AI and digital twins on the factory floor: how drug makers are using them in 2026
Eli Lilly credits AI with producing more medicine than it otherwise could; agentic AI is halving deviation investigations. What is real, and what it needs.
In this article
“Pharma 4.0” was a theme of the 2025 International Pharma Convention at Pharma Asia. In 2026, some of the world’s biggest drug makers are showing what it looks like in practice.
Eli Lilly: more output from the same plants
Lilly uses AI to predict batch outcomes, tune process parameters and detect deviations, and builds digital twins — virtual copies of production lines — to simulate scale-up and find bottlenecks before changing equipment, Pharma Manufacturing reported in June 2026. In one case, a digital twin’s prediction was tested in a pilot and then implemented, sharply shortening a bottleneck step. Chief Information and Digital Officer Diogo Rau said: “We literally made more product last year than we possibly could have without AI.” In October 2025 Lilly also partnered with NVIDIA on a supercomputer intended partly for manufacturing digital twins.
Lilly’s Scot Lindsey stressed that explanations matter: “We don’t want any black box.”
Agentic AI in quality
In July 2026, Megha Sinha of Kolter AI told PharmTech that agentic AI systems in validated GxP environments can cut deviation investigation time by 50% or more — detecting out-of-specification results, pulling batch records and drafting root-cause analyses. She put the market for plant-level digital twins at more than $1 billion, while describing the infrastructure as “half built.”
Research tools reaching industry
Cambridge’s research centre in Singapore (CARES) and A*STAR’s Institute for Infocomm Research have developed an AI platform that builds digital twins from real-time plant data and flags anomalies such as abnormal tank levels before they escalate.
What it needs
| Foundation | Why |
|---|---|
| Clean, connected data | AI cannot learn from paper records |
| Sensors and PAT | Real-time measurements feed the models |
| Validated systems | GMP requires explainable, controlled software |
What a digital twin actually is
A digital twin is a computer model of a real process or piece of equipment that is kept up to date with data from the real thing. In a pharmaceutical plant it might represent a granulator, a tablet press, a whole production line or a utility such as a water system. Because the model is fed with live measurements, engineers can use it to ask “what if?” — what happens if we change the spray rate, run the press faster or add a shift — without risking a real batch.
Typical uses include:
- Scale-up and technology transfer — predicting how a process will behave on larger equipment.
- Debottlenecking — finding which step limits output, as in Lilly’s example.
- Predictive maintenance — spotting signs that a machine will fail before it does.
- Deviation investigation — reconstructing what happened when a batch goes wrong.
The regulatory side
In a GMP plant, software that affects product quality must be validated: the company has to show it does what it is meant to do, that its data are reliable and that changes are controlled. Established principles such as data integrity (records that are attributable, legible, contemporaneous, original and accurate) and the validation of computerised systems apply to AI as well. Regulators are now writing more specific guidance: the US FDA published a discussion paper on artificial intelligence in drug manufacturing in 2023, and European regulators have consulted on a new GMP annex on AI. The common message matches Lilly’s: no black boxes. A model used for quality decisions must be explainable, tested and kept under control.
Where to start
| Step | What it involves |
|---|---|
| 1. Digitise records | Electronic batch records and laboratory systems instead of paper |
| 2. Connect equipment | Collect data from machines and sensors in one place |
| 3. Analyse | Use the data for trending, alarms and root-cause analysis |
| 4. Model | Build models and twins for the processes that matter most |
Each step brings benefits on its own — fewer documentation errors, faster investigations, better maintenance planning — so plants do not need to wait for a full digital twin to see a return.
What it means for Pakistan
The first step for most plants is digitising records and instrumenting lines — work that also supports DRAP’s documentation requirements. Automation and software suppliers exhibit at Pharma Asia 2026.
Sources
- Pharma Manufacturing, “Eli Lilly taps AI, digital twin technology to boost manufacturing capacity”, 11 June 2026
- PharmTech, “Midyear 2026 check-in: agentic AI cuts deviation investigations by 50%”, 17 July 2026
- European Pharmaceutical Review, “AI digital twin developed to boost pharma manufacturing efficiency”
