Marketing Insights
Most startup marketing capabilities never fully crystallise. They persist in an intermediate state — recognisable patterns that work but never stabilise into transferable processes.
You decide to "do content marketing." Week one, you write a blog post. Week three, you write another. Week six, you've missed two weeks because a customer emergency took priority. Week ten, you write one more, slightly different in tone because you've learned something. Week fifteen, your co-founder starts contributing, but with a different approach.
Is this a process? A capability? A mess?
The answer is: it's all three simultaneously. Most startup marketing lives in that intermediate state. Not chaos, not system. Not in the polished world of editorial calendars and content pillars. Not in the aspirational strategy deck that describes what marketing should look like. But in the messy, half-formed, partially standardised, frequently interrupted reality of what marketing actually is when one or two people are doing everything under continuous resource pressure.
This post is about that reality. When I studied thirteen marketing leaders navigating exactly this condition, I found something the strategy frameworks don't account for: most startup marketing capabilities never fully crystallise. They persist in an intermediate state I've termed the persistent proto-routine. These are recognisable patterns that work but never stabilise into transferable processes, because the resource pressure never lets up long enough for formalisation to occur. And those capabilities that do crystallise can collapse through four distinct mechanisms: fear, impatience, external pressure, or environmental shift. Each of these breaks the feedback loop that keeps the routine alive. What follows is the empirical account of how marketing capabilities emerge, get stuck, and fragment, grounded in routine dynamics theory and sixty-nine coded passages from original research.
Key Concepts in This Article
Ostensive-Performative Gap: A core concept from routine dynamics theory (Feldman & Pentland, 2003\) describing the structural mismatch between how leaders describe their marketing processes (the ostensive routine — "how we do content marketing") and what actually happens in practice (the performative routine — what someone would observe on a given Tuesday). In routine dynamics, this gap is not a failure to be corrected but the generative engine of capability innovation. This research applies the framework to startup marketing leadership, finding that all thirteen participants exhibit the gap — and that it drives capability construction under constraint.
Persistent Proto-Routine (Stage 2.5): The concept of proto-routines — partially formed organisational patterns that have not yet stabilised — appears in prior routines literature. This research extends that concept with a specific finding: in startup marketing under chronic resource constraint, proto-routines don't resolve. They persist as a stable intermediate state — recognisable as a pattern, partially standardised, and durable over time, but never fully crystallising into a reliable, transferable process. Identified in 16% of routine-coded passages. Designated "Stage 2.5" between ad-hoc activity and established routine.
Routine Bundler: A theoretical account, grounded in routine dynamics, of the emergent "GTM engineer" role identified in practitioner discourse — in which a single individual performs demand generation, content production, campaign experimentation, and performance tracking routines simultaneously, enabled by AI tooling that collapses skill barriers between traditionally separated functions.
AI Co-Constitution: Drawing on sociomaterial practice theory (Orlikowski, 2007), this finding identifies generative AI tools functioning as constitutive elements of marketing capability, not merely as productivity enhancers. Remove the AI component, and the routine as currently configured ceases to function. Observed in 6 of 13 participants.
Meta-Routine: A concept from organisational theory describing a routine whose purpose is sustaining other routines — budget reallocation, CRM maintenance, stakeholder management. This research finds meta-routine activity dominates the coded data: twenty-three percent of all coded routine passages describe meta-routine activity, making it the largest single coding category in the startup marketing context.
Activation Trap (Routine-Level): The structural mechanism by which demand-generation routines crowd out brand-building routines at the level of daily practice. Demand routines are more materially embedded (in CRM configurations, reporting dashboards, investor expectations) and produce stronger reinforcement signals, causing brand capability to degrade through under-practice rather than deliberate neglect.
Every marketing leader has two versions of their marketing. There's the version they describe (the ostensive routine, in academic language) which is what you'd say if someone asked "how do you do content marketing?" You'd describe a process: we identify topics based on keyword research, write posts on a regular cadence, distribute through LinkedIn and email, measure engagement and pipeline impact.
Then there's the version that actually happens: the performative routine. It's what you'd see if you followed that same leader around for a month. Opportunistic content when bandwidth allows. Distribution when someone remembers. Measurement when the CEO asks.
