Milena Traikovich has spent her career at the leading edge of demand generation, helping businesses navigate the complex intersection of data analytics and performance optimization. With a background rooted in high-stakes lead generation, she possesses a unique vantage point on how emerging technologies shift from fringe experiments to core business drivers. Today, she views the rapid ascent of artificial intelligence not merely as a technical upgrade, but as a societal inflection point reminiscent of the early days of email marketing. In this discussion, we explore the friction between corporate caution and individual innovation, the nuances of competing large language models, and the specific strategies marketers can use to future-proof their careers. Our conversation covers the shift from top-down to bottom-up innovation, the necessity of personal “sandboxes” for testing AI capabilities, and the practical technical competencies required to stay relevant in an increasingly automated landscape.
Corporate adoption of AI often stalls in committee due to security and privacy concerns, yet innovation is currently bubbling up from individuals rather than large enterprises. How can a marketer effectively bridge this gap, and what specific steps should they take to build a personal “sandbox” for experimentation?
The current landscape of AI adoption is fascinating because it has completely flipped the traditional script of technological evolution. In the early days of email marketing, innovation was a top-down affair where massive enterprise-level companies had the capital and infrastructure to pioneer new tools, which then trickled down to the rest of us. Now, we are seeing a “bubble-up” effect where individuals and mid-market players are the ones pushing the boundaries while the super-enterprise giants are paralyzed by red tape, information security protocols, and privacy fears. To bridge this gap, you have to stop waiting for a corporate mandate or a formal training program that might not arrive for another eighteen months. You should treat your professional development like an independent research and development lab by identifying a “niggling” problem in your personal life—something low-risk but data-intensive—and applying AI tools to solve it. My personal sandbox was my wine cellar, a collection of 300 bottles built over 20 years that desperately needed an inventory system. By tackling a personal project like this, you can safely explore the limits of various platforms without risking sensitive company data or violating any internal security policies. It allows you to build a level of fluency and intuition that will make you the go-to expert the moment your company finally opens the floodgates for AI integration.
Different large language models often yield wildly different results when processing visual data or pattern recognition. When a tool produces “hallucinations” or incorrect data during a task, what is your process for diagnosing whether the failure stems from the model’s underlying infrastructure or the specific context provided?
Diagnosing an AI failure requires a mix of skepticism and methodical testing, much like troubleshooting a leaky funnel in a demand gen campaign. When I first tried to inventory my wine collection using ChatGPT, I spent about thirty minutes meticulously photographing labels and uploading them, expecting the model’s pattern recognition to do the heavy lifting. Instead, the system began hallucinating spectacular errors, such as claiming I owned a 1999 Screaming Eagle Cabernet Sauvignon valued at $2,985—a bottle I certainly didn’t have in my racks. To diagnose this, I had to look at whether I was providing enough context or if I was simply asking the tool to perform a task outside its core architectural strengths. In that specific instance, the failure wasn’t about the prompts; I had given ChatGPT the same context and images I eventually gave to other models, yet it still returned inaccurate data. This suggested an infrastructural limitation in how that specific version handled fine-grained visual pattern recognition for niche consumer goods. When you hit a wall like this, you have to pivot and test the same parameters across different models to see if the error persists; if one model succeeds where another fails using identical input, you’ve identified a fundamental quirk of the model’s infrastructure rather than a flaw in your own process.
Using personal projects—like inventorying a private collection—can reveal the unique quirks of tools like Claude, Gemini, or ChatGPT. How do you determine which specific AI model is best suited for a particular business objective, and what metrics do you use to verify the reliability of their outputs?
