing work, the same lesson applies: the quality of the result depends heavily on the quality and accessibility of the underlying information.

The next AI advantage is connected judgment

AI models are already powerful enough to analyze a substantial amount of business information and carry on a long, useful conversation about it. There will be improvements, but marketing teams don’t need to wait for some future model before starting.

The constraint is access.

Can your AI see the information required to understand the decision? Is that information accurate, connected, and current? Can your team explain what the numbers mean? Can you protect customer privacy while still giving the system enough context to be useful?

If the answer is no, better prompting won’t solve the problem.

Start building the relationships and infrastructure that make responsible data access possible. Then use AI to analyze the information, expose patterns, and give your team a stronger first draft of the strategy.

The marketer’s role isn’t disappearing. But it is moving upward—from manually gathering every number to deciding which questions matter, which evidence is trustworthy, and what the business should do next.

That’s the work I’d focus on now: pick one important marketing decision, bring IT into the room, connect the data, and see what becomes possible when AI has something real to analyze.

For the human side of that transition, I’d also revisit the Human Edge Framework. The data gives AI leverage. Your judgment determines how that leverage gets used.

AI in marketing is limited by the availability of data for strategic analysis and recommendations. Unlocking data and building relationships with IT are crucial for leveraging AI effectively in marketing. Small and medium-sized companies have an advantage in leveraging data for strategic decision-making. Takeaways Data availability is the key limitation for AI in marketing Building relationships with IT and unlocking data are crucial for leveraging AI in marketing Chapters 00:00 The Limitation of AI in Marketing 01:14 Unlocking Data for AI in Marketing 02:05 Challenges in E-commerce and Inventory Management 03:28 Complexities in Marketing Decision-Making 05:48 The Power of AI in Strategic Analysis 06:11 Collaboration with IT for Data Access 07:03 The Role of Data in Strategic Decision-Making 13:19 Advantages for Small and Medium-Sized Companies

