Why marketing needs a decision infrastructure for AI

AI excels where structure already exists. Context graphs could help marketing capture decision logic and make AI systems more reliable.

Table of Contents

    Spy on Any Website

    Get traffic data and keyword intel on competitors instantly.

    When people ask why AI has been so successful in code generation, the answer isn’t mysterious. Programming languages are structured systems. They have syntax, grammar, modularity, version control, testing protocols and shared conventions that engineers learn early and reinforce daily. Tasks can be decomposed, interfaces are defined and dependencies are explicit.

    When an AI model is trained on code, it operates within an environment that already has deeply standardized patterns and well-understood constraints. AI performs well in these domains because the infrastructure of meaning already exists. 

    Marketing is different. It’s often described as both art and science, but in practice it operates on partially documented logic, data and that leader with a strong opinion. Taste, timing, perception, risk tolerance and lived experience all influence decisions. The rationale for why a campaign pivoted midstream, why a claim was softened or why an audience was excluded often lies in five-minute Slack exchanges, verbal reviews or the instincts of experienced leaders.

    Unlike engineering teams, marketing organizations are rarely trained in a shared, modular decision language. In my career, the term “campaign” has over a dozen meanings depending on the organization or industry. 

    The time spent deciding on the right shade of blue for a logo is unfathomable. Engineers can break down tasks, assign components and reassemble them because they operate within a formalized structure that is teachable and machine-readable. Marketing teams collaborate more fluidly. Ideas collide, evolve and shift based on nuance that isn’t always captured in overt data fields.

    This is precisely why context graphs matter in marketing.

    Context graphs as marketing’s decision infrastructure

    If AI can excel where structure already exists, the opportunity in marketing is to build the missing structure around decision-making. Not to remove creativity or judgment, but to make the logic behind those judgments durable, discoverable and agent-ready.

    • How much of the taste and institutional knowledge inside a global marketing organization is actually documented in a way another human could reuse? 
    • How much of it exists in the right system for an AI agent to reference before generating or executing? 
    • How much of it is structured enough to compound over time?

    When that reasoning remains fragmented, AI in marketing remains assistive at best and risky at worst. When that reasoning becomes structured context, AI can begin to accelerate collaboration, shorten feedback loops and raise the floor of quality across teams.

    Context graphs are emerging as a way to make decision logic durable, queryable and usable by both humans and machines.

    At a practical level, a context graph connects data about entities such as customers, campaigns, products and markets with the rules, policies, constraints, approvals and reasoning that shape decisions. It captures not just outcomes, but decision traces over time.

    This includes things like:

    • What inputs were considered at the time of a decision.
    • Which policies or guardrails applied.
    • Whether an exception was granted and by whom.
    • What precedent influenced the choice.
    • What happened as a result.
    • With two or more sources with the same data, which is the best answer.

    Context graphs can function as a new system of record, one that sits alongside transactional systems but serves a different purpose: preserving organizational reasoning. Rather than storing only the current state, they retain the conditions and the logic that led to it.

    The Foundation Capital article on context graphs frames the concept this way. The closest example of this I have personally used is Glean.

    This isn’t about turning marketing into code. It’s about giving marketing a decision infrastructure strong enough to support intelligent systems.

    10X your SEO with Semrush for Enterprise.

    The world’s most powerful SEO platform, purpose-built for Enterprise.

    Request demo

    Where AI breaks down and why marketing exposes it first

    Marketing teams introduce AI into content, targeting, offers and optimization and almost immediately. human guardrails reappear. Reviews, escalations, quiet overrides. The output may align with the data and still feel wrong because brand nuance, regulatory interpretation, historical missteps and internal risk tolerance aren’t captured in structured form. What looks like intelligence is missing the memory of how trade-offs are actually made.

    The challenge is that this knowledge rarely lives where machines can use it. It lives in briefs, in Slack threads, in five-minute debates, in the instincts of people who’ve seen what happens when a claim crosses a line. It changes as audiences change. It sharpens as teams connect ideas and adjust hypotheses at the component level.

    If AI is going to participate meaningfully in marketing, it needs access to that living layer of reasoning. Context graphs don’t replace human taste or experience. They capture the logic surrounding it so that precedent, constraint and strategic intent become durable. Without that layer, AI remains reactive. With it, marketing becomes the clearest proving ground for scaling judgment — not just automation.

    The expanding network of why: marketing’s real complexity

    Marketing decisions rarely hinge on a single variable. Every message, incentive, image or journey interacts with thousands, sometimes millions, of dynamic inputs: customer history, channel context, device state, competitive pressure, cultural shifts, regulatory nuance, timing and brand perception. Even in CRM environments that feel structured, the combinatorial complexity is significant.

    Teams manage this through experimentation. A/B testing evolved into multivariate and multi-factor frameworks because simple comparisons rarely explain performance at scale. Experimentation itself is durable and increasingly sophisticated. The constraint is complexity. Isolating a single meaningful variable can take weeks and translating multiple winning tests into a coherent explanation for what should happen next is even harder.

