How to Use AI for Market Research and Product Development

Full video transcript

Hey, how’s it going y’all? Darby here with Gen AI University, and in this video I’m going to be sharing an interview that I did recently with a member of our community by the name of Marty Bradley. He is an AI consultant, and he helps businesses integrate AI systems into their workflows. Marty has been a member of our community for several years now — he was actually one of the first members who went through our initial book challenges and more extensive hands-on workshops, working with myself and our team to help him get the book out of his head and onto paper. That was three years ago, and since then he’s really leaned into the AI consulting front and is doing a lot of cool things. He reached out to me to see if I’d be open to sharing with him and his audience a bit more about how we’re using AI, natural language, and prompts to create products — how we’re doing this for Market Sauce — plus some other insights into the new tools we’ve been leaning into at Gen AI University that are helping us take our ideas and turn them into action and impact.

Marty: Hey Darby, nice to see you again. Hey everybody, thanks for joining us — quick intro, my name is Marty Bradley, I’m a gen trainer and more of an engineer than a strategist lately. So we’re going to be talking with Darby — Darby, why don’t you do a quick intro, then I’ll let everybody know how we actually met, which was like three or four years ago.

Darby: Sure, yeah, thanks for having me here Marty. My name is Darby Rollins, I’m the founder of Gen AI University, a platform we started about three and a half years ago to help people with AI education and implementation training. Back then we got started with Jasper — we’ll cover some of those early days on today’s call. We’ve grown a community over the past few years and are out exploring ways to build and create products and different services, collaborating with others on all these new AI innovations. It’s an exciting time to be in entrepreneurship and business, but also a lot of change and uncertainty to navigate, so it’s good to have a community of folks like yourself along for the ride.

Marty: It’s pretty interesting how we met — it was through a cohort of people writing a business book, I think the title was “Write a Business Book in 10 Days,” and it took me a little longer than everybody else. But it was interesting because you were doing something I hadn’t seen before, writing recipes in the back end of Jasper — you called it autoscripting at the time. As we went through it, I was doing consulting and started thinking I could use this to help people generate other types of things, not just books — so I used it in my agile consulting practice helping people write epics and stories. Then when OpenAI and ChatGPT came out I went full bore, but it was all thanks to you and writing a book. I did get my book written and published on Amazon, but what really stuck with me was the prompt engineering side. So tell us what you’ve been up to — Gen AI University, and I also want to hear about Market Sauce, and how you built it basically on your own without a software development background.

Darby: Yeah, I don’t think I’d ever classify myself as a software engineer, but when AI came out I had a lot of fun figuring out how to use the natural-language side to speak prompts and direct these tools toward an outcome, because otherwise you just get lackluster outputs. That was a lot of the problem people ran into early on with Jasper — a book was a great example of an application, but that piece of content could be anything: podcast scripts, blog posts, LinkedIn articles, all the things that are time-consuming and resource-heavy to produce traditionally. It’s shaken up how people build and structure their teams. One of the things that’s been key to getting Market Sauce to where it is today — still very much an MVP — has been working with our core development partner Chris, who I’d call our chief AI officer. He’s been instrumental in tuning the automations to run at a scale that would be humanly impossible, even if you could get similar outputs manually inside ChatGPT or Claude. There’s a lot of process involved in running the prompts in the right sequence to get the desired outcome, because AI will go off the rails sometimes, and that’s where you need a human guiding the process.

We built this as a product based on frameworks we’ve learned and collaborated on with other experts over the last three years, mainly by structuring and refining prompts — the only “code” in this case is talking to ChatGPT and using language as the code. We got to the point where we could produce 50-page, super-detailed market analysis reports just through that process, but even running five ChatGPT accounts at once you hit limits, so we needed automation to scale it. Over the past year we’ve been digging into a platform called Make very extensively for the backend processing of these automations, creating AI assistants on the fly using OpenAI’s Assistants platform that quality-control the process and keep the automation moving if something breaks.

There are also simpler tools removing a lot of that friction if you just want something lightweight — one I’m exploring right now is called Pickaxe, which lets you create and productize AI tools that plug into OpenAI, Anthropic, and Mistral, in a simple interface similar to building your own GPTs. You upload your own knowledge and instructions, tell it how to act and respond, give it a series of prompts. It’s not as overwhelming to get started with as building extensive Make automations, and it’s especially useful for a consultant with a proven process — after onboarding clients with a few questions, it can generate a strategy document or analysis. Whether you go no-code with something like Make or a lighter tool like Pickaxe, there are a lot of ways to find quick wins — even reducing a 15-minute task to 15 seconds frees up time to reallocate.

