The development sector's obsession with impact evaluation has produced an uncomfortable blind spot. We have become remarkably sophisticated at measuring whether programs work while remaining surprisingly naive about why so many never get the chance to demonstrate their potential. The evidence is sobering: a substantial proportion of development interventions collapse not because their core theory was wrong, but because critical failures occurred in the months before the first beneficiary was ever reached.

These pre-implementation failures represent a systematic waste of resources and opportunity. They are not random misfortunes but predictable consequences of institutional pressures, methodological shortcuts, and a peculiar form of organizational amnesia. Program after program repeats the same foundational errors—weak causal logic, context-blind design transfers, and compressed timelines that eliminate essential learning phases. The pattern is so consistent that it constitutes a structural feature of the development enterprise rather than a series of isolated mistakes.

Understanding these failure modes requires examining the political economy of program design itself. Donor incentives, organizational pressures, and professional norms create systematic biases toward certain types of errors. The evaluation literature has documented these patterns extensively, yet the institutional structures that generate them remain largely unchanged. What follows is an analysis of three critical pre-implementation failure mechanisms—theory of change weaknesses, context blindness, and timeline compression—each of which can doom an intervention before it begins.

A theory of change is supposed to be the logical backbone of any development intervention—a clear articulation of how inputs translate to activities, outputs, outcomes, and ultimately impact. In practice, most theories of change are exercises in wishful thinking dressed in the language of rigor. They contain causal leaps that would never survive serious interrogation, assumptions that have never been tested, and mechanisms that remain entirely unspecified. The document exists, the boxes are connected by arrows, but the underlying logic is hollow.

The most common failure pattern involves what might be called assumption stacking—the implicit requirement that multiple untested propositions must simultaneously hold true for the intervention to work. A conditional cash transfer program, for instance, might assume that households will prioritize children's education over immediate consumption, that schools will have capacity to absorb increased enrollment, that teachers will be present and motivated, and that labor markets will eventually reward the additional human capital. Each assumption carries uncertainty; their multiplication produces interventions balanced on increasingly improbable foundations.

Evaluations of failed programs consistently reveal these logical gaps. J-PAL's systematic reviews have documented cases where the fundamental mechanism of change—the core pathway through which an intervention was supposed to operate—was either never clearly specified or contradicted by basic behavioral economics. Microcredit programs assumed entrepreneurial capacity that did not exist. Information campaigns assumed information was the binding constraint when it was not. Training programs assumed skills were the bottleneck when the actual constraint was access to capital or markets.

The institutional incentives driving weak theory development are powerful. Program designers face pressure to appear innovative while remaining within established paradigms. They must satisfy multiple stakeholders with different priorities, leading to theories of change that are politically negotiated rather than logically derived. The documents become artifacts of organizational compromise rather than genuine analytical tools. Reviewers, often lacking subject matter expertise or time, approve frameworks that would not withstand basic causal scrutiny.

Rigorous theory of change development requires something that development organizations rarely permit: genuine uncertainty acknowledgment and iterative testing of core assumptions. It demands that program designers specify not just what they hope will happen, but precisely why they expect it to happen, what evidence supports each causal link, and what would constitute a falsifying observation. This level of intellectual honesty is organizationally costly and professionally risky, which explains its rarity.

Takeaway

Before accepting any theory of change, identify its three most critical untested assumptions and ask what evidence would be required to validate each one—if this evidence does not exist or cannot be gathered before scale-up, the program is a gamble rather than an investment.

The development sector operates with an implicit theory of context that is demonstrably false: the belief that interventions successful in one setting can be transferred to another with modest adaptations. This assumption underlies the entire architecture of evidence-based development—the systematic reviews, the 'what works' databases, the scaling frameworks. Yet the empirical record shows that context sensitivity is not a secondary consideration to be addressed during implementation; it is a primary determinant of whether an intervention can work at all.

External validity failures are not exceptions but the norm. Lant Pritchett and Justin Sandefur's analysis of development program replication found that effect sizes vary enormously across contexts, often changing sign entirely. A deworming program that produced substantial educational gains in Kenya showed minimal effects when replicated in India. Conditional cash transfers that transformed schooling outcomes in Mexico produced different results in Sub-Saharan African contexts. The heterogeneity is not random noise—it reflects genuine differences in institutional environments, social structures, and economic conditions that fundamentally alter how interventions operate.

