Spec-Driven Marketing

Foreword#

You're already using AI for content. The output is mediocre. You think the problem is the model, or maybe the prompt. It's neither. The problem is that your brief was always vague — human talent just compensated for it. A skilled writer receiving "target: developers, tone: professional but approachable" would fill the gaps with professional judgment, audience intuition, and craft. The output was good enough despite the brief, not because of it.

AI doesn't compensate. It's the most literal executor in history. It does exactly what the input specifies and fills every gap with statistical defaults. A vague brief produces vague output — not because AI lacks capability, but because it lacks the tacit context that human executors provided for free.

This paper describes the infrastructure that fixes the input, not the tool. It introduces a methodology called spec-driven marketing — a way of making the thinking behind marketing activities explicit, inspectable, and improvable before execution begins. The spec is a knowledge object that carries complete business context: who you're reaching, what must change, how the activity achieves it, where it adapts, and whether it worked. Every dimension can be validated, debugged, reused, and inherited.

The methodology is not about AI. It serves human-only teams. But AI made the problem visible, and the assembled spec turns out to be the complete prompt architecture that AI-assisted marketing actually needs.

What follows is the formal argument.

The problem: marketing's missing knowledge object#

Software engineering has the spec. Law has the contract. Finance has the model. Each of these disciplines organizes professional knowledge around a unit that is explicit, inspectable, and improvable. Marketing has no equivalent.

The closest candidate — the brief — describes a deliverable, not a mechanism. A brief says what to make. It does not say what must be true for it to work. The distinction is fundamental: a brief that says "target: developers, tone: professional but approachable" contains no testable assumptions. When the resulting content underperforms, there is nothing to debug. The team cannot distinguish between a wrong strategy, a missed execution, and a failed distribution — because the strategy was never made precise enough to be wrong in a specific way.

This is not a talent problem. It is an infrastructure problem. Marketing lacks a knowledge representation that makes thinking inspectable. The consequence is threefold:

Diagnostic failure. When content underperforms, teams default to surface-level fixes (different headline, different channel) rather than identifying which underlying assumption broke. This is the organizational equivalent of single-loop learning (Argyris, 1977): adjusting output without examining the model that produced it.

Knowledge evaporation. What a team learns about its audience, its channels, and its arguments exists as tacit knowledge in individual heads. When people leave, the knowledge leaves. When teams scale, the knowledge dilutes. There is no mechanism for institutional learning because there is no artifact that captures what was learned.

Coordination collapse. In multi-product or multi-team organizations, coherence depends on senior people carrying shared context. This works at small scale and breaks at large scale. There is no formal mechanism for one team's audience insight to propagate to another team, or for strategic intent to cascade from organizational leadership to individual content decisions.

The System 1 organization#

Kahneman's distinction between System 1 (fast, intuitive, pattern-matching) and System 2 (slow, deliberate, analytical) thinking applies not only to individuals but to organizations. Most marketing organizations operate on organizational System 1: decisions are made through pattern recognition, experienced intuition, and "does this feel right?" heuristics. This is not a criticism — System 1 is efficient, and experienced marketers' intuitions are often correct.

The problem is that organizational System 1 is invisible. When it works, nobody can explain why. When it fails, nobody can explain why either. The thinking behind a decision lives in the intuition of the person who made it and is inaccessible to colleagues, successors, or the organization's institutional memory.

The spec is a mechanism for engaging organizational System 2 where it matters most: before execution begins. It forces the deliberate, explicit articulation of assumptions that would otherwise remain intuitive. This does not eliminate System 1 — experienced judgment remains essential during execution, creative development, and real-time adaptation. But it ensures that System 2 has been applied to the foundations (who are we talking to? what do we believe about them? what must change?) before System 1 takes over in the execution (how do we say it? what creative choices do we make?). The spec is not a replacement for intuition. It is a checkpoint that ensures intuition is operating on examined premises.

