Figure 1
Reported GenAI efficiency gains remain concentrated before translation
Distribution of 153 coded literature segments by development phase and efficiency type — systematic review of 100 peer-reviewed studies (2018–2025)
⚠ Segment counts reflect publication frequency, not measured or validated effect sizes
Time efficiency (reported)
Cost reduction (reported)
Discovery
Phase I
114 time  ·  25 cost
n = 139
Evidence maturity
Highest evidence density; predominantly in-silico / proof-of-concept; prospective biological validation limited
Preclinical
Phase II
2 time  ·  8 cost
n = 10
Evidence maturity
Sparse; largely model-based simulations; limited prospective validation or standardised methods
Clinical
Phase III
0 time  ·  4 cost
n = 4
Evidence maturity
Very sparse; no reported time savings; regulatory, recruitment and workflow constraints unresolved
Source: Adapted from Riemer & Freund (2026), “Generative artificial intelligence in pharmaceutical drug development,” Intelligent Pharmacy, doi:10.1016/j.ipha.2025.12.006. Bars are proportional to coded segment count per phase (n = 153 total segments across 100 studies). Segment counts were derived from qualitative content analysis (MAXQDA); they do not represent estimated effect sizes or commercial impact.