How Dan Cumberland Built an AI System That Eliminates the Blank Page Problem

HSB - Ep. 114 Dan Cumberland

Hector [00:00:00] Hello, and welcome to another episode of the How to Scale Business Podcast. My name is Hector Santiesteban, and I'm your host. Our special guest is Mr. Dan Cumberland, and he's an AI strategist consultant. We're gonna talk about something that we hint at on a lot of episodes, but we're gonna really dig in to how all businesses might be able to leverage it, where things are going.

Hector So, Dan, thanks for hanging out.

Dan Thanks so much for having me, so pumped to be here.

Hector So, you know, usually I get into how someone got into their industry and such, and I'm gonna change that question because the industry is so, so new, so fresh. I don't know that it's as important as: at what point did you realize that AI was gonna be something that you were gonna go all in on?

Hector Was there a moment or a thing that happened that said you should, you know, dive in?

Dan Such a good question. It was kind of a slow progression, but I'll tell you how it went. I was running a venture studio build with a core group of about 12 of us, building software projects.

Dan This [00:01:00] was 2019, 2020, 2021, and then it wrapped up due to differences with business partners. I became a fractional CMO, and I was pretty burned out from holding this team together, launching these products. And that was about the time that ChatGPT came on the scene as a consumer-facing tool.

Dan I had dabbled in AI in the venture studio with some of our software. We built the first long-form video to short-form video tool called Video Snapp. It doesn’t function now, unfortunately. But I knew about AI from a developer standpoint. With ChatGPT, it was like, oh, this is something that I can use as a non-technical marketer.

Dan And so I leaned in heavy, and I was like, can I build a team differently than how I'd run the agency? I developed some unique workflows and was very much an early adopter. But I think the moment where the light bulb came on for me was about two years ago. I was running marketing for a couple software companies doing 5 million to 20 million—kind of that range was the sweet spot where I ended up—and I started talking to [00:02:00] friends about what they were doing in their businesses with AI and kept seeing, oh, I'm doing something in a much different way than other people.

Dan So I started just very informally coaching and helping people. All along the way I was kind of frustrated with my practice as a fractional CMO just because of scalability. It's like you either have to become an agency or you're capped at what you can do. I was looking for something more scalable.

Dan That's when the light bulb moment hit—maybe this is something I could do where I could do one-to-many much more easily and have a bigger impact on businesses more immediately than just having a CMO relationship with them.

Dan So that's when that pivot started, about two years ago now.

Hector Yeah, it's amazing how quickly things have progressed. We've been using Notion for years now and have run all of our podcast production out of there, and we've published thousands of episodes, and just now they've integrated their AI into it. I feel like it's a totally new thing, right?

Hector It feels like I'm going from [00:03:00] driving a stick shift to having an automatic, and I'm like, what am I supposed to do with my hands?

Dan It's super cool to see tools like Notion that already had legs—was a household brand—but were just waiting for AI. It feels like it was the right container, and now you put AI into it and it's a rocket ship. It's just fun, as a technologist myself, to see tools like this become AI-enhanced and the capabilities that it unlocks.

Hector When it comes to how a business owner might start approaching it, do you think the best way to go in is to talk about tactics and some cool things that people might be able to implement right away? Or do you think that thinking about the strategy, the mindset, the philosophy, and how you're using it and what you're using it for is a better place to start for business owners?

Dan Well, both are really important.

Dan I think a two-prong approach is really the best way to do it. But I think starting with the thinking and how we think about these tools [00:04:00] informs the practice and what we do with them. You have to have both because they're such generalist tools. You can get them to do so many different things, but you don't know what they can do until you've seen it done before, or you have enough experience to connect the dots for yourself. You either need to see someone else's dots connected or have enough experience to do that work.

Dan Both are important. I could dive in—I have some frameworks that could be helpful there if that's useful.

Hector Yeah, I'd love to know: if you're thinking about the mindset, how should someone be thinking about using it? Before we get into some of the tactics, what are some of your thoughts on that?

Dan I love it. The first thing I want to start with is a very high-level view of how these tools function.

Dan They are extraordinarily sophisticated prediction algorithms. They're much like when you're texting on your phone and it suggests a word to come next—because you said “roses are,” it's going to [00:05:00] auto-complete “red.” That's what these are doing, but in a much more advanced way.

Dan At a high level, we need to understand that all they are doing is looking at how language works, translating all that into zeros and ones, and then translating back what words, sentences, and ideas should come next.

