I study how the criteria used to evaluate technological innovations
emerge, persist, and change.
I am on the 2026–2027 academic job market.
In many industries, what counts as a successful innovation is defined
by criteria that seem objective but are shaped by strategic choices.
When those criteria fall short of capturing the actual performance of
innovations in market deployment, who proposes new evaluation
criteria — and why are some firms able to do this while most are not?
My dissertation develops a theory of evaluative evolution:
how evaluation criteria of innovations change, which firms drive that
change, and what happens to the technological trajectories and
competitive dynamics of innovations following that change.
I study these questions in the pharmaceutical industry, where clinical
trial endpoints determine which therapies reach patients — and where
the wrong criteria can mean approved drugs that don't actually help.
I bring training in law, biostatistics, and strategic management,
which is why I see evaluation criteria as simultaneously technical
design choices and sites of strategic contestation. To study these
questions at scale, I build custom AI/LLM research pipelines —
including retrieval-augmented systems that outperform state-of-the-art
general-purpose models on domain-specific tasks. I have been
invited to present my research at major pharmaceutical firms.
Research Interests
Innovation Strategy
Evaluation of Innovations
Organizational Learning
Pharmaceutical Industry
Strategic Management of Intellectual Property
Methods
Causal Inference / Econometrics
Natural Language Processing
LLM/RAG Pipeline Design
Mathematical Modeling
Medical Concept Classification Systems
♦
Research
Dissertation: Evaluative Evolution
The dissertation consists of three interconnected studies.
The first asks which firms recognize and propose new criteria
when existing ones fail. The second examines how evaluators decide
which proposals to endorse — a problem structurally harder than
evaluating innovations under fixed criteria, because the very
dimensions being proposed are new. The third traces what happens
after endorsement: how new criteria redirect technological
trajectories, and why the firms most likely to adopt are not always
the ones who benefit most.
Because newly endorsed criteria are themselves satisficing — good enough
to guide decisions, but never fully capturing real-world
performance — evaluative evolution is recursive, and each new
criterion sows the seeds of its own eventual revision. These three
studies open a broader research program — including how evaluators
learn through the evaluative evolution process, the durability of
industry convergence around new criteria, the dynamics of
successive evaluative cycles, and the extension of evaluative
evolution to medical devices, financial regulation, and
environmental standards.
Job Market Paper
Who Proposes New Evaluation Criteria for Innovations? Firm Capabilities and the Second-Order Learning Problem
Yunxiang Bai
Criteria are heuristics — imperfect proxies for how innovations will
actually perform in the real world — and should evolve too. Yet firms
also need stable criteria for adaptive learning: to search for better
solutions, you must hold the definition of "better" constant. Who
then proposes new criteria, and why?
In many industries, nonmarket evaluators — regulators, standard-setting
bodies, certification agencies — establish the criteria by which
innovations are judged. These criteria are structurally fallible: no
measurable attributes can fully capture real-world performance. Yet
criteria typically persist despite their fallibility, because the same
learning mechanisms that make innovation-level optimization adaptive
actively prevent criteria-level questioning. I theorize that criteria
proposals require overcoming three structural barriers: criteria sit
outside the hypothesis space of normal search, failures are attributed
to innovations rather than to criteria, and adaptive search generates
evidence that confirms existing criteria while preventing evidence
that might challenge them. Visible criterion failures — such as FDA
withdrawal of traditionally approved drugs — open windows where
firms with specific capabilities can overcome these barriers. Using
quarterly firm × therapeutic area panels from 2008–2025, I find
that scientific knowledge, cross-domain breadth, and market
deployment experience each amplify novel endpoint proposals
following withdrawal shocks.
Working Papers
Vision or Delusion? How Evaluation Sequence Anchors the Assessment of Novelty
In preparation for submission to Strategic Management Journal
Organizations select against novel ventures even when they
explicitly seek novelty. The literature diagnoses this as a problem
of obscured vision — evaluators fail to see the upside. But
evaluators do score both upside and risk. This study argues that
the penalty arises not only from how they see each dimension, but
also from how they synthesize conflicting dimensions into an overall
judgment.
Evaluating novel ventures is challenging as evaluators struggle
to reconcile the upside potential of a venture with its execution
risks. While prior work focuses on how evaluation procedures shape
evaluators' relative attention to the opposing dimensions, we
examine the sequence of key evaluation criteria and explore how it
can shape the overall assessment of novel venture ideas. Using
proprietary data from a startup evaluation platform and two
pre-registered experiments, we found that when evaluators consider
upside potential before risk, their evaluation anchors on the
upside potential, thus favoring novelty. Conversely, when risks
are considered first, their evaluation anchors on risks, thus
disadvantaging. This paper contributes to research on idea
evaluation and entrepreneurship by highlighting the role of a key
structural component, the sequence of criteria, in evaluating
novel ideas.
Presented at SMS Annual Conference, Istanbul, 2024
Engaging in public science creates knowledge that rivals can freely
use — so does it ultimately help or hurt the publishing firm? The
literature has treated this as a single tradeoff, but tracing four
decades of knowledge flows reveals that the answer depends on a
temporal distinction the literature has not drawn.
Research on science and innovation highlights how firms' scientific
engagement shapes knowledge flows determining who captures returns
to innovation. Yet, whether science tilts these flows toward the
publishing firm or its rivals has not been directly tested. This
study abductively explores this question by tracing patent citation
flows for 170 biopharmaceutical firms over four decades. In contrast
with existing literature treating the appropriability implications
of science as a single tradeoff, this study reveals that the answer
depends on temporal perspective: under a retrospective lens, firms
sustaining ongoing science capture roughly twice the benefit rivals
do, while under a prospective lens, science at invention creates
contested opportunities whose firm advantage materializes only at
longer horizons. Exploratory evidence suggests science helps firms
retrieve knowledge from spillovers.
Mitigating Nonattendance Using Clinic-Resourced Incentives Can Be Mutually Beneficial: A Contingency Management-Inspired Partially Observable Markov Decision Process Model