Drug discovery gets the headlines. Four other functions are saving money today.
Ask a pharma leadership team where AI is creating value and most will point at drug discovery. Ask their finance team where the savings actually landed this year, and the answer is somewhere far less glamorous. We compared four day-to-day functions — drug safety reporting, medical writing, clinical trials, and manufacturing — against drug discovery, to see which ones are saving money now and which are still a bet.
01 The baseline
What AI in drug discovery has actually delivered so far
Discovery is the benchmark every other AI claim in pharma gets measured against, so start with its scorecard.
AI-discovered molecules: trial and approval rate
| Label | Value |
|---|---|
| Clear Phase 1 | 85% |
| Clear Phase 2 | 40% |
| FDA-approved to date | 0% |
- 80-90% Phase 1 clearance. Statistically in line with the industry average — not the step-change the discovery narrative implies.
- ~40% Phase 2 clearance. Also roughly at baseline. The harder efficacy bar hasn't moved yet.
- Zero approvals, early 2026. The verdict on discovery AI is still years away, settled by pending Phase 3 readouts.
02 Function 1
Drug safety reporting: the largest proven saving
Pharmacovigilance is high-volume, rules-heavy and expensive — exactly what AI is good at, and the function with the least guesswork in its return.
Drug safety reporting: how much can be automated
| Label | Value |
|---|---|
| Case reports that can be automated | 85% |
| Faster detection of safety signals | 60% |
- $200-500M spent on drug safety a year. Typical for a major pharma company — the base the saving is calculated against.
- $80-300M estimated saving a year. What automating 70-85% of case reports implies, per company.
- Regulators have started to set rules. The CIOMS Working Group XIV published a 2025 consensus on AI in drug safety, so the compliance question is no longer open-ended.
Medical writing and regulatory drafting work the same way: high-volume, document-heavy, and reviewed by a person before anything ships. That is the safest use of generative AI in a regulated setting, which is why the savings there show up as hours saved and shorter turnaround within a few quarters rather than years — even though the evidence base is thinner than in drug safety.
03 Function 2
Clinical trials: real savings, if your data is reliable
Spending here triples by 2030. What holds teams back is not the AI — it is incompatible data, bias, and trust.
AI-in-clinical-trials market size
| Label | Value |
|---|---|
| 2025 | 2.7$B |
| 2030 (est.) | 8.5$B |
- 60-70% of trials will use AI by 2030. Up from a small minority today.
- $20-30B a year in estimated savings by 2030. Industry-wide, across patient recruitment, monitoring, and analysis.
- Patient recruitment costs down up to 70%. And recruitment timelines down up to 40% — the two figures moving fastest.
04 Function 3
Manufacturing and supply chain: proven, but barely deployed
Half the work on the floor could be automated with AI. Fewer than one in five companies actually run it at scale.
Manufacturing AI: what could be automated vs. what actually is
| Label | Value |
|---|---|
| Work that could be automated | 50% |
| Companies running AI at scale (2025) | 19% |
- Up to 50% of the work could be automated. McKinsey's estimate across demand forecasting, inventory, and supply planning.
- Only 19% run AI at scale today. The quickest wins — digital twins and predictive maintenance — are proven but barely deployed.
- That 31-point gap is the opportunity. Every point of it is money still sitting on the table.
05 Side by side
How the four functions compare on payback time
Day-to-day functions pay back in 12 to 24 months. Drug discovery takes years. Both numbers matter when you split a budget.
$25.7B
pharma AI market by 2030
McKinsey, up from ~$4B today
0
fully AI-discovered drugs approved
As of early 2026
12-24 mo
payback time, the four functions above
Multi-year
payback time, drug discovery
Fix the data first, then scale the pilots
If these functions pay back within 12 to 24 months, why do so many pilots still stall?
Because what blocks them is rarely the AI model. It is the state of the data underneath it and the controls around it. Connecting scattered data sources and finding the right people have overtaken executive backing as the hardest part, and under GxP rules every AI output has to be validated, auditable, and explainable. A pilot that cannot show real users and hours saved within two quarters is stuck — and fixing it means fixing the data, not swapping the model.
We help manufacturers capture the savings that are already proven: the drug safety case load, the manufacturing line, the regulatory documents — built on connected, well-governed data so a pilot can grow past the demo. QB SYSTEMS® handles the lab and bioprocess data layer where much of that groundwork starts. Drug discovery will play out on its own timeline. The four functions above pay back now.
Methodology & sources
Figures above are drawn from McKinsey (“How pharma is rewriting the AI playbook”), BCG/Wellcome (Jayatunga et al., Drug Discovery Today, 2024), Clinical Leader/IQVIA, and the CIOMS Working Group XIV 2025 report on AI in drug safety monitoring. Cost and efficiency figures are third-party or vendor estimates, not measured results from any single company — treat the ranges as a direction of travel, not a benchmark to hold a specific program against.