The Drug Discovery Landscape: Economics, Failure Rates, and Strategic Frameworks

The Drug Discovery Landscape: Economics, Failure Rates, and Strategic Frameworks

A new molecular entity approved in 2024 cost, on average, between $1.5 billion and $2.6 billion to discover, develop, and register—depending on whose methodology you trust and whether you capitalize failure costs. The 2014 Tufts Center for the Study of Drug Development figure of $2.558 billion (in 2013 dollars) has become the industry's most cited number, but it is now more than a decade old and almost certainly an underestimate; Paul et al. and later revisions using the same framework peg the inflation-adjusted figure north of $2.8 billion when accounting for post-approval costs and the higher failure rates observed in the 2010s. Whatever the exact number, the trajectory is unambiguous: cost grows, success rates do not, and the structural drivers of this divergence are deep enough that no single intervention—better assays, AI, larger libraries, more biomarkers—has bent the curve. This opening chapter establishes the macroeconomic, scientific, and organizational context within which every subsequent decision in this book is made. We will dissect Eroom's Law, quantify attrition by phase and therapeutic area, introduce the Target-to-Patient pipeline that scaffolds the rest of the book, formalize the portfolio theory that any serious program manager must understand, and conclude with the four trade-offs that recur at every decision gate.

Eroom's Law, R&D Productivity Decay, and the Structural Drivers of Cost Inflation

The observation that pharmaceutical R&D productivity has been declining for at least seventy years is usually presented as a curiosity, but it is in fact the single most important fact about the industry. Eroom's Law—Moore's Law spelled backward, because the trend is the inverse of what we see in semiconductors—was formalized by Jack Scannell and colleagues in 2012 in *Nature Reviews Drug Discovery*, though the underlying pattern had been visible in industry data since the 1950s. The standard formulation states that the number of new drugs approved per billion US dollars of R&D spending has roughly halved every nine years since 1950, even after adjusting for inflation and improvements in purchasing power. By 2010, that ratio had fallen to approximately one-tenth of its mid-century value. The trend has not reversed in the subsequent decade, despite enormous investments in high-throughput screening, combinatorial chemistry, structure-based design, genomics, and machine learning.

Why does this happen? The naïve explanation—that regulators have grown arbitrarily more demanding—is wrong. The FDA approves the overwhelming majority of applications it receives for drugs that demonstrate a favorable benefit-risk profile; the bottleneck is upstream, in the pipeline itself, not at the agency. The deeper causes are structural, and they are worth enumerating carefully because they shape every chapter that follows.

The "low-hanging fruit" exhaustion effect. The first generations of modern pharmaceuticals were discovered by testing a small number of compounds against a small number of targets in a small number of diseases. Aspirin, the sulfonamides, the statins, the proton pump inhibitors, the SSRIs, and the first-generation tyrosine kinase inhibitors were all discovered against well-validated biology using assays that were good enough to find active compounds. As the easy wins were harvested, the industry moved to harder targets with less validated biology, harder chemistry with less tractable binding sites, and harder diseases with greater competitive pressure. The marginal return on each additional dollar of R&D has fallen because the *opportunity set* has become worse, not because the science has become less powerful.

The "better than the Beatles" problem. Once a therapeutic area has an effective standard of care, new entrants must demonstrate superiority or non-inferiority on increasingly demanding endpoints. Depression is the canonical example: a new SSRI in 2024 must beat or match escitalopram, which is generic, off-patent, well-tolerated, and clinically effective for 60-70% of patients. The bar is set by the best available therapy, not by placebo, and the bar rises every year as generics of previous winners enter the market. This phenomenon is most acute in cardiovascular disease, diabetes, and psychiatry, and least acute in rare genetic disorders and emerging infectious diseases.

The "throw money at it" fallacy. Because capital is fungible and the scientific problems are real, the industry has responded to declining productivity by increasing spending. The top twenty pharmaceutical companies collectively spent approximately $140 billion on R&D in 2022, up from approximately $80 billion in 2010 (inflation-adjusted). The expected output—new molecular entities approved—has hovered around 40-50 per year for the entire industry for two decades. Marginal dollars have not produced marginal approvals because the binding constraint is not effort but opportunity: there simply are not enough well-validated, druggable, commercially interesting targets to feed the pipeline at the rate the industry would like. The growth in R&D budgets has therefore been absorbed by the increasing cost of clinical trials (larger, longer, more globally distributed), the increasing cost of failure (more elaborate preclinical packages, more safety studies, more biomarkers), and the increasing cost of success (longer development times for high-need indications, more combination studies, more pediatric commitments).

The regulatory ratchet. Regulatory standards have, in fact, tightened—but in a specific way that interacts with the other factors. Phase III trials have grown larger, longer, and more expensive. Cardiovascular outcome trials for diabetes drugs, required after the 2008 rosiglitazone controversy, added hundreds of millions of dollars to each development program. Pediatric study commitments, REMS programs, and post-marketing requirements have shifted cost from pre-approval to post-approval, but the total cost to the sponsor has risen. The shift toward orphan drugs and accelerated approval pathways has partially offset this, but the offset is incomplete: orphan trials are smaller but still expensive per patient, and the post-approval confirmatory studies required for accelerated approvals are themselves major development efforts.

The capital cost of failure. The Tufts methodology, which is the source of the $2.6 billion figure, includes the cost of capital—the opportunity cost of money tied up in R&D that might have been invested elsewhere. Because clinical development now takes 10-15 years, and because the probability of success for a Phase I entry is approximately 10%, the expected capital cost of any individual program is enormous. A program that fails in Phase III after spending $500 million has cost the company not just $500 million in direct outlay but the time value of that money, which at a 10-12% cost of capital adds another $200-400 million in capitalized cost. The capital cost of failure is the largest single contributor to the headline R&D figure, and it grows automatically as development times lengthen, even if direct costs were held constant.

Taken together, these five drivers explain why Eroom's Law has been so resistant to disruption. None of them is a single point of failure that a new technology can fix. The "AI will solve drug discovery" narrative, whatever its long-term merits, cannot reverse the fact that the opportunity set is finite, the bar is rising, and the capital cost of failure is built into the development timeline. What AI can do—and Chapters 9, 10, and 14 will explore this in detail—is shift the productivity curve *at the margin* by reducing the cost and time of certain discovery activities, particularly hit identification, lead optimization, and ADMET prediction. Whether the marginal improvements compound to a structural bre