Milena Traikovich is a seasoned leader in demand generation who specializes in bridging the gap between high-level marketing strategy and technical execution. With a deep background in analytics and performance optimization, she has spent years helping businesses transform fragmented data into cohesive, high-quality lead-nurturing engines. Today, she shares her insights on why marketing needs a “decision infrastructure” to truly unlock the potential of artificial intelligence.
In this discussion, Traikovich explores the structural differences between engineering and marketing, the necessity of capturing “institutional memory” through context graphs, and the shift from simple A/B testing to a more complex “network of why.” She also outlines how decision-making can be treated as structured data to create a more durable and intelligent marketing ecosystem.
Engineering teams rely on standardized syntax and modularity, while marketing terms like “campaign” vary wildly across industries. How does this lack of structure limit AI effectiveness, and what practical steps should leaders take to create a shared, modular language that both humans and machines can interpret?
The success of AI in software engineering isn’t an accident; it’s because programming languages are built on explicit dependencies, version control, and modularity. In marketing, we often operate on “partially documented logic,” where a single term like “campaign” can have over a dozen different meanings depending on who you ask. This lack of structure forces AI to guess the intent behind our actions, leading to inconsistent outputs. To fix this, leaders must stop treating marketing as just “art” and start formalizing our definitions and processes. We need to decompose our tasks just like engineers do—breaking a brand launch down into modular components with defined interfaces. By standardizing our vocabulary and documenting the “why” behind a pivot or an audience exclusion in a way that is machine-readable, we give AI the infrastructure of meaning it needs to perform.
Critical logic often hides in Slack threads or verbal reviews rather than structured systems. How can organizations implement context graphs to capture decision traces—such as policy exceptions and historical precedents—and what specific metrics demonstrate that this institutional knowledge is actually compounding over time?
We currently lose an incredible amount of “taste” and institutional knowledge to five-minute Slack exchanges and verbal huddles. Context graphs solve this by acting as a new system of record that sits alongside our transactional tools to preserve organizational reasoning. Practically, this involves recording exactly what inputs were considered during a decision, which policies applied, and whether a specific exception was granted by a stakeholder. You know this knowledge is compounding when you see a measurable shortening of feedback loops and a “raised floor” of quality across the entire team. Instead of junior staff making the same historical mistakes, the system flags precedents, ensuring that the logic used six months ago is immediately available to inform today’s execution.
Human guardrails and manual overrides often reappear when AI fails to grasp brand nuance or internal risk tolerance. How can these subjective instincts be translated into machine-readable constraints, and what is the workflow for ensuring an AI agent refers to this “living layer of reasoning” before it executes a task?
When AI produces something that “feels wrong” despite being data-accurate, it’s usually because it lacks the memory of how we handle trade-offs. To bridge this, we have to translate our brand nuance and regulatory interpretations into structured context that an agent can query. The workflow involves moving policies from static PDF documents into active inputs within the AI’s operational loop. Before an agent generates a content piece or selects an audience, it must check the context graph to see the “living layer of reasoning”—the specific claims we’ve softened in the past or the risks we aren’t willing to take. This ensures that the AI isn’t just reacting to the loudest statistical signal, but is navigating the same subtle guardrails that a human expert would use.
Marketing outcomes are influenced by a complex network of variables, from customer history to cultural shifts. How do you move beyond simple A/B testing to build a “network of why,” and how can teams isolate the specific hypotheses or creative wagers that truly drove a performance lift?
Simple A/B testing is often too reductive to explain why a campaign actually worked in a world of millions of dynamic inputs like device state and competitive pressure. Building a “network of why” requires us to codify our specificity at the component level—stating exactly what language we expect to resonate and the strategic wager behind a specific creative choice. When a lift occurs, we shouldn’t just credit the “campaign”; we should record which hypothesis was under pressure and why one signal carried more weight than another. By documenting these conflicting signals and the resulting leadership overrides, we create an interconnected graph of assumptions. This allows us to move beyond surface-level optimization and understand if a win was due to a narrative arc, perceived credibility, or a specific cultural alignment.
Decision infrastructure serves as a connective layer for existing tools like CRMs and CDPs. How does this shift the way marketing operations handle approvals and static policies, and what are the long-term strategic benefits of treating every high-level choice as structured data?
This shift transforms marketing operations from a series of manual checkpoints into a more coherent, automated flow. Instead of approvals living in emails, they become structured data points that link the “what” in your CRM to the “why” in your decision infrastructure. The long-term benefit is continuity; when a key leader leaves, their judgment and the precedents they set don’t disappear—they are baked into the system. It allows the martech stack to store the conditions and logic that led to a state, not just the state itself. Treating choices as data means that governance moves from being a reactive hurdle to a proactive, referenced input, making the entire organization more durable and better prepared for AI-driven scale.
What is your forecast for marketing decision infrastructure?
I believe we are moving toward a future where marketing “insight” itself becomes a scalable asset. We will stop seeing AI as a replacement for human creativity and start seeing it as a way to ensure our best ideas and hard-won lessons compound over time. As we move away from static documents and toward dynamic context graphs, the organizations that win will be the ones that have successfully encoded their unique “taste” into their infrastructure. Marketing will become the primary proving ground for scaling human judgment, allowing teams to move at unprecedented speeds because they are no longer guessing—they are operating with a collective, machine-readable memory.