In the research, every single participant articulated marketing practices as they should operate that diverged from what they actually did. All thirteen showed this mismatch. This isn't a finding about a few leaders falling short. It's a universal structural condition.
The brand investment gap is the most consistent pattern. One startup-native leader puts it starkly: academics prescribe sixty percent brand investment, yet brand is "not the focus area for me at all" because "it's hard to quantify." A corporate-transplant leader reports that "the minute you speak to anyone outside of marketing about brand, they glaze over, they check out emotionally." This requires a political strategy: first demonstrating demand-generation results before brand investment becomes organisationally tolerable. Another declares "brand first, absolutely brand first" while acknowledging that a ninety percent budget cut would make brand work impossible.
One participant resolves this tension in a way that illuminates the broader dynamic: "100% of your time should be spent on brand... but you do not have the luxury of running campaigns that are brand only." The resolution isn't choosing brand over demand or demand over brand. It's collapsing the distinction. Brand gets embedded within demand-generation activities so that every piece of activation work also does brand work. This isn't a compromise. It's a structural innovation forced by the gap between what you know you should do and what your resources permit.
A second systematic gap appears in measurement. The ideal, as one corporate-transplant leader describes it: "You track everything, your cost, your internal investments, make it all very tangible." The reality: CRM management was "a side job" before eventually receiving dedicated resource. This is the experience paradox in action. The leader's corporate background provides the measurement ideal but not the resource to implement it. Another participant offers the most consequential observation: "The power of politics surpasses by far data... even for data-driven organisations." This suggests something uncomfortable: measurement routines can exist performatively. Data is collected, dashboards are maintained, reports are generated. But they produce no actual change in decisions. The routine operates. Its intended function is displaced by organisational politics.
Here's the reframing that matters: the gap isn't failure. It's the engine of capability innovation. Leaders who recognise the mismatch between their imported patterns (from corporate experience, industry advice, academic frameworks) and their actual conditions are forced to construct novel approaches. This is the metacognitive paradox made productive: knowing what you should do but can't creates the tension that drives the construction of what you actually can do. The ostensive-performative gap is not a problem to be solved. It's a condition to be navigated. And navigating it is precisely how new marketing capabilities emerge.
The Data at a Glance
The routine dynamics re-analysis coded 69 passages across all 13 participants, producing the most granular view of marketing capability mechanics in the research programme. The breakdown by routine category:
The 2.5:1 demand-to-brand ratio at the routine level shows the activation trap operating structurally, not just cognitively. Brand doesn't just receive less budget. It receives less practice.
The data documents three distinct pathways through which ad-hoc activity crystallises into recognisable routines.
One startup-native leader built the company website in Webflow, teaching themselves development through ChatGPT. This was a pragmatic response to having no budget for a developer. It's effectual logic in action: start with available means, produce a tangible result, and let demonstrated outcomes attract formal resource allocation. An advisory participant describes the same pattern: grow proof points through low-budget execution until results justify investment. The routine emerges from demonstrated results, not planned investment.
A corporate-transplant leader established a budget-based growth motion: "if you put this kind of amount of money in, we get this amount of money out." Others built routines around shared KPIs or Google Ads performance data. The metric creates the anchor. It's the measurable signal that stabilises the routine by providing clear feedback. This pathway is characteristic of corporate-transplant leaders who bring structured measurement habits from enterprise environments and adapt them to constrained contexts.
One startup-native leader conducts audience-habitat analysis before selecting channels. They map where target buyers actually spend attention before investing in channel-specific routines. The routine emerges from environmental fit rather than industry convention. This is cognitive search applied to channel strategy: the leader gathers environmental intelligence before committing resource, reducing the risk of building routines in channels where the audience doesn't exist.
These pathways suggest that capability emergence isn't random. Recognising which pathway you're on helps you support the emergence rather than disrupting it with premature formalisation.
Most discussions of marketing processes assume two states: you either have a process or you don't. Strategy or chaos. System or improvisation.
The data reveals a third state that's far more common than either extreme. It's the persistent proto-routine.