The proposition: the spec as knowledge object#

Spec-driven marketing introduces the marketing spec as the missing unit. A spec is a structured, explicit representation of the thinking behind any marketing activity — a content piece, a campaign, an event, a partnership, a product launch. It defines:

Who you are reaching (the audience's current state: situation, behavior, constraints, and mindset at the moment of encounter);

What must change (the specific shift in belief, understanding, emotional state, or attitude this activity must produce);

How the activity achieves it (the structural and tonal rules the execution must follow — including what is prohibited);

Where it adapts (how the core argument or experience changes shape for different channels and contexts);

Whether it worked (verifiable criteria, both pre-launch and post-launch).

A critical clarification: the spec's value is not in what it contains. Good marketers already think about their audience, their angle, and their evidence. The spec's value is in what it externalizes. The same thinking, when it remains in a practitioner's head, is invisible to colleagues, inaccessible to new team members, untestable by reviewers, and lost when the practitioner leaves or moves on. When that thinking is written into a spec, it becomes inspectable, challengeable, transferable, and improvable by anyone — not just the person who first developed it. The spec doesn't create better thinking. It makes existing thinking operational.

The three-failure diagnostic#

The spec's defining property is debuggability. When a spec-driven activity underperforms, the team can trace the failure to a specific category:

Each diagnosis produces a different corrective action. This is what distinguishes spec-driven failure analysis from conventional post-mortems, which typically conflate all three into "it didn't work, let's try something else." The diagnostic transforms marketing from a single-loop discipline (adjust outputs) to a double-loop discipline (examine and revise the governing assumptions).

Differentiation from existing practitioner frameworks#

Marketing practice already has structured planning artifacts: the creative brief, the message house, the brand key, the positioning statement. A legitimate question is whether the spec adds anything these frameworks do not. The answer requires addressing the closest existing ancestor directly.

The Get/To/By framework. The state-transition logic at the core of the spec — audience in state A, shifted to state B through mechanism C — is not new. It is the fundamental structure of the Get/To/By (or Get-Who-To-By) framework that has been the dominant strategic device in advertising briefs for decades, strongly associated with BBDO and widely taught across the industry. In its proper form, GTB defines: Get [an audience] Who [are in a specific state or face a specific problem] To [shift to a new state or behavior] By [a strategic mechanism or proposition]. The spec does not claim to invent this logic. It inherits it.

What GTB provides is a summary line — a concise articulation of strategy, typically one to three sentences, designed to align teams on creative direction. When used well, it captures the audience, the creative task, and the strategic mechanism in a form that is quick to read and simple to understand. Its popularity is deserved: few frameworks express the core strategic question more concisely. The spec's contribution is not the logic itself but the infrastructure around the logic that GTB does not provide.

Industry experience suggests that GTB is misused more often than it is used well — producing empty loops like "Get people to buy our product by showing them it's great" that are structurally complete but strategically vacant. The typical diagnosis is that practitioners need to "be more strategic." But this is an exhortation, not a mechanism. The spec provides the mechanism: a structure that makes empty loops detectable (the compiler catches a content contract that merely restates the intended shift), a diagnostic that identifies where the strategy broke when it fails (the three-failure diagnostic), and a knowledge system that accumulates what the organization learns (the library). GTB tells you to think strategically. The spec makes strategic thinking inspectable, validatable, and improvable.

The relationship is analogous to the difference between pseudocode and compiled code. Pseudocode captures the logic. Compiled code makes it executable, testable, and debuggable. Both express the same intent. Only one is operational. GTB is the pseudocode of marketing strategy. The spec is the compiled form.

Beyond GTB, other practitioner frameworks occupy adjacent but distinct spaces. The comparison with each reveals specific capabilities that existing frameworks lack and the spec provides:

Debuggability. A creative brief defines an audience and a message. A GTB line captures a strategic direction. Neither provides a diagnostic mechanism when the output underperforms. Was the audience definition wrong? Was the message right but the execution off? Was the execution perfect but the channel wrong? These frameworks cannot answer because they were never structured to separate these dimensions. The spec's five-component structure makes each dimension independently examinable.

Composability. A message house provides a hierarchy of messages. A brand key provides a positioning framework. A GTB line provides a strategic summary. All are standalone artifacts that must be manually interpreted and applied by each new project. The spec is composable: audience models, channel modules, and structural templates are reusable components that assemble into project-specific configurations. The same audience model can drive a blog post, a conference talk, an email campaign, and a product launch without being rewritten each time.