Dan Sanchez: The biggest limitation with AI and marketing is no longer the intelligence of AI. It's not even the wrapper or the tools that you give it. It's not even the context, given it the general frame of the business and all the general necessary things that anybody would need to know in order to do marketing for your business. It's the data. It's the raw amount of information in order for it to actually do strategic analysis in order to make real recommendations. It's the same kind of data that any consultant you hire would ask for and have to dig deep into in order to make very strategic choices for you. And that's what AI needs as well. I saw something just two weeks ago that blew my mind. And I've been thinking about it ever since then. In fact, it wasn't even two weeks ago, it was just one week ago, and I've been doubling down on this ever since. For a lot of people, this might not even be new, but for me, it hit different. It was something that was awe inspiring and slightly terrifying at the same time. So today I want to talk about how you can unlock your data And what you're going need to do as a marketing leader in order to make the most of this shift when it comes to AI. Welcome back to the AI-driven marketer. I'm Dan Sanchez. My friends call me Danchez. And I find that more and more it's not about how we prompt AI anymore. It's about the all the resources and access and data we give it. and the more I do, the more I find that AI is making great strategic choices. Now it's not In need it's still going to need your help. It's still going to need your guidance. And we're gonna talk about that in a bit. But the thing that I saw just a week ago was AI properly equipped and given a whole database of information. As you know, I work for a company called Trim Healthy Mama. It's an e-commerce company selling dozens and dozens of different health and wellness supplements. And what I've learned getting into the e-commerce space is that e-commerce Has some I mean, every every business model has its own difficulties. E-commerce, it's the logistics and all the problems that can go wrong when you have lots and lots of skews. And we probably have 400 skews, probably about really 65 unique products and then repackaged in different sizes and flavors and all that kind of stuff to be 400 different skews. And each product for us, I mean, we're t we're dealing with with products that spoil. It's not like vegetables where it spoils in weeks, but you know, it spoils in a year or two years, depending on the ingredients. And we gotta keep track of that. We gotta keep track of inventory. We gotta keep track to the fact that we buy and store some of the ingredients of many of these supplement products. And a lot of products will share from the same ingredient stores that we have. So it's a it's it's it's honestly the the biggest crux of e-commerce, especially around supplements, is managing all this inventory, figuring out what the heck you should be selling when, because some of it's going to be going bad at the current run rate or the current purchase rate of it, and you need to sell it way faster before it all expires. Some of it is you have way too much inventory and need to sell it down because you ordered too much because you thought you would sell more. Some of it is you're running low and you need to reorder a lot faster than you thought. And all the different co manufacturers of this stuff or your own manufacturing lines are deliver at a different speeds. So that creates a whole amount of complexity when thinking about what to promote this week or this month in order to bring revenue forward, but not cr screw yourself three months from now when you're running out of your top seller, which has happened. And that's the complexity. If you're in a lead gen game, just know the product game isn't easier. It is nice that we're selling things that people like collagen, people start taking that daily. They got to come back and buy it the next month. That's the upside of this game that I'm in now is that I have repeat purchases. I have subscriptions that I can sell. So that's all well and good, but the hard part is knowing what to sell and when when it when you have that many products to sell. And every business has problems like this. If you're selling enterprise SaaS and you have a very complex purchase cycle, well, there's complexities all throughout the funnel and all kinds of data points from your sales team. Right? It's no it's and it's it's to the point where it's like marketing's got multiple channels and every account's being hit in different ways at different times, different replies are going out. There's a whole bunch of complexity in enterprise sales too. It's just that it's not stuck in inventory and trying to manage all the different products. You're stuck trying to manage the complexity of the sales pipeline. Every business has some kind of data complexity somewhere. And this is what I saw recently, because this is something where you honestly can't fix by yourself and you're a marketing leader, you're not working alone. And this is the good part. We do not have to figure out everything by ourselves. And I certainly didn't I had a lot of help figuring out this particular issue that we were running into. And of course you can have just hire more people to help you manage the complexity and each person's in charge of something else. One person's in charge of reordering, one person's in charge of tracking pricing, one tr person's in charge of forecasting and making sure all the inventory levels, one person's good at You know, just dealing with all the different things, but you're not always at a place and a size where you can have a dedicated person for all the little tiny nuances of making sure all that data is properly formatted and analyzed so that you can make the right decision. So what do you do? This is where AI is now, just now becoming incredibly, incredibly powerful. And you can do it all for a twenty dollar a month subscription. Even better if it's a hundred dollar a month subscription 'cause you Once you start doing this, you find you're gonna be using AI way, way more. So I worked with IT and actually IT really started taking on the charge of this well before I asked any questions about it. only because they got tired of helping of dealing with all the data issues. So our IT department, Ron Brouwer specifically, I need to have him on the show and interview him because I think he'd be really helpful and insightful for a lot of you. He is our CTO and he's not even like a tech background. He's kind of a jack of all trades CTO, which is why he does really well with tech, because he doesn't just understand the tech side, he understands all the operational sides of this particular company, which means he's not getting stuck in the weeds on the technical details. He's often implementing things that are really helpful. This was one of the big ones for us recently. As we work off of WooCommerce, which, you know, WooCommerce is great because you can customize it, but it's not like i as as stable as something like Shopify, but we've been on it for a long time. We're we're dealing with it. But a lot of this data stuck in WooCommerce and ShipStation and all the other different little software pieces we use in order to manage the whole ecosystem of this e commerce store. And the data's everywhere and the data's complicated, you know, like th all these things don't line up correctly. So he created one database that syncs everything. And pulls in everything and makes sense and compares everything because he was doing it manually for so long, he knew exactly how to vibe code the thing in order to do it. Internally, we call it countracula. He he's got some humor, you know? and it's great. It's a database, we have a big dashboard we can log into to check it. But eventually, like he can only vibe code so many solutions for all of us who have all kinds of questions about the data. Have you ever run into that in a meeting where your boss like, well What's what's the amount of days it takes to somebody go from second purchase to third purchase? What's our lifetime value for our twenty twenty-four cohort? It's like dude, like pulling any of that information before now was incredibly difficult. But if you can get a solid database, and just recently he was able to supply our whole team with an MCP to that database. So that means whether we're using Codex or Chat GPT, whether we're using Claude, it doesn't really matter now. We just tap into the MCP and ask it questions. And it's read-only, so the day AI can go and not make any changes to the database, and it's all anomalized customary data, so we're not pulling anything sensitive. It's just our numbers. And now we can ask it whatever we want. And I'm telling you, when I saw what it could do, it like blew my mind. For example, we're in a meeting of trying to figure out how to get more revenue forward, right? Have you ever been in that meeting? We've