A proto-routine meets some but not all of the criteria for a genuine routine. It's recognisable as a pattern. You can see that the leader does something similar each time. It's partially standardised. Some elements repeat while others vary. It persists over time. But it never fully crystallises into a reliable, transferable process. It works, but it requires the founder's full attention every time it executes.
Approximately sixteen percent of the coded routine passages describe this state. One startup-native leader captures it precisely: "DIY... lacking processes... rushing scenarios, lack of budget, lack of time." The activity is recognisable but never formalises because the resource pressure never lets up long enough for formalisation to occur. This isn't a transitional phase. It's a stable condition under chronic constraint. The research framework designates this as "Stage 2.5" between ad-hoc activity and established routine.
Another leader's repeated attempts at account-based marketing illustrate the proto-routine's frustrating persistence. The team targets fifty to sixty enterprise logos, running ABM campaigns through multiple iterations as "strategies have been evolving with a lot of gurus" offering conflicting approaches. Sales cycles exceed twelve months. Buying committees span CIO, procurement, finance, and CEO levels on million-dollar investments. The ABM routine never stabilises because the contextual complexity defeats every standardised playbook. The leader acknowledges: "I don't think we have found a real kind of a model or a structure which I could say really works." The proto-routine reappears each time they try, producing recognisable patterns that never quite become reliable processes.
A third case shows proto-routine failure from a different angle: a solo founder attempted cold email outreach following Y Combinator's recommendation to "write them manually so it's hyper personalised." The messages were ignored. The founder questioned whether the problem was volume. "If I sent 100 emails someone should react, and if that doesn't happen, maybe I should send a thousand, I don't know." But they never invested in the infrastructure the routine required: multiple sending domains, email warm-up tools, automated sequencing. The ostensive pattern existed. Industry advice says to do outbound. But personal commitment to the performative execution never developed. The routine never stabilised because the founder couldn't tolerate its characteristically slow, uncertain feedback loop.
This is where many founder-scale startups get stuck. They can execute proto-routines: patterns of activity that produce recognisable outcomes. But they can't systematise without additional resource. And the additional resource doesn't arrive because the proto-routines are just functional enough to sustain the business without ever becoming efficient enough to free up capacity for investment. The dynamic has structural parallels with the activation trap: activity that is good enough to survive but not good enough to grow. The data doesn't track long-term outcomes of proto-routines, but the pattern of partial functionality under chronic constraint appears stable rather than transitional.
If capabilities are things you do rather than things you have, then they can stop. The data documents four distinct mechanisms through which marketing routines degrade.
One startup-native leader halted AI-assisted content production. It was a routine that was technically feasible and practically effective. But they stopped because of speculation about Google penalising AI-generated content. "During that period there was so much speculation about Google's going to punish you if you use generative AI... we held off." The routine degraded not through failure but through risk aversion. The feedback loop collapsed because fear of future consequences overrode present-tense evidence of the routine's effectiveness.
Another leader describes killing competitor advertising campaigns after a single week because they "didn't deliver any leads right away." The team's immediate reaction was simple: "Oh, it doesn't work. Just kill it." They didn't investigate whether the failure point was ad copy, landing page conversion, or the application process itself. The diagnostic loop that would allow routine improvement got truncated by short-term performance pressure. The routine never had time to stabilise because evaluation happened before the feedback could arrive.
A corporate-transplant leader describes how private equity governance systematically degrades creative and strategic routines. "It kills creativity... kills the ability for the team to really think about what are our strongest messages." The time horizon imposed by PE governance compresses the performative space available for routines that require extended development: messaging development, brand narrative construction, thought leadership. The participant's experience suggests that one routine's degradation can propagate through cascading dynamics. Degraded messaging quality leads to weaker demand performance, which produces poor metrics, which triggers further budget pressure. The data captures a single case rather than a generalisable mechanism.
One startup-native leader invested heavily in SEO optimisation based on industry advice. They invested "relative to our size we've done a lot of investment on that front." But the results were "next to nothing" after twelve months. The underlying problem was clear: "The market is moving so fast that there is no written rules yet, so the advice is to invest heavily on it." The routine failed not through execution problems but because the market shifted underneath it. This is a distinctly dynamic-capabilities challenge. The environment changed faster than the routine's adaptation cycle.