Accretiveness. Existing frameworks are static. A positioning statement is written once and updated infrequently. A creative brief is project-specific and discarded after launch. A GTB line is written per campaign and does not feed learning back into a persistent knowledge system. The spec feeds into a learning loop: every post-mortem updates the component library, every update improves future specs. Knowledge compounds rather than restarting with each campaign. No existing practitioner framework provides this dynamic.

The spec does not replace these frameworks — it can incorporate them. A team that uses GTB can recognize the spec as the operationalized, expanded form of their existing strategic logic. A team that uses message houses can embed their message hierarchy into the spec's content contract. A team that uses brand keys can derive spec constraints from their brand key. The spec is the operational layer that connects strategic frameworks to individual execution in a way that is inspectable and improvable.

Two separable claims#

It is important to distinguish two levels of argument in this paper, because they carry different degrees of certainty and invite different types of critique.

The theoretical claim: Marketing lacks a formal knowledge object — an artifact class with defined properties that makes thinking inspectable, transferable, and improvable. This absence is the root cause of diagnostic failure, knowledge evaporation, and coordination collapse. The claim is that marketing needs such an artifact, regardless of its specific design.

The design proposal: The five-dimensional spec (audience state, intended shift, content contract, channel adaptation, success test), the three-failure diagnostic, the library model, the inheritance mechanism, and the compiler concept represent one design that fills this gap. It is the design this paper proposes and defends. But it is an engineering proposal, not a logical necessity. The components could be refined, the diagnostic taxonomy could expand, the inheritance model could evolve. What cannot change without abandoning the methodology is the principle: marketing thinking must be externalized into a knowledge object with formal properties before execution begins.

This separation matters for intellectual honesty. The theoretical claim is strong and, to my knowledge, unaddressed in existing marketing methodology. The design proposal is grounded in practice but open to iteration. Critiques of the specific design ("why five components and not six?") do not threaten the theoretical claim. Critiques of the theoretical claim ("marketing doesn't need a knowledge object") would require demonstrating that the three consequences — diagnostic failure, knowledge evaporation, coordination collapse — have other root causes or are not real problems.

Methodology, not system#

A critical distinction: spec-driven marketing is a methodology — a way of working — not a system, a tool, or a process framework. This distinction determines what is fixed and what is variable.

What is fixed is the principle that marketing thinking must be made explicit, inspectable, and improvable before execution begins. The spec as the unit that carries this thinking. Debuggability as the defining property. The three-failure diagnostic as the analysis framework.

What is variable is everything else.

Marketing philosophy. The methodology does not prescribe what the spec contains at the conceptual level. A team that thinks in belief shifts writes specs around belief transitions. A team that thinks in emotional brand states writes specs around emotional state changes ("from 'this brand is irrelevant to me' to 'this brand reflects how I see myself,' verified by recall association in post-exposure survey"). A team that uses jobs-to-be-done writes specs around job completion. The methodology demands that whatever model you use, you make it explicit and testable. It is agnostic to the model itself.

Organizational structure. The methodology works in a one-person team (you write specs for yourself) and in a 200-person department (specs cascade across organizational layers). The mechanism of cascade — inheritance, ownership, conflict resolution — adapts to the organization. The principle remains: decisions flow down as constraints, learnings flow up as model updates.

Tooling. The methodology is tool-agnostic. Specs can live in Notion, Jira, YouTrack, Google Docs, or a shared folder of text files. The value comes from making thinking explicit, not from the container it lives in.

AI integration. AI is an accelerant, not a requirement. Specs improve human-only workflows (clearer briefs, better reviews, institutional memory) and AI-assisted workflows (better prompts, more targetable agents, validatable output). The methodology predates the AI moment; the AI moment makes it urgent.

Three properties of the spec as knowledge object#

The spec's value derives from three properties that briefs, decks, and tribal knowledge do not have.

Portability. A spec can be handed to a different executor — a new team member, a freelancer, an agency, an AI agent — and produce structurally equivalent output. Not identical output (creativity varies) but output that serves the same mechanism: same audience, same intended shift, same evidence logic. The spec is the portable unit of marketing intent. A brief is not portable in this sense because it relies on shared tacit context that doesn't travel with the document.