Dan With that in mind, all that training is based on the internet, books—anything that can be fed to it—to the point where these AIs have sucked up the entire world. Now they're making more content with AI just to train more AI. When you think about how these tools work and what you want to do with them, you need to know it's either going to default to its training—its prediction of what comes next—or you need to swing that pendulum toward the output that you want. That's why we see all these AI “tells,” and people talk about AI slop—things that sound like AI because we've seen certain combinations of phrases enough times that it feels too familiar.

Dan The challenge for anyone trying to use AI at a high level is to get the AI to swing in your direction by giving it enough context. Context is everything when it comes to AI. It's what AI knows about you, what you're trying to do, who your audience is, the task at hand, and all of those things. There's a framework that I like to use called the POWER framework.

Dan It's an acronym that helps me keep it organized. It's a checklist you can go through anytime you're approaching a problem or a task with AI to make sure it has all the context it needs. Before I get there, I want to say that a year from now we'll probably be having a different conversation about AI and context, because this is the thing every AI company is trying to solve for. That's why ChatGPT has memory. Claude not too long ago introduced their memory function. They're trying to learn about the user so they can give different information to the AI when you're using it that's more true to who you are. But along with that, there are times when we need to give it more.

Dan We have a more specific outcome we're looking for. We're trying to speak or create something in a different way than we've created in the past. Or if we work with clients, we're trying to create something for someone else. How do we give AI everything it needs to create that kind of content?

Dan So the POWER framework is: P is the person or persona—who is it, what are they about, what's their mission, vision, values? O is the others—you can think of this like an ICP. Who's the audience? In marketing, think ICP. In internal and operational things, who's receiving it—team, client, etc. W is the work—what is the actual work product output? How long is it? What's the format? What would a template look like? E is expression—the style, the language, the reading level, the syntax, all of those kinds of things. And then finally R is references—examples of other successful versions of a similar output.

Dan As you're approaching any task—marketing is easy to talk about because it's the common denominator across businesses, like a LinkedIn post—go through and think about whether you've given AI everything it needs to know about the person the writing is from, the others on the receiving end, the work itself, the expression and style, and references of what success looks like.

Dan I think that's a really important place to start. Then we can get into specifics of what use cases people are using and what I'm helping clients build in my coaching community. I'm coaching business leaders and founders and applying AI in their specific context, so lots of references from that side as well.

Hector People go in there and say, “Write me a LinkedIn post,” and then they wonder why it doesn't turn out well. I think the next progression is moving from it really being only this LLM thing—this place where you can go and get text done—to more of these things where it's doing stuff for you. The easy bridge is the videos and some of these things—the pictures—even how it's really [00:10:00] able to create images and videos easily. The next layer is that when you can start to train these things, they can not only give you information, but they can go out and do things for you, and you can 100 percent integrate them into different workflows.

Hector It seems like that's something you're excited about too. Why don't you talk about that as well?

Dan People get so excited about two things: agentic workflows and automations, and those go hand in hand. Let me start by defining those.

Dan An agentic workflow—or an AI agent—is like an LLM with extra abilities to go and input things, activate things, open things, close things. Examples of this: in ChatGPT they have a tool right now called Operator. Who knows what it'll be called when this goes live because they're always growing.

Dan You could tell it to, you know, go and make me a dentist appointment. If your dentist has a web interface where it can book an appointment, it may be able to do that. But right now we're at such a rudimentary level that [00:11:00] these things aren't all that great. I kind of doubt that Operator could actually successfully do that. I've played with Operator, and every time I play with it I'm like, “Yeah, that didn't really work,” and I could have gotten it done so much faster.

Hector The way that I've likened it is: this thing is so smart and so dumb. It is so specific in what you tell it. If you ask it to do anything else, it's like you've broken the whole thing. Just don't move off of this thing. If you need one agent to do everything, it's not really good. The real magic is where you can get one agent to hand it off to another thing, to hand it off to another thing. You have these different [00:12:00] agents talking to each other—almost like a team—to create a better product.

Dan I think that's where it's gonna go behind the scenes. We're getting a hint of that even with ChatGPT-5, which has their model chooser where it's choosing the model for you, so you don't have to choose. Behind the scenes there will be agents that know how to do specific things like book an appointment, collect data, fill out a form, and of course much more sophisticated than that. Right now it can feel like a fragile system.

Dan With that said, there are agentic things you can get AI to do. An example I run in my business: I subscribe to a handful of emails from media outlets that are looking for people to comment on things—media opportunities where I could guest on a podcast or comment on an article to get backlinks or opportunities to be in front of an audience.