These four mechanisms share a common structural feature: they all involve the collapse of the routine's feedback loop. Effective routines operate through recursive cycles of performance, observation, and adjustment. Degradation occurs when this loop is disrupted. By fear, by impatience, by time compression, or by environmental velocity.
This is the finding that caught me most off guard, and the one with the most immediate implications for anyone building marketing capability right now.
Six of thirteen participants have built marketing routines in which generative AI is a constitutive element, not an optional enhancement. Remove the AI component and the routine as currently configured ceases to function. Though the leader could rebuild it using alternative, typically more expensive means.
One leader runs a complete SEO routine through founder plus ChatGPT plus a thousand-dollar contractor. Another produces entire marketing campaigns as a solo operator using "literally just me and AI." They generate qualified leads at five to six dollars each. The campaigns include strategy, persona development, landing page copy, logo design, brand guidelines, banner creative, and media buying. A third wrote video scripts, generated AI voiceover through a Motion Array subscription, and hired four freelance editors at roughly a hundred dollars per video. They acknowledged: "the end result is not great, but for a bootstrap solo founder, the cost was well worth it".
Three participants (all startup natives) independently describe the same emergent role: the "GTM engineer." One defines it as "someone who understands how to build systems that generate demand, capture that demand, bring people into the sales cycle." In routine dynamics terms, this is a routine bundler. It's one person performing demand generation, content production, campaign experimentation, data enrichment, and performance tracking simultaneously. AI tooling has collapsed the skill barriers between traditionally separated functions. The minimum unit of routine performance shifts from the team to the augmented individual.
This creates a tension. Capability innovation accelerates. A solo founder can produce at a volume that previously required a team. But capability fragility increases. You're dependent on tools you don't control, on AI parameters that change without notice, on pricing models that can shift overnight. Participants develop AI-specific knowledge (prompting techniques, aesthetic parameters, workflow sequences) that is tacit, personal, and non-transferable. If the founder leaves, the AI-augmented routine leaves with them. This is the network-mediated capability problem in concentrated form: the routine's constitutive elements are bundled in a single node.
A related finding: acquiring artifacts doesn't create capability. One participant bought Clay (a data enrichment tool) but couldn't implement it due to "lack of knowledge and lack of priority." The tool that was supposed to accelerate the team's data enrichment routine remained unused because the routine around it was never developed. Conversely, another participant reverse-engineered Clay's functionality in N8N "for $300 instead of $2000" (C1\_T2). The pattern is consistent. The routine constitutes the capability, not the artifact. Tools without routines are inventory. Routines with improvised tools are still capabilities.
Here's a number that reframes how you think about marketing leadership: sixteen of sixty-nine coded routine passages describe the continuous effort required to sustain existing routines. That's twenty-three percent, the largest single category.
Not building new capabilities. Not launching campaigns. Not strategic planning. Just keeping what you've already built from falling apart.
One leader describes CRM maintenance as starting as "a side job." Nobody on the team had CRM expertise, so tracking, data quality, and campaign attribution were bolted onto existing roles. The team repeatedly attempted to implement proper tracking but "never really got there, not 100%." Eventually the function received dedicated resource, but the data quality problems persisted. Another leader calls budget management "a daily problem." The team must continuously "dive into the books, scrape together funds from different buckets," reallocating existing budget rather than securing new investment. The marketing routines are too new to produce the ROI calculations that would justify incremental spend. A third must continuously "move around activities, deprioritise things, move funding" to maintain routine viability within a complex stakeholder environment.
These accounts describe meta-routines (routines for sustaining other routines) that consume substantial leadership attention. And they explain something that every marketing leader in a startup has felt but rarely articulates: the feeling that you're running just to stay in place. You're not imagining it. Twenty-three percent of all coded routine passages in the data describe maintenance activity. It's the largest single category. While coding frequency isn't a direct measure of time allocation, the pattern suggests that routine maintenance consumes a significant share of leadership attention that capability planning models rarely account for.
