On organizational imprinting, strategic pivots, and what business research is actually for

I spent the last few days reading six papers on organizational imprinting and strategic change. This is preparation: I am about to start a PhD in management focused on strategic redirection in Chinese deep-tech ventures, after a decade of investing in exactly those companies. Before I let the field’s vocabulary reshape how I think, I wanted to record how I think now — as someone whose undergraduate training was in electronic engineering, with a later detour through strategic marketing at Cardiff, and with a portfolio of hard-tech companies whose founders I have watched pivot, refuse to pivot, and occasionally run out of cash while deciding.

What follows is not a literature review. It is a set of reactions, filtered through three habits I cannot switch off.

The lens

Counterfactual reasoning. Temporal precedence is not causation. Before I accept that X caused Y, I want to know whether Y would have happened with X removed. Most management papers describe a sequence and call it a mechanism; the counterfactual is rarely constructed, and often cannot be.

High-dimensional projection. Organizational capability is not a static trait you can read off a founder’s biography. It is a low-dimensional shadow of a system evolving in a much higher-dimensional state space — technology, capital, customers, regulation, timing. When a paper isolates one variable and finds it “explains” an outcome, my first question is what was held constant, and whether it could have been.

Signal-to-noise and transfer functions. A founder’s ego, mood, or rhetoric is real. So is thermal noise. The question is whether it is the transfer function of the system or a perturbation the system filters out. Confusing the two is the most common error I see in qualitative work.

These three habits compose into a model I keep returning to, borrowed from systems architecture.

Bare metal, hypervisor, guest OS

L0 — bare metal. Shannon limits. Thermodynamics. Process yield. The physics of a transistor at a given node. The date on which the bank balance hits zero. These constraints are indifferent to narrative.

L1 — hypervisor. The virtualization layer that human societies build by contract and consensus: money, corporate law, property rights, accounting standards. This layer is real and consequential, but it is constructed, and it can be rewritten — slowly, by institutions, or suddenly, by war or regulation.

L2 — guest OS and applications. Management narratives. Founder psychology. Brand mythology. Organizational identity. Public relations. This is where most management scholarship lives.

The recurring failure mode of social science, as I see it from the outside, is mistaking an error message thrown by the guest OS during resource contention for a law of the underlying hardware. Everything in the guest layer runs fine until a non-maskable interrupt arrives — cash exhausted, physics not cooperating, a supply chain severed by geopolitics — and then the elegant narrative is cleared from memory in a single cycle. Investors live at the interrupt boundary. Scholars mostly do not.

Physics envy, and why I think it is backwards

There is a standing hierarchy among the sciences, and everyone in the social sciences knows their place in it. Physics is a real science because nature is precisely modelable: the same experiment run twice gives the same answer, and the answer is a number. Social science is soft because people are messy. The field’s response to this ranking has a name — physics envy — and it is visible in the reflexive reach for formalism: regressions on constructs nobody measured, Greek letters carrying no dimensions, significance stars purchased from a sample of twenty-nine.

I accept the diagnosis. I think the ranking it responds to is backwards, and the layer model is why.

Natural science runs on the physical machine. It has hardware access. A physicist can put a thermometer on the thing, a scope probe on the node, a counter on the events; the instrument touches the phenomenon and nothing sits in between. Better still, the experiment can be rerun with exactly one parameter moved, because the machine is the researcher’s to reset.

Social science is a process running inside a virtual machine, trying to characterize the machine from inside. From inside a guest you cannot read the host’s die temperature, its fan curve, or how many other tenants are on the box. Your clock drifts and you cannot tell whether the cause is your own workload or a noisy neighbour you will never be shown. Every instrument you do have is itself virtualized — the survey, the interview transcript, the accounting standard — an emulated device faithfully reporting whatever number the hypervisor decided to present. And you cannot reset the machine and run it again, because there is only one instance and you are in it.

So the underdetermination that makes social science look soft is not a failure of rigour. It is a property of the observation position. Social science is not the easier science with looser standards; it is the harder one, attempting inference with no access to ground truth, and it is harder in a way that importing more formalism does not fix.

That has a consequence for how the field’s failures should be read. Faking the missing instruments — running statistics on quantities that were never observed — is one way of refusing to say “I cannot see the host.” Abandoning causal claims altogether and retreating into narrative and discourse is the other. Both are evasions of the same admission. The honest move, and the standard I hold these six papers to, is to state which layer a claim lives on and how much confidence that layer’s instruments actually support.

With that established, the papers.

1. Marquis & Tilcsik (2013): the geological map

Imprinting: Toward a Multilevel Theory. Academy of Management Annals.