Dan I get hundreds of these opportunities [00:13:00] every day in my inbox. I don't have time to go through all of them. So I built an agentic workflow. I run it through Make—you could run it through Zapier or n8n, any automation tools—or a script if you know scripting. It watches my inbox, and when any of these emails arrive, they receive a specific tag. Anything with that tag triggers the process. It reads those opportunities, breaks that email—which might have 25 opportunities—into individual items. Each chunk is fed through a prompt that evaluates it based on who I am, my expertise, and what kind of opportunities I'm looking for.

Dan It gives a pass/fail grade based on metrics I put in: big enough audience, topic close enough to my expertise, etc. If any pass my threshold, I get an email about that. Instead of me manually parsing hundreds of opportunities every day, I have three or four that I look at. If I decide it's a good opportunity for me, I respond. That's a somewhat agentic workflow—it's doing something for me and making decisions for me. It's not just “go out and do this thing,” but it's an example of giving these tools specific roles and tasks that run for us in the background.

Hector I want more of your favorite use cases because unlocking people's mindset around this is valuable. “I didn't know you could do that,” so I want to hear more of that. But before that, I think it's helpful to talk about how AI can also be the ultimate gaslighter, ever. South Park has a really good scene about this. Talk about some of the dangers of AI or things to think about in that regard.

Dan Totally. Similar to how I said these are predictive algorithms, they're also instructed to make the user happy. Part of it is these are businesses running these things. They want you to feel good when you use their tool so you'll stay on the subscription. We have a relationship with these tools. Even if you don't have a name for your AI—it's just a tool—some people have names and refer to their AI, and we want to be cautious about that.

Dan They are tools akin to media—to social media platforms. When I open a social media platform and see the feed, there's psychology going on. It's built by product managers trained in methodologies to keep you hooked and engaged. That's part of what's going on behind the scenes.

Dan We have to be careful because AI is going to want to make you happy. If you ask it—this happened to me not too long ago—I was in a contract negotiation disagreement with a client. They were saying it was one way; I was saying it was another. I was presenting evidence that we had done all the things we said we were gonna do, and they were skirting the issue. I brought my conversation to AI and asked, how was my tone in this conversation? I felt frustrated and wanted a pressure check: am I just mad? I wanted to be civil and kind and not come across as a hothead.

Dan It read through and was like, “Oh yeah, you're doing great, and they're so off base,” etc. I said, okay, give me some critical feedback—and then it totally flipped completely the other direction: “You're being so by the book and getting muddled into details,” etc. I was like, well, which one is it?

Dan You can get around that with quality prompting about the kind of feedback you're looking for, but it has a tendency: if I asked for feedback, it wants to tell me it's good. If I asked it to be critical, then it gets really critical. How do we prompt well to end up in the middle? First, something I call meta-prompting: before you start a task, have a chat with AI about that task in order to create a better prompt for what you're going to do.

Dan What I should have done is start another chat and say, “Hey, I'm doing this negotiation. I want to make sure I'm not overly biased. Help me write a prompt that gives me an unbiased assessment of how I'm doing.” Get into the details, have it print a prompt, then take that prompt to another chat and do the analysis. That would've helped avoid the gaslighting of “am I all good or all bad,” because it seemed like I was either one or the other.

Hector It will tell you exactly what you want. I'm thinking people are just getting around to figuring out how to search on Google. Most people didn't. They had 15 years—maybe 20. Now we're asking people to get even better. Someone told me recently that search is really great at being broad, and AI is really great at being narrow. If you go broad with AI, you're not going to get what you want, whereas if you go broad with Google, that's the way to search—to go broad. If you get too specific, then Google gets funky. It's [00:19:00] the opposite with AI.

Dan Yes, totally. That's why they call it GEO—which I kind of hate—Generative Engine Optimization. It's AI optimizing your content and web presence to show up in AI search results. I hate the name because it's still just SEO—just a different kind of search. But the conversion rates are like 40 percent. If your product shows up in a search on ChatGPT, it's like 40 percent conversions, which is nuts compared to Google—single digits if even.

Dan That's exactly what you're saying. Because it's so specific—serving up exactly what someone is looking for. That's also something I teach in my cohort. Again, meta-prompting—using AI to help you use AI. Sometimes we don't know exactly what we're looking for, so we end up with broader searching even through AI. Start with another chat where you're talking about what you're trying to do before you take the output of that chat [00:20:00] to another chat and actually do the task. That can help AI help you get ultra-specific so you can get the specific answers you need.

Hector One of my favorite things is like, write something that one of your things is gonna know what to do—be able to communicate with one of your kind. What other favorite use cases do you have when it comes to AI, whether it's small-level stuff, quick and easy things, or even high-level workflows?

