The model for clinical development is structurally incompatible with the direction of the industry.

Consider that 53% of current phase 2 and 3 trials are delayed, according to a recent Bain & Company survey of 120 sponsors, contract research organizations (CROs), and trial site professionals. Most say trials are more complex than they were three years ago. And 76% expect complexity to increase over the next three years, whereas only 2% foresee relief.

%

of current phase 2 and 3 trials are delayed

Despite a decade of investment, the bottlenecks of patient identification, recruitment, and site activation are stubborn. AI is shortening discovery timelines, but without a redesigned development model, those gains are evaporating.

A fundamentally different model is possible today, but it takes more than using AI to optimize trials around the margins. Leading innovators will build a connected evidence-generation system and reimagine clinical development from first principles: the science, the patients, and the evidence needed for regulatory and commercial success.

The cost of inaction is high. If a competitor moves from early signal to registrational-quality evidence by compressing or replacing a conventional intermediate trial, latecomers will find the market already claimed. Formulary position, physician habit, and payer relationships will be locked in for the life of the asset.

The question, then, is who will set the benchmark and who will scramble to meet it.

Why the current system isn’t sustainable

Large pharma companies’ development infrastructure simply wasn’t built for today’s landscape. Several forces are stressing the model beyond its limits.

• The clinical pipeline is expanding, and competition is intensifying. The number of drugs in phase 2 and 3 is up roughly 9% year over year. Agile biotechs are running more phase 3 trials. And rapid early-stage trials in China are compressing the time from proof of concept to competitive threat. Speed is of the essence: The market can absorb only so many winning franchises, and after first-in, value drops sharply for subsequent followers, especially those without material improvements and heavily resourced commercial strategies.
• The gap between trial demand and eligible patients is getting wider. The continued rise of biomarker-driven and targeted therapies is narrowing eligible populations, even as expectations for trial diversity grow. Sponsors are simultaneously shifting enrollment outside the US for savings, patient access, and faster site activation and recruitment. Globally, regulators are increasing requirements for locally representative populations and region-specific evidence. Bain’s survey shows that average US enrollment in clinical trials has fallen about 28% over the past decade, with 45% of sponsors and CROs expecting further decline.
• The evidence-generation system is fragmented. Endpoint, comparator, and patient population choices often determine differentiation, reimbursement, physician adoption, and commercial success years before launch. Yet most organizations treat trial design as a regulatory exercise rather than a strategic imperative. Teams operate in silos, and insights gained in one stage rarely inform the next. Preclinical findings don’t consistently shape trial design. Commercial and market-access perspectives arrive too late to matter. The fragmentation costs time and lowers the probability of success.

%

decline in average US enrollment in clinical trials over the past decade

%

of sponsors and CROs expect further decline

What a different model looks like

Clinical development is ripe for scaled AI adoption, given its rich structured data, repeatable processes, and direct P&L benefit from even modest cycle-time reductions. But achieving real results requires moving beyond disconnected AI tools to specialized agents—across every development, regulatory, and commercial function—working from a shared evidence base. Agents exchange information, challenge assumptions, identify inconsistencies, and escalate consequential decisions to accountable experts.

Imagine how AI can help replace today’s trial-by-trial framework, connecting five activities into an AI-fluent, end-to-end R&D system:

1. Choosing the right asset, indication, dose, and patient population. A connected evidence-generation system draws on preclinical, clinical, real-world, and competitive data to prioritize the assets and indications with the greatest probability of success. AI-enabled modeling integrates pharmacokinetic and pharmacodynamic (PK/PD), exposure-response, efficacy, and safety data to optimize dose and regimen. AI also redefines how sponsors identify patients, combining genomics, imaging, digital measures, clinical history, and real-world data into multidimensional responder profiles. Prospectively validated, those profiles increase treatment-effect size, reduce unnecessary patient exposure, and enable smaller, faster trials.
2. Designing an integrated clinical, regulatory, and market-access evidence strategy. AI agents synthesize evidence as it arrives and reassess proposed endpoints, comparators, treatment settings, follow-up periods, patient-reported outcomes, and health-economic measures. They integrate the evolving expectations of regulators, health technology assessment (HTA) bodies, payers, clinicians, and patients across major markets. The result is fewer downstream amendments and protection against a trial that supports approval but doesn’t generate the comparative or economic evidence required for broad reimbursement and adoption.
3. Ensuring chemistry, manufacturing, and controls (CMC), bioanalytical, and supply readiness. AI surfaces manufacturing or analytical constraints before they result in delay. For complex modalities, CMC and bioanalytical readiness may increasingly determine the development path, rather than enrollment alone.
4. Executing an adaptive, patient- and site-centered trial. AI-powered site selection and patient matching compress time to first patient enrolled. Prespecified adaptive designs reduce waste by allowing prospectively defined modifications based on interim evidence while preserving statistical validity. External or synthetic control arms supplement or reduce the size of conventional control groups, with regulatory agreement. Patient-level simulation and digital-twin approaches help test trial assumptions, forecast clinical trajectories, and identify hypotheses for prospective validation. Instead of waiting for final analysis, continuous monitoring across adverse event and laboratory databases surfaces risks early enough to intervene and protect timelines. AI-generated site queries run in the background, accelerating database lock.
5. Turning evidence into decisions, submissions, and portfolio learning. Decision cadence shifts from milestone-based to event-based. Agents keep the evidence package current and, when a predefined threshold is crossed, assemble the supporting analysis and call for a decision meeting. Senior leaders attend review meetings only when there’s real evidence to act on. And every trial event, from a protocol amendment to a manufacturing deviation, updates the models for subsequent trials. This converts institutional knowledge into a compounding competitive advantage, instead of leaving it trapped within functions and documents.