Brand maintenance presents a distinctive challenge within this pattern. One corporate-transplant leader built a thought leadership routine: technical articles demonstrating what the team knew, published consistently over years. It "took a very long time but I think it did work... it's still there even though we got acquired." The routine persisted through an acquisition because the continuous investment had embedded it into the company's market legitimacy. But another leader reports the consequence of abandoning brand routines: the team stopped advertising on their brand name, stopped attending events, and "when we lacked doing that, it impacted our growth... we underestimated the value of brand." The routine wasn't cancelled through a strategic decision. It simply stopped being performed. The capability degraded. A third advocates maintaining "a small allotment on brand for the long-term vision so you don't end up chasing leads forever".
The consensus across participants is clear: brand routines require continuous maintenance but are the first to be sacrificed under resource pressure. The activation trap, viewed through the routine dynamics lens, is a structural feature of the routine ecology. Demand-generation routines produce measurable outcomes that reinforce their persistence, while brand-building routines produce diffuse, long-horizon outcomes that provide weak reinforcement signals. Demand-generation routines are more embedded (in CRM configurations, in reporting dashboards, in investor expectations) and more materially supported (by automation tools, by attribution models) than brand routines.
This post applies routine dynamics theory to marketing practice. That sounds similar to several adjacent ideas. It isn't.
This is not marketing operations. Marketing operations focuses on making existing processes more efficient. It streamlines workflows, reduces friction, improves throughput. Routine dynamics asks a prior question: how do those processes come into existence in the first place? The emergence pathways and proto-routine concept describe what happens before there's anything to optimise.
This is not marketing automation. The AI co-constitution finding is not about automating tasks. It's about AI functioning as a constitutive element of the capability itself. Automation implies a pre-existing manual process that gets mechanised. Co-constitution means the routine was never possible without the AI component. It's a new kind of capability, not a faster version of an old one.
This is not agile marketing. Agile marketing prescribes planned iteration cycles: sprints, retrospectives, backlogs. The emergence pathways documented here describe organic crystallisation under constraint. Nobody in the data ran a sprint retrospective and decided to build a routine. They tried something, noticed it was working, and gradually recognised a pattern. The process is emergent, not managed.
This is not a capability maturity model. Maturity models (Gartner, Forrester, and their many imitators) assume linear progression from ad-hoc to optimised. The persistent proto-routine finding directly contradicts this assumption. Stage 2.5 is not a waypoint on the road to Stage 3\. It's a stable condition under chronic constraint. Some capabilities persist indefinitely in a partially-formed state, not because the organisation hasn't progressed, but because the resource environment never permits full crystallisation.
This is not growth hacking. The routine bundler concept describes how one person sustains multiple routines through AI augmentation. Growth hacking describes rapid experimentation to find scalable channels. The difference is ontological. Growth hacking asks "what works?" Routine dynamics asks a different question: "how does 'what works' become 'how we work'?" That's the process by which a successful experiment becomes a reliable, repeatable capability.
This post engages with three theoretical traditions that converge on a single claim: capabilities are not things you build and then have. They're things you do and must keep doing.
Routine dynamics (Feldman & Pentland, 2003). The foundational framework distinguishes between the ostensive aspect of routines (the abstract pattern, the "how we do things" narrative) and the performative aspect (the actual enactment, what people really do on Tuesday afternoon). Feldman and Pentland's central insight is that the gap between ostensive and performative isn't a deficiency to be corrected but a generative feature of organisational life. The tension between intention and action drives adaptation, innovation, and routine evolution. The marketing leadership data provides what may be the richest naturalistic illustration of this dynamic in an entrepreneurial context: thirteen of thirteen participants show the gap, and the gap drives capability construction.
Process ontology (Tsoukas & Chia, 2002). Process ontology argues that organisations are not stable entities that occasionally change. They are ongoing accomplishments that require continuous effort to maintain. Stability is the thing that needs explaining, not change. Applied to marketing capability: the question isn't "how do capabilities change?" but "how do capabilities persist?" The answer, from both theory and data, is through continuous performative investment. When the doing stops, the capability disappears. The four degradation triggers documented above are empirical cases of this dynamic.
Sociomaterial practice theory (Orlikowski, 2007). The AI co-constitution finding connects to Orlikowski's argument that practice is always entangled with material artifacts. The sociomaterial view argues that agency is distributed across humans and artifacts. The tool shapes what's thinkable, doable, and routine. When six of thirteen participants cannot execute their marketing routines without generative AI, the tool isn't supporting the routine. It's constituting it. Remove any element and the capability ceases.