The mechanisms of context blindness are multiple. Program designers often lack deep knowledge of implementation settings and rely on surface-level assessments. Donor organizations, seeking to demonstrate learning and replication, create pressure to export 'proven' models rather than invest in context-specific design. Professional networks circulate success stories that emphasize replicable elements while downplaying the contextual factors that may have been decisive. The result is systematic overconfidence in transferability and systematic underinvestment in local adaptation.

Consider the institutional prerequisites that successful programs often take for granted. A health intervention may assume functional referral systems that do not exist. An agricultural program may assume land tenure security that is absent. A financial inclusion initiative may assume regulatory frameworks that have not been developed. These institutional foundations are rarely explicit in program documents but are absolutely essential for the causal mechanisms to operate. Importing the visible components of an intervention while ignoring its invisible institutional supports is a recipe for failure.

Genuine context adaptation requires investment that development timelines and budgets rarely accommodate: extended formative research, pilot iterations, and genuine engagement with local implementation partners. It requires acknowledging that the 'model' from elsewhere provides at best a starting hypothesis rather than a blueprint. Most fundamentally, it requires organizational humility—the recognition that expertise in program design does not automatically translate to expertise in a specific context.

Takeaway

Treat every program design from another context as a hypothesis requiring local validation rather than a proven model awaiting implementation—the burden of proof for transferability should be on those claiming it, not those questioning it.

Development programs operate within funding cycles that are fundamentally misaligned with the time required to design, test, and refine effective interventions. The typical three-to-five-year project horizon, combined with pressure to demonstrate early results, creates systematic incentives to skip the iterative learning phases that determine long-term success. This timeline compression is not an implementation challenge to be managed but a structural feature that predictably undermines program effectiveness.

The evidence from successful interventions tells a consistent story about time requirements. BRAC's graduation approach, now widely replicated, emerged from over a decade of iterative development in Bangladesh before achieving its documented impacts. The Progresa/Oportunidades conditional cash transfer program in Mexico underwent years of design refinement, including extensive piloting and adjustment. These programs succeeded not despite lengthy development processes but because of them. The iteration allowed designers to identify and correct problems that would have been fatal at scale.

What gets eliminated when timelines compress is precisely what matters most: the learning loops that allow programs to adapt to implementation realities. Piloting becomes a formality rather than a genuine test. Monitoring systems are designed to report rather than to inform adaptation. Staff training is abbreviated. Community engagement processes are truncated. Each shortcut is individually defensible under time pressure; collectively, they hollow out the intervention's capacity to function.

The political economy of donor relationships reinforces these pressures. Program managers who request additional design time risk appearing uncommitted or incompetent. Donors who accept extended timelines risk criticism for slow disbursement. Success metrics emphasize outputs delivered on schedule rather than outcomes achieved through careful implementation. The incentive structure systematically rewards the appearance of action over the substance of effectiveness.

Breaking this trap requires fundamental changes in how development effectiveness is conceptualized and measured. It requires donors to accept that appropriate timeline investment is not inefficiency but a prerequisite for impact. It requires program managers to resist pressure for premature scale-up and to document learning value explicitly. Most importantly, it requires the sector to internalize that speed and effectiveness are often inversely related—that the fastest path to impact is frequently the one that allows adequate time for getting the intervention right.

Takeaway

Build explicit learning phases into program design with protected timelines and budgets—any funder unwilling to invest in adequate piloting and iteration is effectively asking you to waste resources on predictable failures.

The pre-implementation failures examined here are not mysteries. They are well-documented, repeatedly analyzed, and thoroughly understood. The development sector possesses extensive knowledge about why programs fail before they begin—weak theories of change, context blindness, and timeline compression appear in evaluation after evaluation. This knowledge has had remarkably little effect on practice because the institutional pressures generating these failures remain unchanged.

Addressing these systematic weaknesses requires more than methodological improvements. It demands restructuring incentives, timelines, and accountability frameworks across the development enterprise. Donors must accept that rigorous program design requires time and resources that conflict with pressure for visible action. Implementing organizations must develop the institutional capacity and political cover to push back against unrealistic expectations. The evaluation community must extend its scrutiny backward from impact assessment to design quality.

The alternative is continued waste—not just of financial resources, but of opportunities to improve lives. Every program that collapses due to predictable pre-implementation failures represents beneficiaries who might have been served by a better-designed intervention. The knowledge to prevent these failures exists. What remains is summoning the institutional will to apply it.