What it says. Imprinting has three elements: a brief sensitive period (founding, transition), an environment that stamps its features onto the entity, and persistence of those features across later environmental change. The authors carefully separate imprinting from path dependence (contingent events plus increasing returns that lock in) and from cohort effects (shared ongoing experience of a generation). They add two dynamic concepts: sedimentation — successive imprints layering and conflicting — and exaptation — a trait imprinted for environment A later becoming the core advantage in environment B.

What I think. The framework matches my field intuition. Imprints are habits internalized as culture and working style; organizations keep executing them after the environment changes and after the exit cost has fallen. That is precisely what distinguishes an imprint from path dependence, and the distinction is worth keeping.

Two problems.

First, the empirical base for individual-level imprinting leans on studies of people who entered a profession during a downturn and remained risk-averse for life. I cannot read those results without seeing a selection effect. The people who survived a recessionary entry were not necessarily reshaped by it; they may simply have been the risk-averse subset that got selected and retained. Without a comparison against people who entered in the same period and left, “imprinting” and “selection” are observationally equivalent.

Second, exaptation is unfalsifiable as usually deployed. Any success can be reconstructed afterwards as a clever repurposing of some earlier trait. To make the concept do work, you need a pre-registered cut: define the imprint before the new environment appears, define the control group, and check whether firms carrying the imprint outperform those without it. Absent that, exaptation is a story-completion device.

A smaller point on multiple imprints. Serial founders carry stamps from several periods, and the stamps are a function of time — but transmission from founder to organization is not lossless. It is a signal through a noisy channel, attenuated and filtered by boundary conditions: the board, the investment agreement, the heterogeneity of co-founders. Papers that treat founder-to-firm transmission as roughly complete are modeling a channel with zero loss, which no engineer would accept.

2. Becker (2025): the Zeiss myth and the hidden reservoir

Mechanisms of Organizational Imprinting: From Entrepreneur to Organization. Administrative Science Quarterly.

What it says. Founders do not need a modern managerial blueprint. They need a task blueprint — a way of doing the core work — and that alone can shape the organizational skeleton. At Zeiss this was Carl Zeiss’s obsession with precision manufacture combined with Ernst Abbe’s scientific, mathematical approach to optical design. Four micro-mechanisms (early decisions, direct teaching, role modeling, and the legal charter of the Carl Zeiss Foundation) fixed the blueprint into structure, behavior, and product. The evidence for persistence is striking: after 1945, East and West Zeiss developed in isolation for forty years, yet their patent portfolios remained highly correlated (0.93).

What I think. The paper is well executed and the archival work is serious. My objections are conceptual.

Concept boundaries. Without a strict hierarchical definition, “task blueprint” collapses into concepts evolutionary economics already has — technological paradigm, operating routine. A new label is not a new mechanism.

Origin versus maintenance. Abbe teaching two hours a day is a plausible origin mechanism. But after 1945 both Abbe and Zeiss were long dead, the Jena works had been dismantled, and the West German operation was rebuilt by 77 people on a field in Oberkochen. Micro-interaction with the founders had been zero for decades. What actually maintained the organization through that discontinuity was Weberian bureaucracy and the cold legal text of the Foundation statute. The paper’s micro-mechanisms explain how the imprint was formed; they do not explain how it survived, and the survival is the paper’s headline result.

The atomistic fallacy, or crediting the reservoir to the ladle. Look at where high-precision manufacturing actually clusters: the semiconductor ecosystems of Taiwan, Japan, Korea, and mainland China; the machine-tool and optics traditions of Germany. These are not collections of founder miracles. They are the output of national and cultural imprints built over centuries — Humboldt’s research universities, the guild and apprenticeship system, a tolerance for slow accumulation of tacit skill. Zeiss dipped a ladle into that reservoir. The paper attributes the reservoir’s capacity to two men. The right unit of analysis for “why does this organization tolerate zero defects” is not the founder; it is the society that produced both the founder and the workforce that would accept his standards.

I want to name this plainly, because it is the paper’s central defect and it is not a small one: this is an attribution error. The archival work is excellent and the persistence result is real, but the causal weight has been assigned to the two men who are visible in the archive rather than to the conditions that made two such men possible and gave them a workforce willing to hold a tolerance nobody was checking. Founders are legible — they leave letters, lecture notes, statutes. Centuries of accumulated craft norms leave nothing an archive can file. Assigning causation to the part of the system that generated documents is a systematic bias, not an oversight, and it is how a study of a national manufacturing tradition ends up as a biography.

3. Crosina, Pratt & Lifshitz (2024): identity games in the co-working space

A Part of, or Apart from, Me? Linking Dynamic Founder-Venture Identity Relationships to New Venture Strategy. Administrative Science Quarterly.