Marty: What I’m seeing is the last couple of years AI was almost like a toy in large organizations — fun to play with but nobody put it seriously into their customer workflow. I think as we move into 2025 that’s starting to change, people need real apps that can handle the load, and once you have to budget for it, you have to prove ROI. I want to hear more about what’s under the hood, without giving away your whole architecture, because I think these are the things people are struggling with. Another benefit I’ve seen is that as subject matter experts learn this technology, even if they don’t build a production system themselves, they can better describe what they need built and help lead their development staff toward a more robust solution.

Darby: As the landscape changes, being able to speak to where a tool adds value matters more, because there’s a lot of obscurity and tools that aren’t backed by the right infrastructure to handle larger-scale clients. We’re a small team, and we realize there’s a lot to handle around data and privacy when building any app — that was true before AI, but now you need to be extra careful, especially with private or sensitive information. When ChatGPT first came out there were cases of people uploading a document and worrying whether other companies could learn what was in it. Enterprise accounts and agencies generally address this now — not training on your data, letting you delete it — but anyone building this inside their own organization should be aware of where that conversation currently stands with whatever platform they use.

Understanding where time is currently being spent is important for identifying where AI should be focused. We don’t work with a ton of large enterprises — most of our clients are consultants, smaller agencies, or business owners using these tools with one or two team members — so we’re looking for inefficiencies in a smaller-scale process: what’s manual, what’s costing time, and which activities add the most value relative to the input required. Traditionally, putting together a marketing brief or structured report required reading reviews, running focus groups, and a lot of time. With Market Sauce we’ve built a system that compresses that time — give it information about your business and service, and it produces the output.

Let me show a bit of what this looks like. This is still very much early — I’m building it out right now for a platform that will be available for marketers and consultants, so you’re one of the first people seeing this, Marty. This is built on Pickaxe — we’re also building our own platform and back end separately, and this is one way we’re starting to build out these workflows and plug them in, including for clients who want a white-labeled tool. In this interface I have two GPTs I’ve created: the first is “The Mentalist 2.0,” named after our process, for getting a quick snapshot of a market based on some content or information. I’ve built an elaborate series of instructions and prompts behind it for a specific scenario — understanding buyers in a market and how a product, service, or company can meet them where they’re at.

For this demo I used Claude this morning to come up with a random business name — “Myo Mend,” a mushroom-based biodegradable packaging company that creates custom packaging solutions and offers consulting services to companies transitioning to sustainable packaging. As the user, I add in the business information — you can add files like customer interviews or additional context — and clicking submit generates the full report. Some of our reports end up being 50-60 pages using Make, which can be overwhelming, so we’re working on more simplified, modular frameworks.

Walking through the output: it starts with an ideal customer profile — eco-conscious brands and e-commerce companies committed to sustainability — a SWOT-style analysis of strengths, weaknesses, and threats, then goes deeper into who the customers are, their core values, emotional drivers, desired benefits, and specific market segments and sub-segments. For a marketing consultant working with brands transitioning from non-eco-friendly plastics to sustainable packaging, this gives the language and messaging to speak to that audience, and helps validate whether the messaging is hitting the right pain points and goals. As feedback comes in from the market, you refine and evolve the messaging over time — this is one example of what a comprehensive first pass looks like, and you can go deeper on any section.

All of this came from one input pasted from a Claude generation. The more information you bring from your own business and customer insights, the more it informs the messaging — and if you’re working with multiple clients, generating multiple reports is straightforward. The second tool, the buyer’s brief conversational thread, takes the Mentalist output and opens a conversational dialogue with the AI to continue refining, simulating your own focus group for messaging and ad content. We’ll be building additional frameworks inside Market Sauce for more content, messaging, and insights, all built on the back end in Pickaxe — I’m not sharing an affiliate link, just showing that this is a powerful, cost-effective tool.

What matters is what’s going on in the background. For the buyer’s brief chat simulation, the AI needs to understand its role first — in this case, it’s a clone of me as the author, trained on my book, helping people execute the book’s content based on their buyer personas. That book (available at buyersbriefbook.com or Amazon) literally contains the recipe for what’s going on behind this GPT — the first half is examples of inputs and outputs, the second half is the actual instructions for the prompts. Every prompt has been detailed and crafted to build on the next, and what you saw is about half of what a full report would be. Even the smaller five-page snapshot reports operate on the same core questions — the better you understand your audience, goals, and industries upfront, the more you can focus and organize that information.

Most businesses know who their market is, but there are often other segments right under their nose that their messaging or service is completely missing. Having a different perspective — and then getting real feedback from humans — is important; I’d view the AI as a co-creation partner where you might refine a blueprint 50 times before it lands on the right word. This applies beyond marketing too — recruiters use similar tools for building recruiting profiles of who they want to hire. This process would take a professional copywriter hours if not days to weeks, depending on depth. Even if you don’t use our tool, doing this with AI in some form gives you a competitive advantage today.