Composability. Specs compose from reusable components. An audience model, once validated, can be plugged into dozens of specs across different campaigns and activity types. A channel adaptation module, once developed by a channel expert, can be inherited by any spec targeting that channel. This is the library model: individual components are validated independently and assembled into project-specific configurations.

Composability also operates vertically. A strategy-level spec (which audience segments, which bets, which constraints) decomposes into campaign-level specs (which sequence of shifts, which activities, which channels), which decompose into artifact-level specs (this piece, this audience, this shift). Each level inherits from the one above. If a strategic constraint changes, it propagates. If a strategy cannot decompose into specs — into specific influence points with specific audiences and specific intended effects — it was never operational. It was a slide in a deck.

Accretiveness. Specs accumulate institutional knowledge. Every post-mortem that updates an audience model, refines a belief shift, or discovers a channel-specific constraint adds to a persistent, searchable body of validated knowledge. This knowledge compounds across projects, quarters, and team changes. A new team member reads the spec library and understands, in hours, what the organization has learned about its audiences over months or years.

This is the property that makes spec-driven marketing an organizational methodology rather than an individual skill. Individual marketers have always carried implicit specs in their heads — experienced practitioners know their audience, their angles, their evidence. The methodology's contribution is making this knowledge organizational: explicit, transferable, and improvable by anyone, not just the person who first developed it. This is the operational implementation of Nonaka and Takeuchi's knowledge spiral (1995): tacit knowledge (practitioner intuition) becomes explicit knowledge (the spec), which becomes systematic knowledge (the library), which generates new tacit knowledge (updated practitioner judgment informed by the library).

The spec as context assembler#

The three properties above describe what the spec is. This section describes what the spec does in practice: it functions as a context assembler — a composition node where every piece of business context relevant to a marketing activity converges by reference.

Consider a social media post. In a conventional workflow, the person producing it has access to a brief (maybe), a brand guidelines document (maybe), their memory of the last team meeting, and whatever they personally know about the audience. The gap between what the organization knows and what reaches the executor is enormous. Context is lost at every handoff, every interpretation, every assumption that "everyone knows what we mean."

In a spec-driven workflow, the spec for that same post assembles its context from the library:

Audience state → references a validated audience module (e.g., embedded-c-developers-v3), which carries its full history of validation and refinement across prior campaigns.

Intended shift → inherits from the campaign-level spec, which inherits from the product strategy, which inherits from the organizational strategy. The post's intended shift is traceable to the top.

Content contract → inherits brand-level constraints (voice, prohibited patterns, evidence standards) and adds artifact-specific rules for this particular post.

Channel adaptation → references a channel module (e.g., reddit-embedded-communities), built by someone who deeply understands that channel and reusable by anyone writing for it.

Success test → inherits campaign-level success criteria, adds artifact-specific proxies.

The spec for a single post is therefore not a standalone document with five fields to fill in. It is the leaf node of a context tree. You can look at that one spec and trace every decision: why this audience (audience module, validated by three prior post-mortems), why this constraint (brand spec, inherited from org level), why this format (channel spec, refined by channel team), why this shift (campaign spec, which decomposes from product strategy). The spec carries its own justification chain. Nothing is assumed. Nothing requires someone to "just know" the context.

This means each of the five dimensions in any given spec can be written fresh (unique to this artifact, based on new thinking), referenced from library (a validated, reusable component — an audience model, a channel module, a brand voice spec), inherited from parent (cascaded from campaign, product, or organizational level), or composed from a combination: inherited constraints plus artifact-specific additions. This is where the library earns its value. Every validated component that a spec references is a piece of organizational knowledge the individual executor does not need to rediscover.

Business context and information context#

The assembly model surfaces an important distinction. The spec provides complete business context: the strategic rationale, audience model, brand constraints, channel norms, campaign objectives, success criteria, and institutional knowledge relevant to the activity. This is the why and the what for of the activity.

What the spec does not provide is information context: the raw material the executor works with — product documentation, customer quotes, competitive data, technical examples, research findings. In the worked example later in this paper, the content contract specifies "one real before/after code example on STM32, showing register-level patterns." But the actual code example is not in the spec. It comes from engineering, from the product, from a real codebase.