What it says. Founder identification with the venture has two dimensions: intensity (how thick the rope) and construal level or psychological distance (how long the rope). Two paths follow. Founders who are psychologically very close (“I am the company”) experience setbacks as identity threats, respond by abstracting and increasing distance, and end up diversifying into a portfolio of projects to spread ego risk. Founders who are more distant (“I am an entrepreneur”) experience setbacks as objective threats, respond by moving closer and treating the venture as a child to protect, and end up specializing.

What I think. This is the weakest paper of the six, and I want to be specific about why rather than merely dismissive.

Start with the rope. Identification is given two dimensions — intensity, “how thick the rope,” and psychological distance, “how long the rope” — and from that image the entire theory unrolls. But a rope is not a measurement, it is a feeling about a measurement. Nobody weighed a rope. Thickness and length are not independently observed; they are read back out of the same interview in which the founder described how they felt about their company, and then treated as two orthogonal axes because ropes happen to have two obvious properties. The metaphor is not illustrating a construct that was established elsewhere. It is the construct. A sensory intuition has been installed in the place where a scientific conclusion is supposed to sit, and once it is there it generates predictions — short rope plus setback yields diversification, long rope plus setback yields specialization — with a tidiness no actual founder population produces.

The theoretical foundation is thin, and the strategic question has been flattened into a psychological one.

Self-esteem is a hidden variable in every founder. It is not the transfer function of strategy. The deep-tech founders I have backed are, almost without exception, results-driven to the point of coldness; they diversify or specialize because a customer signed or did not, because a process hit yield or did not, because the runway supports one product or two. Attributing scope decisions to identity-threat regulation reverses the order of magnitude of the forces involved.

The sample makes this worse. Twenty-nine founders, all from co-working spaces, without institutional venture funding and without full-time employees. These are individuals operating outside both constraints that make strategy hard: no capital with contractual claims on their decisions, and no physical technology that can fail. When such a founder takes on consulting work or invests in a friend’s restaurant, that is cash-flow survival. Coding it as “defensive diversification driven by identity threat” is teleological reconstruction — the theory decided what the data meant before the data arrived.

I do not doubt the interviews happened or that the founders said what they said. I doubt that what they said is the cause of what they did.

4. Kaplan (2008): framing contests in the fiber crash

Framing Contests: Strategy Making Under Uncertainty. Organization Science.

What it says. Under genuine uncertainty, data cannot speak for itself, so strategy-making becomes a contest between coalitions over which frame will govern interpretation. Frames are simultaneously cognitive constraints and political weapons. The study follows an optical-networking firm through the 2001 telecom collapse, documenting the tactics — data dumping, attacks on opponents’ legitimacy, and frame realignment through bridging, amplification, extension, and transformation.

What I think. This is the paper in the set with the most explanatory power, and it is the one I would hand to a first-time CTO.

Resources are finite, every functional head has a professional bias toward their own domain, and most of them are arguing in good faith about how the company survives. Framing contests under those conditions are not pathology; they are how a management team allocates a physical bus among competing devices.

The character who stays with me is Hugh, the technical leader. He was, by the paper’s account, right about the technology. He was also unable to operate. He presented a 238-slide deck to executives who needed a decision — an act of cognitive coercion, whatever its content. When marketing pointed out that carriers had no money to dig trenches, he could not translate his position into the commercial frame, walked out, and lost the decision window. I do not find his fate unjust. Science pursues truth about the physical world; a business creates value for shareholders within a bounded time and budget, and must solve this period’s survival problem before next decade’s technical one. Someone who cannot compile physical truth into the language of ROI and cash flow can be an excellent scientist and should not be a strategist.

That compilation step is, I think, the real skill Kaplan’s paper is pointing at without naming. The strongest technical strategists I have worked with advance their core technology through partial compromise — a stripped-down product to get a reference customer, a reverse merger to get a listing, an external validation to buy internal credibility. They accept that the hypervisor layer is where resources are allocated, and they learn its syntax.

5. Kirtley & O’Mahony (2023): the add/subtract state machine

What Is a Pivot? Explaining When and How Entrepreneurial Firms Decide to Make Strategic Change and Pivot. Strategic Management Journal.

What it says. A pivot is not a discrete, single-moment reversal. It accumulates through successive reconfigurations of activities, resources, and attention. Across seven clean-tech hardware ventures and 93 strategic decisions, 77% (72 of 93) were decisions to stay the course. Strategic exits (subtraction: dropping a product with no replacement) were triggered entirely by internal problems; strategic additions were overwhelmingly triggered by external opportunities. Genuine pivots typically passed through a strategic vacuum — a period after subtraction and before a replacement addition.