all been in that meeting as marketers, right? We like, we need more revenue this month than the next month. What are we gonna do? Well, we asked Claude and it went and pulled the data and came up with the most beautiful promotional plan that I've seen in a long time. And because I've been working in it now, I know how hard it is to do what it did. To the point where I was reading this promotional plan and it was justifying what to promote, when to promote it, looking at inventory stores, looking at when pr products would be expiring, looking at all the different variables, even looking at all the historical promotions we've done over the last three years, which wasn't even in the databases in it a Google sheet that we had that wasn't synced up to anything and didn't have any of the the revenue numbers posted to it, but it had week by week what re what promotions we had done with like a one sentence description of it. So we could actually say, hey. Based on your performance over the last three years, here's the three plays that actually move the needle for you. You're like, gosh, data. Wow. it was amazing to see it look at decisions and quantify everything and use that to make strategic choices about what to promote, when and why, and in what way. It didn't do all the creative, it just put the baseline reports together of what to do week by week in order to move the most num move the most revenue forward without sacrificing selling the wrong thing at too much of a discount later. And it was magic. I'm telling you. It was like looking at it and being like, I couldn't be able to pull this off if you gave me a whole week to try to track down all the numbers and make the most strategic choices. It quantified and analyzed and put together a strategic plan that I was not capable of doing with the current size of my team. And in that very moment I became a little terrified as I was like, my gosh, I am going to lose my job. I won't. Because humans still add a lot to the equation. I still have to be able to read it and understand where it went off, where it still needs to like sh where where it still needs to be double checked, where I might be able to str take inspiration from what it recommended and still tweak it a little bit because it didn't know what I knew about our audience, because it hasn't talked to them, right? I've actually talked to them and actually have relationships with them. Shoot, I'm married to the target audience of this customer of this company and therefore it can Tweak it to the place where I know it needs to be. But the fact that it gave me a starting point that was actually like empirically accurate and strategically sound was such an amazing place to start from. Cause then you could layer at it and you can continue the conversation because it's AI, right? You can talk it through and be like, well, I love your idea here, but this other idea, but the next week is actually off because of X. Let me fill you in on the context. And you can, and it. says, yeah, of course. If I would have known that I wouldn't have made that recommendation. And these are the on the kinds of conversations you can have now, but it's all because it has access to real live data. And I'm telling you, the first time you really hook it up to a data source like this, running the current frontier models, it's going to change the game for you. And you're gonna see what I saw and you're gonna be amazed and slightly terrified. But remember, the human edge is still at play here. The hu only humans can have core values, only humans can have convictions, relationships with other humans, and have the years of experience of the pains and the highs and lows of going through all the different marketing campaigns to actually be able to make sound judgments. a eighteen year old kid with this technology won't be able to execute it the same, nearly the same as a forty year old with twenty years of experience doing this thing. And that should bring some rest to you. But just know that when you start hooking up AI to data, it starts changing everything. So what I recommend is start to build the relationships you need in order to consolidate the data into a source like we did at Trim Healthy Mama. I'm not gonna give you the step by steps on how to do it because it's gonna be diff different for every every one of you. Some of you have a lot of information stuck in a SAP. Some of you have it in a CRM. Some of you have it in Google Drive. It's like it's information's all over the place. It's going to be a collaborative effort across departments in order for you to achieve this. But I promise you, those who achieve this sooner will start making more strategic s decisions much earlier. And this is a game of speed to a large degree. You also This is also again one of those things that's gonna favor the small and medium sized companies. Specifically, I think the smaller medium-sized companies that can move fast on this, but actually have enough data and infrastructure to be able to have data to go off of. You know, companies that have been around five, ten years have way more data in order to analyze and know what works and what doesn't work, versus your startup that has almost no data and only can rely on public records and benchmarks. This is a major advantage to existing companies that already have some traction in the market. So if that's you. This is an advantage. All the small players have had a lot of advantages. This is where medium-sized companies really start picking up steam because they have the data to go off of. So start building those relationships. If you haven't started talking to your CTO and your IT team already, start collaborating on this stuff because once you start working on this together, the whole thing starts to be way more fun. way more collaborative when you're talking to each other and bouncing ideas off each other, you're actually pulling in data. I can already see it from my team because again, he gave MCP access to operations, procurement, my marketing team and the IT guys have it. So they can test my ideas and be like, well according to and they can actually say according to this. Which sounds kind of a like it would be annoying, but it's not because before we were just doing it based on g on feels right. We were coming up with ideas based on feels, but now your team is gonna come to you with Data, which is a much better starting point than being like, well, I saw a competitor do so and so having no data of how it performed for a competitor, right? This is a much better step forward. But it's gonna take you as a marketing leader to start building those relationships and working on these collaborative things in order to build the database that you need. Because if you don't do it, then you won't be able to get what you need out of the database because IT thinks it needs to be one-sided data. I actually had to go back to our CTO and be like, hey, there's It's got all the sales data, but it has none of the customer data. How can we get the customer data in there without making it a privacy issue? Because I don't want to send all the customer data to Clot or Chat GPT, right? So he figured out how to anomalize it, not very hard, and get it in there. And now I'm running. Now I'm calculating all kinds of customer lifetime values, which is always kind of a difficult thing to measure, right? But now I can do it based on cohort, based on product, based on purchase rate and all kinds of different things. And it's changing the game already. I just pulled a report this morning. I'm like, I wish I would have known this earlier. This is going to be amazing. This makes it so much easier to figure out where the bottleneck is, how to approach it. And then not only can you take all your current benchmarks, you can actually go and tell it to do research on what the real world benchmarks are for companies just like Heuristic to see where you're off, where you're way ahead, and where you really need to shore it up in order to make strategic decisions. This is how a consultant works. And now you have one in your pocket. But You have to get the data together in a way that AI can access it, read it, and actually analyze it for you to put that raw intelligence to work. Again, it's no longer about the power of AI. It is now plenty powerful enough to do all the analysis you would ever need to do in marketing. Maybe it gets better in the future, but honestly it's it can't get that much better. It has all the context it needs in order to analyze a pretty good sized company now. It can take all of it in and deal with it and have a fairly long conversation about the data. and it's no longer about the wrapper and the tools. It has all the tools it needs in order to actually give you what you need. Now, maybe there's more gains to be had on the execution side of all this, but as far as strategic business analysis, everything's available now. And it's about you giving it what it needs in order to help you move forward.