Marty: A couple things stood out — I hadn’t thought about using this to figure out a target market of potential employees. And for small and medium business owners who know they need to market but assume their existing customer group is the only audience, a tool like this gets them a first draft and forces them to answer the underlying questions, which is itself a learning process. The other point — in the old days of software development, taking that knowledge from your book and making it queryable would have meant building a relational database. Now with AI, businesses with a lot of unstructured data or existing instructional content can just use that as the basis, without needing to structure it that way first. I do something similar with activity guides in my own book that I pull in to help answer client questions.

Darby: The organization of knowledge that already exists is probably the most important thing — in any consulting arrangement, it’s often about figuring out what resources you already have scattered around and getting them into one source of truth. But it’s not about throwing everything into one chatbot — quantity isn’t the goal, it’s quality, even though it takes time to put that data together. The payoff is in how much time you save once you can execute the process with everything organized. Without a clear sense of the expected ROI or time savings, it’s hard to justify the effort going in.

Marty: That’s no different than the old AI/ML process — you always start with the data to figure out what you have. The difference now is that data usually doesn’t need heavy reformatting. Talking through ROI and getting everyone aligned on the value is important, especially for getting approval from CFOs and executive teams at larger organizations.

Darby: What I’m seeing from organizations putting AI initiatives together is that it works best when one person or a small team is dedicated to being the innovators, leading the charge and spending time developing use cases — it’s not something everyone on a team can spend all their time doing, or they end up down rabbit holes. Having at least one dedicated person seeing the opportunities and presenting ideas leads to results. On the other hand, competitors who are open to AI initiatives are more than likely moving on their own front, and the speed of implementation can create real gaps. I don’t recommend adopting AI just to reduce headcount, but roles are shifting — a content writer, for example, can shift into being the person who oversees a team of AI content writers and edits their output, which is probably the fastest way to see benefits from these tools.

This isn’t a tool problem, it’s a people problem. For example, building an online course deck used to take me a couple of days thinking through audience, organization, and graphics. Working through it with ChatGPT, I can generate a course section draft in about 15 minutes, then spend my remaining time on edits and on the parts that matter — whether a human will understand what’s being said, what the script behind the slides should be. That’s where you want to spend your time: thinking about the learner experience, not assembling slides for two weeks. If everyone found one deliverable in their job and figured out how to automate part of it — with Claude, Copilot, or otherwise — they’d teach themselves the technology, have to dig into the deliverable deeply enough to explain it to the AI, and create documentation that makes it easier to hand off work to someone else.

Marty: We’ve all been promoted into a new role while still carrying the old one because there’s never enough documentation to hand it off. If people think about it as making their job easier, it’s not that AI is going to replace them — not yet, anyway. You still need human expertise for final edits and quality judgment, and for the emotional read on how something lands. If people are afraid of losing their job, they should be more afraid of someone else who understands AI better coming in above them. The upskilling gap between people using these tools and people who aren’t is going to matter — if I’m hiring and everything else is equal, I’m choosing the person who knows how to use the tools.

Darby: Did you see the study — I think it was Microsoft — that found around 70% of hiring managers would prefer to hire someone with AI skills but less experience over someone with more experience but no AI skills? It’s already affecting hiring, and people are also using AI to screen applications and to stand out in the application process, which some members of our community have started doing.

Marty: Absolutely. Is there anything else you want to talk about?

Darby: Just how fast everything moves in this space — it’s hard to say where things will be in six months. But at the end of the day, for anyone looking at using these tools to build products or get more clients, it comes down to doing a solid assessment of your current workflow, identifying the core task taking up your time, getting clear on the steps behind that outcome, and then figuring out where AI can help — you can literally have a conversation with ChatGPT describing your role and asking for ways to leverage AI inside that workflow. If documentation is part of your process, there are AI-driven tools that will document your process from monitoring your screen and generate SOPs from that — I’ve been putting that off forever, but a tool called Guide comes to mind, though there are others. From a role standpoint, I also have GPTs that take a role and break it down into its core parts and key deliverables, then break those down into tasks to understand what can be automated.

Marty: That’s absolutely true, that’s what people should be looking at. If people want to find you, what’s the easiest way?

Darby: You can find me through our Gen AI University newsletter, sent out weekly or so, and across social media where we post AI education content, events, and insights. I’m also on LinkedIn and other socials at Darby AR Rollins.

Marty: And if you want to reach me, LinkedIn is the best way — just Marty Bradley. Hey Darby, I really appreciate talking with you, I love how you’ve been looking at this over the years, thanks for jumping on.

Darby: You bet Marty, great to be here, thanks for having me on.

There you have it, folks — wrapped up the interview with Marty Bradley. If you found some insights or takeaways you can put into action, let us know in the comments what stood out to you or what interviews and topics you’d like to see more of, and make sure you’re subscribed to the channel with notifications on so you know when we drop new interviews, trainings, and demos. Head on over to Gen AI University to subscribe to our newsletter for more resources on leveraging AI to get inside the mind of your market.