Innovative companies are already pushing the boundaries of trial design. The US Department of Health and Human Services’ Operation TrialBlazer is supporting the shift, with draft guidance from the FDA on quantitative systems pharmacology (QSP)-based dose selection for first-in-human trials and updated master protocol frameworks for more adaptive designs.

AI also broadens access to these approaches, enabling smaller competitors to generate models and simulations that once required large functional teams and resources. Formation Bio, an AI-native biotech, built clinical development around an integrated platform. It autonomously generates and evaluates thousands of trial design scenarios, drafts study builds in minutes, and surfaces critical signals as trials run, allowing Formation Bio to go from deal close to first participant dosed in a phase 1 trial in five months.

Related brief

Clinical Trials Face Persistent Challenges—and New Opportunities

Bain’s survey of 120 sponsors, CROs, and trial site professionals shows persistent pressure on recruitment and site capacity, but also an opportunity to redesign work around AI.

How to redesign clinical development

Few organizations are moving fast enough. That’s because adopting a different model requires a level of internal risk tolerance and willingness to change that most haven’t built yet.

Nearly half of sponsors and CROs are piloting AI in select functions, but only 16% are redesigning roles and workflows around it. Piloting without restructuring generates local efficiency gains, not a more efficient organization.

First movers are transforming core processes with leadership sponsorship, investment, and sustained focus. They’re taking three steps to implement a continuously learning, AI-fluent R&D system.

Build a connected evidence-generation system

Decades of proprietary research and clinical trial data at large biopharma companies are underutilized. When organized, connected, and accessible to AI, institutional knowledge becomes a durable competitive asset.

Start with bounded projects that can later connect: AI-modeled QSP-based dose selection, protocols informed by regulatory and market-access needs, or real-time data cleaning.

Then form the connective tissue. The system requires cross-functional teams responsible for evidence quality and consistent data standards end to end; governance with decision gates; and interoperable platforms that make evidence accessible and analyzable. It also requires centralized owners of data standards, governance, and privacy, as well as AI and R&D experts who build multistep agents.

Redesign clinical development from first principles

A redesigned model not only improves sponsor speed and cost but also reduces patient and site burden. Trials that are easier on patients and investigators alike can recruit faster and generate more complete, representative data. The most powerful reframe is the blank page: What evidence do we need to make a confident decision? What information types, how many patients, over what time frame, and judged against whose expectations?

Leaders work backward from this integrated foundation, making faster, evidence-based investment decisions. AI-enabled biological models, human-relevant experimental systems, and predictive toxicology platforms can identify efficacy and safety signals earlier in the R&D cycle. Evaluating multiple candidate indications in parallel, with shared infrastructure and prespecified biomarker thresholds, helps identify the most promising programs faster. Explicit stage gates allow companies to terminate lower-probability programs early and reallocate capital to the highest-value options in real time.

Redesign can also compress time between phases. On average, 20 to 30 months elapse between the start of one clinical phase and the next. But there are four approaches that could soon shorten timelines:

• Internal decision making. Few organizations are examining where lag time accumulates between phases and how they make go/no-go decisions.
• Innovative trial designs. Adaptive trials and seamless phase 2 and 3 designs can eliminate the gap entirely for appropriate programs.
• Critical path management. Sponsors can optimize timelines across development activities so no part of the organization waits for a lagging element.
• Real-time regulatory engagement. AI-enabled sponsor-regulator ecosystems could eventually support continuous review and real-time dialogue without compromising scientific rigor. The FDA is already piloting this approach with AstraZeneca and Amgen, using Paradigm Health’s technology platform. Agency officials have suggested this model could potentially reduce overall clinical trial timelines by 20% to 40%. Designing early trials to support FDA Breakthrough Therapy designation can also lead to more frequent agency engagement.

FDA officials have suggested AI-enabled sponsor-regulator ecosystems could potentially reduce overall clinical trial timelines by 20% to 40%. 

Design the workforce and partnership model deliberately

Realizing this vision calls for a fundamental decision about what to build vs. buy: Which capabilities are core to competitive advantage and owned internally, and which are better sourced through AI-native CROs, specialized technology providers, and platform partners? The answer points to the company’s source of durable differentiation.

It also reshapes the internal organization. Today, large study teams coordinate across functions through meetings, handoffs, and periodic reviews. Tomorrow, smaller cross-functional teams will work alongside agents that track enrollment and site performance, reconcile operational and clinical data, identify emerging safety or protocol-compliance signals, and recommend interventions. Study managers shift from coordinating execution to overseeing agents that surface risks and recommend interventions, freeing them to focus on strategic judgment and critical decisions. The clinical research associate role is unbundled into a monitoring agent and a site-facing relationship owner who engages investigators and coordinators in areas of intense study competition and gathers unique insights through direct conversations. Medical directors spend more time on study design and strategy.

Leading sponsors will start redesigning roles, incentives, and team structures today, typically with an executive-level champion and initial functional centralization to catalyze uptake. They also understand that none of this means unsupervised automation. Decisions on patient selection, dose, safety, protocol, and regulatory claims require validated models, auditable data, explicit human accountability, and monitoring for bias and model drift. It’s also critical to ensure external AI systems don’t expose proprietary data or conflict with regional privacy requirements. Winning organizations will pair greater automation with stronger scientific, regulatory, and quality governance.

The competitive unit is no longer the individual trial or asset. It’s a system that reengineers clinical development with AI as the backbone. When innovators move the benchmark, companies that haven’t started redesigning will have to catch up, with less flexibility and higher cost. Those that have will already be efficiently advancing innovative therapies, with better patient, regulatory, and commercial outcomes.

The authors would like to acknowledge and thank the broader team that contributed to this point of view, including Brittany Rodriguez and Kristin Moneyron.