What I think. The finding is real and the myth it punctures deserves puncturing. Organizations are inert. Most pivots are forced by external pressure rather than chosen by insight. Silicon Valley’s iconography of the nimble, lean-startup pivot describes a small minority of cases.

But the sample is doing a lot of the work. The authors chose clean-energy hardware — tooling, long development cycles, high sunk costs — because it makes strategy observable. It also guarantees that strategy looks slow and cumulative. Move to a domain with near-zero marginal cost and code that can be rewritten overnight, and single-point, high-speed pivots are not a myth; I have watched them happen in AI application companies in the space of a quarter. The paper’s conclusion generalizes exactly as far as its physical inertia assumption holds.

I also object to some of the relabeling. The paper describes a prototype too large to fit in a travel suitcase, or a core metric a thousand times worse than a competitor’s, as “belief conflicts,” and claims on that basis to go beyond classical behavioral theory. Read as what they are — measurements — these are negative technical performance gaps, the most literal form of feedback a hardware company can receive. Renaming a measurement as a cognition does not make it one.

More broadly, this is a paper that spends years and considerable methodological care establishing something every operator knows: people and organizations resist change, and redirection is mostly beaten out of them by the outside world. I will return below to whether that is a criticism.

6. Hampel, Tracey & Weber (2020): geek community backlash and the romance of narrative

The Art of the Pivot: How New Ventures Manage Identification Relationships with Stakeholders as They Change Direction. Academy of Management Journal.

What it says. Ventures that grow out of a dedicated user community — here, the Impossible Project reviving Polaroid instant film — trigger feelings of betrayal when they turn toward the mass market. The community splits into attackers and doubters. The firm responds with “identification reset work”: eliciting sympathy from attackers by exposing its technical struggles and survival crisis, and mythologizing the Polaroid heritage and product obsession to reassure doubters.

What I think. The causality is inverted, and the inversion is the kind that a marketing training should immunize you against.

Customers resumed paying because the emulsion chemistry improved. Imaging yield went up, development time came down, and a second-generation film shipped that actually worked. That is objective physical value. The paper attributes the recovery to the emotional value of executives confessing struggle on social media. For an industrial consumer product, the ordering is not ambiguous: technology readiness drives adoption, and communications manage the noise around it. The paper has recorded a natural TRL progression and credited it to public relations.

The arithmetic tells the same story. Three and a half thousand enthusiasts cannot carry the fixed cost of a large legacy film plant in any financial model. Moving to the mass market was not an identity choice; it was survival by division. Managing the old community was reputational hemostasis — preventing a small group of opinion leaders from poisoning conversion rates in the larger market — not an “art of identification.”

There is a broader tendency here, and I will state it plainly because I expect to spend the next several years arguing with it from inside. A strand of Western management scholarship is in love with constructivism and discourse analysis. It treats the retrospective accounts of interviewed executives — accounts produced by people with every incentive to gild their own decisions — as the causal record of what happened. The result is a closed narrative loop: the story explains the story.

Stacking them up

Two of the six earn their place on the shelf. Kaplan (2008) is the most useful thing I read: it describes, without romance, how a management team under uncertainty actually allocates a scarce bus among competing devices, and it is honest that the winner of that contest is decided politically rather than empirically. Marquis and Tilcsik (2013) is the right map at the right altitude — sensitive periods, sedimentation, exaptation — even though exaptation needs counterfactual discipline it does not get, and even though the individual-level evidence cannot separate imprinting from selection. Kirtley and O’Mahony (2023) sits between: the finding is real and the agility myth deserved puncturing, but the sample’s physical inertia is doing work the theory then claims as its own.

The other three are where I part company. Becker (2025) is the best-executed and the most wrongly attributed — a study of a national manufacturing tradition delivered as a biography of two men. Hampel et al. (2020) has the causality inverted: the emulsion chemistry improved, and the recovery was credited to the confessions posted while it improved. Crosina et al. (2024) reasons from a metaphor to a mechanism on a sample from which capital and physics have both been removed.

Four ways to write a paper that says very little

Reading them together, the failures are not six different mistakes. They are four, and they recur.

Choosing an angle that excludes the forces that matter. Study founders who have no institutional capital and no physical technology, and strategy will look like psychology, because the two things that normally dominate strategy have been sampled out. Study only slow hardware ventures, and pivots will look cumulative, because the domain guarantees inertia. In both cases the setting was picked to make the phenomenon observable, and the setting then quietly supplied the finding.

