Drug Discovery Is Running a New Clock

Will Drewes

Founder, Fern Strategy · 3 min read

Drug Discovery Is Running a New Clock

The standard drug discovery timeline is 4-5 years from target to Phase I. That number is so baked into pharma planning that it shapes headcount, capital allocation, and partnership structures. A few recent cases suggest the assumption deserves a hard look.

Sumitomo Pharma: 12 Months to Phase I

DSP-1181, an OCD drug candidate developed with Exscientia, moved from target identification to Phase I entry in roughly 12 months. That is approximately one-quarter of the conventional timeline. The molecule was AI-designed, and the preclinical work that typically consumes years was compressed into a single calendar year.

This is not a rounding error. A 4x compression in discovery time changes the math on portfolio size, capital deployment, and how many shots a mid-size pharma can take in a decade.

Insilico Medicine: Under $2.6M to Phase II

INS018_055, targeting idiopathic pulmonary fibrosis, went from target discovery to preclinical candidate selection in about 18 months. Total discovery cost: under $2.6 million. It reached Phase IIa, where results showed a 98 mL improvement in forced vital capacity.

The cost figure is the one worth sitting with. Traditional discovery campaigns routinely run into the tens of millions before a candidate even reaches the clinic. At $2.6M, the break-even calculus on early-stage bets shifts considerably.

Phase I Success Rates: 80-90% vs. the 40% Baseline

As of December 2023, 21 AI-discovered drug molecules had completed Phase I trials. Their success rate: 80-90%. The historic industry average for traditionally discovered compounds sits around 40%.

That gap matters because Phase I attrition is expensive and demoralizing. If AI-designed molecules are genuinely clearing Phase I at twice the rate, the downstream economics of the entire pipeline change, not just the discovery budget.

An Early Clinical Signal Worth Watching

A June 2025 trial of an AI-designed cancer molecule added to standard hormonal therapy reported 81% tumor reduction across 31 participants with measurable disease. The sample is small. One trial is not a trend. But 81% in 31 patients is the kind of signal that earns a larger study, and the fact that the molecule itself was AI-designed is part of the story.

What This Means for Pharma Operations Leaders

These outcomes share a structural feature: AI is handling the parts of discovery where the rules are stable but the inputs are messy. Protein interactions, molecular property prediction, target validation. The reasoning is hard; the rules do not move much. That is exactly where agents earn their keep.

The governance question is not whether to use these tools. It is whether you can audit what they produced. Regulatory submissions require traceability. If your discovery pipeline runs on a black box, you will hit that wall at IND filing, not before. The teams seeing these results are not just moving fast. They are building in the documentation layer from the start, because the FDA will ask.

Before committing to a platform or building internal infrastructure, the two-week diagnostic question is simple: can you reconstruct every decision the agent made, and would that reconstruction satisfy a reviewer? If the answer is unclear, that is where to start.