Pharma’s Billion Dollar Problem: Trials That Move At The Speed Of Paper

By Simon Arkell, Founder and CEO of Ryght AI

Medicine has entered the age of AI. The same way ChatGPT can generate hundreds of social posts in seconds, AI models are churning out drug candidates at record pace. Hundreds of startups and big pharma teams have invested in the tech and are generating novel molecules and targets at scale. Yet very few of these AI-created drugs have reached patients. The hold-up is not in discovery. It’s clinical trials, a slow, manual, paper-heavy and expensive process that has changed very little over the last 100 years.

Eight leading AI drug-discovery companies alone reported 31 drug candidates in human trials by 2024. That is real progress. But it hasn’t been translated into approved medicines yet. Even the most advanced cases, like Insilico Medicine’s AI-designed TNIK inhibitor for pulmonary fibrosis, are only now showing early Phase IIa signals in Nature Medicine. That’s promising, but still years from broad use.

Trials Move At The Speed Of Paperwork

Despite these possibilities, clinical trial startup remains painfully slow. Contracts, IRB reviews, budgets and site activation often drag on for months. Recent site data show median activation times near 8 months at academic centers, with many studies taking years to activate even at community sites. All of this to activate sites that were possibly chosen for the wrong reasons, and therefore may not even be successful once activated. Every delay pushes lifesaving therapies further out.

Once trials actually get up and running, much of the work is still manual: manually pre-screening patients that may be missed due to complexity and volume, missing the narrow enrollment window that may close before a patient even knows about a trial, reading through long, complex protocols to know how to treat a patient, source-data verification by hand and siloed logs for safety, shipping and compliance. The result: long cycle times, higher costs and staff burnout. Industry estimates still place end-to-end R&D costs anywhere from hundreds of millions to several billions per approved drug, even after accounting for failures. Average timelines of 12 years remain common.

Regulators Are Opening The Door, But Slowly

There is movement. The FDA has finalized guidance to allow decentralized elements, telehealth visits, at-home data collection and local labs, to cut friction for patients and sites. And the global ICH E6(R3) Good Clinical Practice update pushes risk-based quality and “build quality in”, not “inspect it in”. These are meaningful shifts, but they need execution at scale.

AI is pushing more and better shots on goal. If the trial system cannot absorb that volume, patients won’t see the benefits. Early readouts show some AI-designed drugs clear Phase I well, but Phase II and III are where time and money stack up. Without a faster trial machine, the discovery revolution stalls in development.

What To Implement Now With AI

While AI has accelerated drug discovery, the same technology can address clinical trial bottlenecks that have persisted for decades. Here are some of the ways leaders can start to implement AI:

Standardize startup and budgets. Use master CTAs and templates enhanced by AI clause libraries. Auto-generate site budgets from the schedule of events. Compress contract redlines so turnarounds are measured in hours, not weeks.

Build live “digital twin” replicas of research sites. Maintain continuously updated models that mimic site capabilities, investigators, past performance and data access. Use AI to match study protocols to the best-fit sites, whether for new studies or rescue scenarios, based on evidence, not guesswork.

Automate feasibility and site activation. Let AI read protocols, draft feasibility questions, contact sites, populate feasibility questionnaires on behalf of the sites, track responses and drive signatures. Humans approve exceptions; software handles the rest.

Make pre-screening automatic at the site. Use AI to surface eligible patients from human input, notes, registries and scheduling systems without heavy and slow EMR integrations. Generate auditable candidate lists with transparent inclusion/exclusion reasoning. At my last company, we integrated tightly with the EMR, and it was useful, but it took too long. Most vendors do only this and can therefore only offer a site network as large as the number of sites willing to use that software and provide that EMR access. This can take years to achieve at most AMCs. There is a better way.

Deliver protocol-adherence (visit planning) apps at the site level. Generate per-patient visit plans and automate procedure and labs checklists. Flag deviations in real time, propose fixes and keep coordinators on schedule.

Deploy predictive analytics. Use machine learning to identify at-risk patients, optimize site performance and detect safety signals earlier in the process.

The Bottom Line

AI has transformed the front end of drug R&D, making discovery faster and broader. Now it’s time to apply the same innovation to clinical operations. Until studies shred paper workflows and pre-digital habits, patients will wait unnecessarily for breakthrough treatments. The technology exists today to standardize setup, digitize data capture, decentralize where safe and manage quality through intelligent risk assessment.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?