The spec is the contract. The source material is the input. They are different artifacts that come together at execution time. The spec defines what the executor must achieve and why. The source material provides what the executor must work with. Confusing them is a common failure mode in conventional workflows — teams that dump product documentation into a brief and call it "context" have provided information without direction. The spec ensures that direction is always present, always explicit, and always traceable to strategic intent. The information context then serves the direction, rather than substituting for it.

The compiler analogy (and its limits)#

The engineering analogy that grounds this methodology requires one important clarification. A software compiler validates code against the rules of the language. It does not evaluate whether the resulting program is a good program. It checks whether the code is structurally valid, internally consistent, and free of known error patterns.

A marketing spec "compiler" operates the same way. It does not predict whether a marketing activity will succeed. It validates whether the spec is well-formed: Is the audience state specific enough to be actionable? Is the intended shift sized for a single activity? Does the content contract serve the intended shift? Are there known anti-patterns present (feature lists without context, superlatives without evidence, tone instructions that contradict the audience state)?

This is a crucial distinction. The methodology does not claim to make marketing outcomes deterministic. Marketing operates in a stochastic domain: audience behavior is probabilistic, market context shifts, creative resonance is partly irreducible. What the methodology claims is narrower and more defensible: it makes the inputs to marketing decisions inspectable, so that outcomes — whether good or bad — produce specific, reusable learnings rather than ambiguous impressions.

The compiler validates the spec. The market validates the outcome. They are different tests at different stages. Conflating them would overstate the methodology's promise. Distinguishing them is what makes it honest.

Like software compilers, the marketing compiler's rules evolve. Every post-mortem that reveals a pattern of spec failure ("specs without prohibited elements consistently produce generic output") generates a new validation rule. Over time, the compiler embodies the organization's accumulated wisdom about what makes specs fail. New team members benefit from every lesson the organization has learned, encoded not as tribal lore but as automated checks.

Relationship to existing work#

Spec-driven marketing does not invent new marketing theory. It operationalizes existing theory that has been difficult to apply systematically.

The intended-shift model (in its belief-shift variant) draws on the Elaboration Likelihood Model (Petty & Cacioppo, 1986) and dual-process theories of persuasion. The audience state concept extends Jobs-to-Be-Done theory (Christensen et al., 2016) from product experience to the moment of content encounter. The three-failure diagnostic implements Argyris's double-loop learning (1977) by making the "governing variables" of marketing decisions explicit and revisable. The knowledge inheritance model parallels Nonaka and Takeuchi's knowledge spiral (1995): tacit knowledge becomes explicit knowledge becomes systematic knowledge. The System 1/System 2 framing (Kahneman, 2011) provides the cognitive model for the organizational shift: from intuitive, invisible decision-making to deliberate, inspectable decision-making at the foundational level.

The contribution is not theoretical but translational. These theories are well-established but largely disconnected from marketing practice. Practitioners don't cite Argyris when doing post-mortems. They don't reference Nonaka when onboarding new team members. The spec is the mechanism that connects these theories to daily practice: a single artifact where persuasion theory, audience modeling, organizational learning, and knowledge management converge into something a practitioner can write in a working session and use immediately.

At the practitioner level, the spec's closest ancestor is the Get/To/By framework (also known as Get-Who-To-By), the dominant strategic summary device in advertising briefs for decades. The state-transition logic — audience in one state, shifted to a new state through a mechanism — is shared. The spec inherits this logic and does not claim to originate it. What the spec adds is the formalization of this logic into a knowledge object with defined properties and a surrounding infrastructure (diagnostic taxonomy, component library, validation layer, inheritance model) that GTB was never designed to provide. The relationship is one of inheritance, not displacement: GTB captured the right strategic question. The spec makes it operational.

Boundary conditions#

The methodology is most powerful in contexts where the audience makes considered decisions influenced by beliefs, evidence, and arguments (B2B, professional services, technology, education, healthcare). The belief shift model is native to these contexts. It's also strongest where multiple people or teams must coordinate on marketing quality without bottlenecking on individual taste or senior review (multi-product organizations, distributed teams, agency relationships). And where the cost of iteration is high: execution takes days or weeks, not minutes; feedback loops are slow; getting it wrong is expensive. Specs front-load the thinking to reduce expensive rework.