    Performance data may show lift but understanding whether that lift came from a specific phrase, a narrative arc, perceived credibility or alignment with a broader brand moment requires judgment. Someone has to articulate the hypothesis, isolate the component under pressure and state what they believe drove the outcome. Those hypotheses are usually summarized at a campaign level. What’s rarely codified is the specificity at the component level: the exact language expected to resonate, the tension intentionally introduced, the strategic wager behind a creative choice.

    Marketing is inherently dynamic because consumer perception is dynamic. Meaning is co-created between brand and audience and evolves as context shifts. What works this quarter may fail next quarter, not because execution declined, but because the environment changed. That living dimension makes marketing powerful and it makes structured reasoning more demanding.

    There’s additional complexity when signals conflict. Quantitative performance may point one direction while qualitative insight suggests another. Brand priorities may compete with efficiency targets. In practice, experienced leaders negotiate these tensions in real time based on precedent, risk tolerance and strategic intent.

    Capturing that hierarchy of reasoning isn’t trivial. It requires formalizing hypotheses when they’re proposed, documenting which component is being tested, recording when a decision was made despite conflicting inputs and clarifying why one signal carried more weight than another. Over time, this builds a network of why — an interconnected graph of assumptions, tests, conflicts, overrides and learning loops.

    That network becomes increasingly valuable as AI systems are asked to collaborate, generate and execute. It allows machines to navigate nuance rather than default to the loudest statistical signal. It enables teams to move faster not because creativity is automated but because reasoning compounds.

    In this light, context graphs aren’t an abstract governance layer. They are a structural response to the reality that marketing operates across evolving human perception and high-dimensional data. Without a way to encode and connect the reasoning behind decisions, AI remains limited to surface optimization. With it, marketing organizations begin to scale insight itself.

    This isn’t about replacing human judgment

    Capturing decision logic for machines often triggers a concern: if we encode the reasoning, do we reduce the need for the people behind it? The goal isn’t replacement. It’s continuity.

    Marketing judgment evolves. Audience expectations shift. New variables enter the environment. What we call best practice is simply the strongest hypothesis supported by available evidence at a given moment. As evidence changes, so should the conclusion. That adaptability is a strength.

    The scientific method works the same way. A claim holds until a stronger, repeatable explanation emerges. Context graphs follow that logic. They record the conditions, assumptions, trade-offs and outcomes tied to a decision at a point in time. As new information appears, that context can be expanded or revised. The record evolves with it.

    In marketing, where taste and experience shape results, the insight forged in debate and collaboration remains essential. The aim isn’t to transplant a human brain into a system. It’s to ensure that when those conversations produce learning, it becomes durable.

    Structured context creates a trace of how thinking changed and what followed. That trace supports deeper iteration and more informed evolution. Context graphs aren’t a concession to automation. They are infrastructure for institutional memory in a world where knowledge compounds through revision.

    What changes in the martech stack

    Context graphs add a connective layer to marketing architecture.

    They don’t replace CRMs, CDPs, DAMs or marketing automation platforms. They link actions to the reasoning behind them, so the stack stores not just what happened but why it happened.

    In practice, this means:

    • Treating decisions as structured data. 
    • Capturing context at the moment a choice is made.
    • Connecting approvals, policies and outcomes across systems. 
    • Making that reasoning accessible to both people and AI.

    Governance shifts quietly but meaningfully. Policies move from static documents to referenced inputs within workflows.

    When reasoning is durable, AI can operate with context instead of guesswork. Scale becomes more controlled because decisions are traceable. The stack doesn’t grow more complex. It becomes more coherent.


    Contributing authors are invited to create content for MarTech and are chosen for their expertise and contribution to the martech community. Our contributors work under the oversight of the editorial staff and contributions are checked for quality and relevance to our readers. MarTech is owned by Semrush. Contributor was not asked to make any direct or indirect mentions of Semrush. The opinions they express are their own.

    Angela Vega
    Director, Capabilities and Operations, Expedia Group

    Angela Vega has spent over 13 years leading marketing transformation at the intersection of technology, operations, and strategy. She’s scaled marketing systems across industries—including financial technology, consumer health, and travel—building high-performing teams, streamlining complex operations, and delivering impact for audiences ranging from 20K to over 130M.

    She has managed, led and/or built 14 marketing tech stacks, executed end-to-end operations models from product creation through campaign delivery, and steered CRM architecture for one of the largest global loyalty launches in travel. Her leadership has unlocked multimillion-dollar savings, accelerated execution, and aligned cross-functional teams across data, tech, and creative.

    Angela’s greatest superpower? Connecting dots that didn’t know they needed to be connected—bridging strategy and execution, creative and technical, global ambition and local reality.

    As a late-diagnosed ADHD professional, Angela is passionate about how AI can support neurodivergent minds, and actively prototypes ways to make work more intuitive, creative, and scalable. She even built a custom GPT to help leaders design their own operating manuals—because great leadership starts with self-awareness.

    Outside of work, Angela mentors emerging talent, advocates for women’s health, and is an alum of The Marketing Academy, one of the world’s most prestigious marketing leadership programs. On weekends, you’ll find her testing new AI tools, coaching, or planning her next global adventure.

    View Author Profile