The routine dynamics re-analysis coded sixty-nine passages across six categories from all thirteen participants, producing the most granular view of marketing capability mechanics in the research programme (C1\_T2).
The demand-brand ratio at the routine level reveals the activation trap operating structurally, not just cognitively. Forty of sixty-nine coded passages relate to demand-generation routines (fifty-eight percent of all coded routine activity). Sixteen relate to brand-building (twenty-three percent). Twelve relate to measurement (seventeen percent). Just one passage directly addresses market sensing. This suggests that market sensing in early-stage ventures doesn't exist as a standalone routine. It operates as an embedded activity within other routines (C1\_T1\_routine\_framework\_ENRICHED).
The two-point-five-to-one demand-to-brand ratio at the routine level shows the activation trap isn't just about how leaders think about resource allocation. It's about what they actually do with their time. Brand activity doesn't just receive less budget. It receives less practice. And since capabilities emerge from practice, less practice means less capability. The activation trap reproduces itself at the level of routine architecture.
Routine maintenance as the largest coding category (sixteen of sixty-nine passages, twenty-three percent) has implications for how we think about resource planning. Coding frequency is not a direct proxy for time allocation. It measures how often participants discussed maintenance, not precisely how much time they spend on it. But the fact that maintenance dominates the coded data suggests it occupies a substantial share of leadership attention that capability planning models rarely account for. The directional implication: when you plan your marketing capacity, recognise that a meaningful portion of your effort will go toward maintaining what you've already built, not building new capability. If you don't account for this, you won't lose capability suddenly. You'll lose it gradually, through the slow degradation of routines that nobody is tending.
This is the strongest theoretical confirmation in the research programme yet its weakest procedurally embedded finding. Thirteen of thirteen participants show capabilities emerging from pragmatic activity, not from pre-planned design. The dominant pattern is emergent rather than prescribed: try something → observe what happens → adjust → try again → notice that a pattern has formed → recognise the pattern as "how we do this" → maintain the pattern until it breaks or succeeds. Some Corporate Transplant participants bring structured elements to this process: budget-based growth motions, Pareto prioritisation. But even these structured approaches evolve through iteration rather than following a fully predetermined design.
The process ontology implication is direct. Marketing capability is not something you build and then have. It's something you do and must keep doing. It "becomes" through repeated action and can "un-become" when action stops. The four degradation triggers are four ways that "un-becoming" happens. The persistent proto-routine is a capability that is perpetually in the process of becoming without ever arriving. Routine maintenance (twenty-three percent of all coded activity) is the cost of preventing un-becoming.
This reframes how you should think about marketing capability development. You're not constructing a building that will stand on its own once finished. You're tending a garden that requires continuous attention to remain productive. The moment you stop tending (because of fear, impatience, external pressure, or environmental shift), the garden begins to revert. The capability doesn't just pause. It degrades. And once the tacit knowledge, contextual adjustments, and relational elements that made the routine work have dissipated, the foundation for rebuilding looks different from what existed before. This is why one participant's thought leadership routine took a very long time to build but remained valuable even through an acquisition. The continuous investment preserved what discontinuous investment would have lost.
If capabilities are ongoing accomplishments rather than permanent assets, and if they emerge through routine dynamics rather than strategic planning, then the question of legitimacy takes on new significance. How do these capabilities become recognised, trusted, and valued by stakeholders? The next post examines how B2B SaaS companies transition from operational trust ("they deliver what they promise") to category authority ("they define what this market means"), and why that transition is both more important and more fragile than most growth frameworks acknowledge.
Next in the series: Legitimacy Transitions in B2B SaaS — From Operational Trust to Category Authority
This post is part of a 10-part foundation series exploring how marketing capabilities emerge under constraint. The routine dynamics framework draws on Feldman and Pentland (2003), process ontology (Tsoukas & Chia, 2002), and sociomaterial practice theory (Orlikowski, 2007), grounded in original empirical research with 13 B2B SaaS marketing leaders. Browse all pillars.
