When Better Isn’t Smarter: How Innovation Accelerators Fall into the Selection Trap
| May 2025
In the race to fund the next big innovation, accelerators often pride themselves on refining their selection processes – vetting smarter, filtering harder, and leveraging data with surgical precision. But what if these improvements are quietly backfiring? Dmitry Sharapov and Linus Dahlander’s new study, “Selection Regimes and Selection Errors,” published in Organization Science (2025), offers a sobering reality check. Through a rare combination of qualitative interviews and rigorous quantitative analysis, they uncover how even well-intentioned, evidence-backed improvements in selection regimes can paradoxically increase both false positives and false negatives in innovation funding.
The Setup: An Accelerator’s Evolution Through Three Selection Regimes
The researchers immersed themselves in the inner workings of an innovation accelerator – an institution charged with selecting high-potential projects for funding. Over several years, they conducted 126 interviews and compiled quantitative data on all 3,580 project submissions. As the accelerator evolved through three distinct selection regimes, the authors were able to map how internal changes shaped the quality and accuracy of decision-making.
Each regime aimed to reduce classic selection errors:
• False positives – funding projects that don’t succeed.
• False negatives – rejecting projects that would have succeeded.
The accelerator first emphasized expert judgment, then introduced structured evaluation rubrics, and finally moved to a track-record-based model, adding more layers of review. Each step was an effort to increase accuracy. Each step, paradoxically, introduced new kinds of failure.
Why Getting Stricter Didn’t Help – and Sometimes Hurt
In the final regime, the organization tried to boost the average quality of applicants by emphasizing prior success (applicant track record) and adding additional screening layers. This sounds rational: reward proven winners, reduce noise, and build a portfolio of reliable innovation bets. However, this very logic produced two unintended and counterproductive outcomes:
1. Mean Reversion Meets Track Record Bias
Evaluators became anchored to applicants’ past successes, overestimating their likelihood of future success. This “mean reversion” meant that projects led by high-performers often performed merely average in the new round, while projects led by underdogs were disproportionately rejected – even when they had breakout potential. The system was seduced by reputation, not results.
2. Within-Type Adverse Selection
The stringent filters caused the applicant pool to self-sort in unexpected ways. Those with flashy resumes but weaker ideas could navigate the filters more easily than unconventional thinkers with transformative ideas. The process unintentionally privileged polish over substance, causing the accelerator to make systematic judgment errors – choosing the wrong type within the right category.
Persistent Errors Despite Smart Intentions
Despite well-designed process improvements, the accelerator continued to experience persistent rates of both false positives and false negatives across all regimes. This wasn’t for lack of trying. The authors controlled for:
• Changes in the applicant pool
• Program design and support effects
• Organizational learning over time
• External market evolution
And yet the errors held firm. The key insight: no single lever of selection reform can substitute for holistic, adaptive judgment. Selection regimes must evolve, but not in isolation. Fine-tuning one area – like applicant evaluation – without corresponding adjustments in others, like feedback loops, cultural norms, or organizational incentives, can introduce blind spots that worsen outcomes.
Lessons for Innovation Managers and Policymakers
For leaders involved in innovation funding, venture capital, or R&D evaluation, this research offers a critical warning. Systematizing selection is not the same as improving it. Emphasizing track records, adding more layers, or increasing formality may look like progress, but these changes often miss the dynamic interplay between process, people, and judgment.
The lesson here is one of humility: innovation cannot be fully predicted, and attempts to engineer certainty into uncertain environments may invite new forms of error. Instead, organizations might benefit from:
• Encouraging diversity in evaluators and applicants alike.
• Creating feedback mechanisms that detect subtle forms of adverse selection.
• Rotating evaluators to reduce anchoring on past criteria.
• Designing for learning, not just filtering.
Conclusion: Selection Is a Moving Target
Sharapov and Dahlander’s study is a rare gem – methodologically robust and deeply relevant. It reminds us that in innovation, smarter is not always better, and tighter control can sometimes erode the very discretion and serendipity that drive breakthrough ideas. If we want to fund the future wisely, we must embrace a more dynamic, reflexive, and systemic approach to selection – one that balances rigor with imagination, structure with flexibility, and track record with vision.
Because when it comes to innovation, the best picks aren’t always the most obvious – and the biggest mistakes often wear the mask of expertise.