The methodology is less native to (but not excluded from) contexts where decisions are associative or emotional rather than deliberative (mass consumer brand), where iteration is cheap and fast (performance marketing with real-time A/B testing), or where the primary mode is creative disruption that resists pre-specification (breakthrough brand campaigns). In these contexts, the methodology can still contribute — emotional states can be specified ("from 'this brand is invisible to me' to 'this brand signals that I have taste,' verified by unaided brand recall and association mapping"), creative hypotheses can be made explicit — but it is a supporting discipline rather than the primary tool.

A boundary condition the methodology must acknowledge: it requires that someone in the organization can articulate audience assumptions explicitly. In mature organizations with experienced marketers, this is a translation exercise — the knowledge exists tacitly, the spec externalizes it. In immature organizations, early-stage startups, or teams entering unfamiliar markets, the knowledge may not yet exist.

In these cases, the spec functions differently: it becomes a hypothesis document rather than a knowledge document. The audience state is not "what we know about this audience" but "what we assume, to be tested." This is still valuable — it makes ignorance visible and structured, which is better than invisible ignorance — but it requires a different orientation. The spec is not encoding validated knowledge; it is framing testable bets. Post-mortems then become the primary mechanism for converting hypotheses into validated models. The methodology still applies; the starting position is different.

Why now: the AI inflection#

The methodology is not an AI methodology. It serves human-only teams. But the rise of AI as a marketing execution layer creates an inflection point that makes the methodology urgent.

Before AI, marketing could tolerate implicit thinking because human executors compensated with tacit knowledge. A skilled writer receiving a vague brief would fill the gaps with professional judgment, audience intuition, and craft. The output was often good enough despite the brief, not because of it. The thinking was invisible but the talent absorbed the cost.

AI agents do not compensate. They are, functionally, the most literal executor in history. They do exactly what the input specifies and fill gaps with statistical defaults. A vague brief produces vague output — not because the AI lacks capability, but because it lacks the tacit context that human executors provided implicitly. The spec is what makes that context explicit and transferable to any executor, human or artificial.

This reframes the AI value proposition for marketing. AI does not replace marketing thinking. It reveals whether marketing thinking existed. If your brief can't be executed literally and produce good output, the problem was never the executor. The problem was the brief. AI just stopped hiding it.

The assembly model makes this concrete. In current AI-assisted workflows, teams give an AI agent a prompt ("write a blog post about X") and perhaps paste in brand guidelines or product documentation. The result is generic because the input is generic: the AI fills every unspecified dimension with statistical defaults — default audience, default tone, default structure, default intent.

An assembled spec changes this fundamentally. The spec does not accompany the prompt. The assembled spec is the prompt architecture. Audience model, brand constraints, channel norms, campaign intent, content contract, success criteria — all formally structured, all referenced from validated library components. The AI agent does not need to guess at any of these dimensions. It needs to execute the contract. The difference is not incremental. It is the difference between an AI operating on "write something about this topic" and an AI operating on "here is the complete strategic context, the validated audience model, the specific intended shift, and the rules for how to achieve it."

The information context (product documentation, technical examples, customer data) then becomes the second input to the AI — the material it works with, governed by the spec's contract for how that material must be used. The spec provides the direction. The source material provides the substance. Together, they give the AI everything it needs to produce output that is strategically aligned, audience-appropriate, and verifiable against explicit criteria — rather than plausible-sounding content that no one can evaluate because no one specified what it was supposed to achieve.

A worked example#

In progress.

The core implications#

If the spec is the unit of marketing knowledge, then several things follow:

Strategy that doesn't decompose into specs isn't operational. A strategy deck that cannot produce a set of influence points — specific audience states to target, specific shifts to achieve, specific evidence to deploy — is not a strategy. It is a narrative without a mechanism. The spec is the bridge between strategic intent and operational reality. If the bridge doesn't exist, the strategy is aspirational.

Quality becomes structural, not taste-based. Review shifts from "does this feel right?" to "does this fulfill the contract?" Taste still matters — but it operates on top of structural quality, not instead of it. This makes quality transferable across reviewers, scalable across teams, and teachable to new members.