Dan Totally. The quick and easy stuff—again, marketing is the common denominator. First, train your AI to speak in your brand voice. That's one of the first things I do with any client I engage with. Whether we use it for marketing or internal operations or client-related stuff, we first need to figure out how to get the AI to communicate in your way. I have a guide I can share—AITrainingGuide.com—that walks you through this process.

Dan Start with that to make sure your AI can communicate in the way you want and in the format you want. Your newsletters are one format, your LinkedIn posts another, your podcast notes a different format. Create specific workflows around those things to save tons of time so that every time you approach a task—whether it's a custom GPT, a project, or, in some tools, an agent—it's ready to do that one thing.

Dan On the marketing side, any content repurposing: workflows I help with are around capturing ideas. When you go to write—like you mentioned, “write a LinkedIn post”—it's going to sound like every other LinkedIn post. But your AI knows who you are, who your audience is, what your offers are, what you're trying to do on the platform. Then you say, “Now look at this podcast interview I just did and pull out three or four ideas I talked about that could be sparks for a LinkedIn post.” It'll pull those out, and then you can develop them into posts. Developing workflows like that keeps you from staring at a blinking cursor. If your AI can write like you, you still need to decide what it should say.

Dan How can we create systems—and even automate them? For example, put your Zoom or note-taker recordings into a specific folder. That folder logs three ideas from every call into a spreadsheet. Now, when you work on content, you've got a spreadsheet ready to go. Pull one of these, drop it into your chat, and start developing your content—or use your human mind and let those entries spark ideas.

Dan On operations, it gets more complex. One client does assessments for independent hospitals. It's a report that's like 100 pages and would take her a couple weeks of intensive work gathering information about needs in the community, etc. It's a big part of her business. I guided her through a process to create an AI workflow. Now she can go from the start of that contract to a deliverable in a couple afternoons of work.

Dan It's bespoke to her and that specific report, but even that goes back to the POWER framework. She has a report as a reference of what the product needs to look like. She has examples that capture the language and expression of how these reports should read. Then she asks her AI to sort through the data and start populating the report.

Dan I've worked with a lot of consultants on client deliverables. One consultant team builds businesses: business strategy, marketing strategy, ICP documents from voice-of-customer interviews, etc. We built their backend so their consultants can move much faster using the same process. We're building a training document and an AI that's trained on what they’re trying to produce so they have a custom GPT (or a Gemini project, or a Claude project) for that specific output. They input client stuff, and it outputs a deliverable they can then fine-tune using their perspective and IP. The same is similar for responding to RFPs or creating proposals.

Dan I personally take sales calls, then run a transcript with a bunch of voice notes—I'll record a brain dump of word spaghetti like, “I think we should do this, this, and this,”—then run it through a prompt to create a proposal draft. Drafting bespoke proposals for custom AI builds, contracts, retainers, or advisory roles can be turned around in 20 minutes instead of a whole afternoon.

Hector I'm glad you talked about some big and some small things because people need a gateway into it.

Dan Yes. I think the next step forward is—Notion AI might make this even better and more accessible—take all those SOPs and put them into AI so your team can be walked through the steps. I do all my books for my business myself. There are things like, how do I do accrual accounting for this payment plan? I have an SOP, but following it step by step makes me want to gouge my eyes out.

Dan I have a custom GPT that's trained on all of my SOPs specific to accounting and where I categorize things in case I can't remember what categories to use. I turn on voice mode and just talk to it: “I've got to set up my accrual system for this payment plan that just kicked off. How do I do that?” It talks me through it, which takes so much friction out.

Dan That's what I love about AI: it's such a tool for knowledge. Someone with specific domain authority equipped with good AI workflows is unstoppable. Helping those people take friction out of their work so they can do more of their zone of genius gets me fired up. Spending less time figuring out accounting so I can go back to helping more people brings me joy, and that's what I love to unlock in folks.

Hector Two more questions for you, Dan. First, if we wanted to get connected with you or follow up, where's the best place for people to go?

Dan I have that [00:27:00] guide, AITrainingGuide.com, that teaches you how to train your AI to do the things you want to do. You can start there. That'll put you into my newsletter—always just hit reply and find me. Or you can find me on LinkedIn. Just search my name, Dan Cumberland. Those are the best places to find me.

Hector And last question for you: in your opinion, what is the secret to scaling a business?

Dan The secret to scaling a business is managing your risks—managing the bets that you're taking. You're taking big enough swings—I'm mixing metaphors here—that you can hit a home run, but not strike out. That's maybe the next lane over from resilience, but it's staying in the game through the ups and downs in order to reach the growth trajectory you're chasing.