Reasoning backwards from the conclusion. The theory arrives first; the data is coded to fit. A founder taking consulting work to make rent is recorded as defensive diversification under identity threat. Executives posting about their difficulties during a two-year manufacturing improvement are recorded as performing identification reset work. Nobody fabricated anything. The coding scheme simply already knew what it was looking for, and interview subjects with every incentive to gild their own decisions were happy to help.

Manufacturing concepts. A “task blueprint” is a technological paradigm plus an operating routine, which evolutionary economics has had for forty years. A prototype that does not fit in a suitcase, or a metric a thousand times worse than a competitor’s, is a performance gap, not a “belief conflict.” Renaming a measurement as a cognition is not a contribution; it is a claim to territory already occupied, and it makes the literature harder to read for everyone who comes after.

Crediting the visible actor. Causation gets assigned to whatever part of the system generated documents. Founders write letters and statutes; centuries of craft norms and an education system leave nothing an archive can file. So the ladle is credited with the reservoir, and it happens systematically rather than occasionally.

None of these are the sins of careless individuals — the archival work in Becker is meticulous, and the interviews in Crosina were plainly done with care. They are the predictable output of doing science from inside a virtual machine while declining to say so. If you cannot read the host, and you will not admit you cannot read the host, you will reach for whatever is legible: the metaphor, the transcript, the founder, the new label. Each of those is a way of producing something publishable in place of something true, and I do not think the resulting papers are worth the years that went into them.

What, then, is business research for?

I have been hard on most of these papers. The natural next question — one I need to answer honestly before I spend years producing work of the same kind — is what management scholarship is for, if the outcomes that matter are determined at bare metal and the exceptional operators who navigate them appear to be born rather than taught.

I have four answers, and none of them is “to discover how to be a genius.”

1. From individual magic to fault-tolerant protocol. Exceptional founders find their way through fog by something that looks like animal instinct. A civilization cannot run on the assumption that such people will appear on schedule. The value of management scholarship is to extract regularities from what they do and compile them into standardized protocols — routines, structures, contracts — so that systems composed of ordinary people can coordinate at scale without collapsing. Nobody needs the protocol to reproduce the genius. They need it to survive his absence.

2. Variance reduction on the normal distribution. Venture capital earns its returns from the extreme tail of a power law: the 0.1% of founders who produce most of the value. But 99.9% of firms are not in the tail, and those firms are where employment and capital preservation actually live. The right ambition for management research is not “how do we produce thousand-fold returns” but “how do we reduce the probability that an ordinary firm goes bankrupt or steps on a mine by twenty percent.” That is a modest goal and a large one.

3. Forensic audit of triumphant myth. The business world is saturated with post-hoc rationalization by winners. Rigorous empirical work — counterfactuals, control groups, longitudinal data — exists precisely to strip the gold leaf off success stories and stop entire societies from being marched into pseudo-science by bestsellers. Kirtley & O’Mahony proving that firms are inert is not trivial because everyone already knows it; it matters because a large industry is built on pretending otherwise. Common knowledge that has been measured is different from common knowledge that has not.

4. Guardians of the virtual machine. Management research will not make anyone a creator of worlds. Its practitioners are, at their best, janitors and protocol architects wearing glasses: people whose job is to keep reminding everyone writing code in the guest OS that there is a register temperature and a power budget and a physical limit underneath, and that the interrupt, when it comes, will not be maskable.

That last role is the one I intend to play. I am entering this field not to learn its narratives but to bring the interrupt boundary into the room — the cash date, the yield curve, the customer who did or did not sign — and see how much of the existing theory survives contact with it.

References

  • Becker, M. C. (2025). Mechanisms of organizational imprinting: From entrepreneur to organization. Administrative Science Quarterly, 70(1), 119–156.
  • Crosina, E., Pratt, M. G., & Lifshitz, H. (2024). A part of, or apart from, me? Linking dynamic founder–venture identity relationships to new venture strategy. Administrative Science Quarterly. Advance online publication.
  • Hampel, C. E., Tracey, P., & Weber, K. (2020). The art of the pivot: How new ventures manage identification relationships with stakeholders as they change direction. Academy of Management Journal, 63(2), 440–471.
  • Kaplan, S. (2008). Framing contests: Strategy making under uncertainty. Organization Science, 19(5), 729–752.
  • Kirtley, J., & O’Mahony, S. (2023). What is a pivot? Explaining when and how entrepreneurial firms decide to make strategic change and pivot. Strategic Management Journal, 44(1), 197–230.
  • Marquis, C., & Tilcsik, A. (2013). Imprinting: Toward a multilevel theory. Academy of Management Annals, 7(1), 195–245.