Marketing becomes a learning discipline. Every spec that fails produces a specific, reusable correction. Every correction updates the library. Every library update improves future specs. This is the compounding loop that marketing has lacked: not optimizing individual campaigns, but building an accumulating body of validated knowledge about audiences, arguments, and channels.

The productivity question inverts. The marketing productivity industry focuses on process: how to move work faster through approval chains, calendars, and workflows. Spec-driven marketing focuses on substance: how to ensure the content of the thinking is sound before the process begins. Process optimizes how. Specs optimize what. The industry has been solving the wrong problem.

Falsifiability#

A methodology that cannot be disproven cannot be taken seriously. Spec-driven marketing makes specific, testable predictions about what should change when it is adopted. If these predictions fail, the methodology is wrong.

Prediction 1: Diagnostic precision improves. Teams using specs should be able to categorize content failures into the three failure types (spec wrong, execution missed, distribution failed) with significantly higher agreement than teams using briefs. If teams using specs show no improvement in diagnostic consistency, the spec does not deliver its core promise.

Prediction 2: Revision cycles decrease. Content produced from specs should require fewer revision rounds than content produced from briefs of equivalent scope, because misalignment is caught at the spec stage rather than the review stage. If revision cycles are unchanged, the front-loading of thinking is not producing the expected efficiency gain.

Prediction 3: Knowledge accumulates and transfers. Teams maintaining spec libraries should onboard new members faster and produce more audience-appropriate content from new members earlier than teams without spec libraries. If new members using the library produce output no better than new members without it, the accretiveness property is not functioning.

Prediction 4: Post-mortem specificity increases. Post-mortems on spec-driven campaigns should produce more specific, actionable learnings ("the audience state was wrong in this specific way") than post-mortems on brief-driven campaigns ("it didn't perform well"). If post-mortem quality is unchanged, the debugging property is not delivering.

These predictions are testable in any team that adopts the methodology alongside a control practice. The honest claim is not that spec-driven marketing guarantees better outcomes, but that it produces better learning from any outcome. If it fails to produce better learning, it has failed on its own terms.

Summary#

Spec-driven marketing is a methodology built on a single principle: marketing thinking must be made operational — tangible, inspectable, and improvable — before execution begins.

The paper advances two claims. The theoretical claim: marketing lacks a formal knowledge object, and this absence is the root cause of diagnostic failure, knowledge evaporation, and coordination collapse. The design proposal: the marketing spec — a five-dimensional knowledge object functioning as a context assembler — fills this gap.

The spec is portable (it survives handoff to any executor), composable (it assembles business context from validated library components and inherits from strategic layers above), accretive (it accumulates institutional knowledge through structured post-mortems), and debuggable (its three-failure diagnostic identifies whether the thinking, the execution, or the distribution failed). Each dimension of the spec can be written fresh, referenced from a validated library, inherited from a parent spec, or composed from a combination — making every individual artifact a leaf node in a context tree that traces from organizational strategy to execution.

The methodology is agnostic to marketing philosophy (it operationalizes whatever model you already have), organizational structure (it scales from one person to a multi-business-unit enterprise), and tooling (it works in any project management system or none). AI integration is an accelerant, not a premise — but the assembled spec functions as a complete prompt architecture, transforming AI from a tool that produces plausible content into a tool that executes strategic contracts.

The contribution is translational: established theories of persuasion, audience modeling, organizational learning, knowledge management, and cognitive decision-making, unified into a single practitioner methodology through the spec as the connecting artifact. The state-transition logic at its core — audience in state A, shifted to state B through mechanism C — is inherited from decades of strategic practice. What is new is the formalization of that logic into a knowledge object with properties that enable organizational dynamics previously unavailable: compounding institutional learning, structural diagnostics, and vertical traceability from strategy to individual execution.

Or, more simply: if you can't inspect the thinking behind your marketing, you can't improve it. The spec makes the thinking visible. Everything else follows.

This is a concept paper, not a finished framework. It describes the theoretical contribution of spec-driven marketing at the methodology level. Companion documents detail the operational implementation: spec anatomy, execution models, adoption playbooks, system architecture, and organizational planning architecture.

© Ilya Petrov, 2026. Draft for review and discussion.