Determining the right tool is less about following a static infographic and more about hands-on experience that reveals the “personality” of each model. During my wine inventory project, I discovered that Google’s Gemini was surprisingly hit-or-miss; it struggled to process the photos and, rather than admitting uncertainty, it guessed incorrectly at the labels, which is a dangerous trait in a business setting. Claude, on the other hand, was a revelation because it didn’t just try to process the data blindly; it actually flagged which photos were too blurry or poorly framed and asked for retakes. This ability to provide meta-commentary on the quality of the input is a massive reliability metric for me, as it shows the model can recognize its own limitations. For business objectives, I look at the “trustworthiness of output” as my primary metric—specifically, how often a human has to step in to correct a foundational error versus a stylistic one. In the wine experiment, Claude successfully built a 300-bottle inventory with high accuracy after I followed its suggestions for better photos, proving it was the superior choice for high-precision categorization tasks. I recommend testing each platform on a small, measurable set of data where you already know the “right” answer, allowing you to calculate an accuracy percentage before committing to a larger organizational workflow.
Marketing leaders increasingly prioritize AI proficiency as a top requirement for future hires. Beyond basic copywriting or simple image generation, what specific technical competencies should a professional master to stay competitive, and how can they demonstrate these practical skills to a hesitant employer?
We are moving past the era where simply knowing how to write a prompt for a blog post is enough to impress a hiring manager; the Litmus State of Email 2026 report already highlights AI know-how as the number one skillset marketing chiefs are seeking. To stay competitive, you need to master the “middle ground” of technical competency, which involves understanding how to organize complex tasks, manage data inputs, and recognize when the AI’s logic is beginning to fray. You should be able to demonstrate an ability to build workflows—perhaps showing how you used a tool to create a multi-layered forecasting model or a custom dashboard that synthesizes disparate data points. I often tell people that investing in paid tiers of these services is like subscribing to multiple streaming platforms; it’s a necessary cost for staying culturally and professionally relevant. To sway a hesitant employer, don’t just talk about what AI can do; show them a finished project that solved a concrete problem, explaining the “why” behind your choice of tool and the steps you took to verify the accuracy of the results. Demonstrating that you have already navigated the trial-and-error phase on your own time shows a level of initiative and risk-management that is incredibly attractive to an organization that is still “stuck in committee.”
Many professionals struggle to find a balance between using AI as a glorified search engine and diving into deep coding. What does an ideal “middle ground” look like in a daily marketing workflow, and how can a person transition from surface-level use to solving complex organizational problems?
The ideal middle ground is a space where the marketer acts as a highly skilled orchestrator, someone who doesn’t necessarily need to host Python servers or live in GitHub, but who possesses the technical literacy to push a model to its absolute limits. Transitioning from surface-level use requires a shift in mindset from “asking questions” to “building systems.” Instead of using AI to find an answer, use it to build a framework—for example, instead of asking for a list of lead gen tactics, ask the AI to help you design a scoring model based on specific behavioral triggers and then have it help you write the logic for that model. You move into solving complex organizational problems when you start using these tools to identify patterns in your data that a human eye would miss, or to automate the tedious parts of data cleaning and categorization. It’s about immersion; you have to spend enough time with these tools to understand their nuances, just as I learned that Claude is excellent for detailed inventory work while ChatGPT might be better suited for different types of business modeling. By making mistakes on personal projects and pushing these models until they break, you develop the intuition needed to apply them to high-stakes business challenges with confidence.
What is your forecast for AI in marketing?
I believe we are rapidly approaching a moment where the “individual-led” innovation we see today will force a massive, somewhat painful realignment within enterprise organizations. We are currently at a critical inflection point where those who have spent the last year or two experimenting in their own “sandboxes” will emerge as the only ones capable of leading the next generation of marketing departments. I expect to see a significant shift away from generic AI applications toward highly specialized, multi-model workflows where different tasks are routed to specific AIs based on their architectural strengths—much like how I realized certain models were better for visual label recognition than others. Eventually, the companies that are currently “stuck in committee” will have to play a desperate game of catch-up, and they will do so by aggressively hiring the people who didn’t wait for permission to start learning. The future belongs to the marketer who is comfortable being the “human in the loop,” possessing the discernment to catch a three-thousand-dollar hallucination before it ever reaches a client’s eyes. My advice for readers is to start your own experiment today, whether it’s organizing a collection, planning a complex itinerary, or building a personal budget, because the skills you gain in the privacy of your own home will be the exact ones that secure your seat at the table tomorrow.
