How to Break Through Pharma’s Innovation Bottleneck with AI

Pharma can’t afford to keep AI experiments in the lab. Emerging industry leaders are accelerating clinical development by scaling cross-functional, human-centric systems.

著者：Satty Chandrashekhar, Jackie Flanagan, Robbie Sandig, Enrique Pinuela, Fanny Perraudeau , Shikhar Kwatra, and Aarti Bagul

Brief

How to Break Through Pharma’s Innovation Bottleneck with AI

Pharma can’t afford to keep AI experiments in the lab. Emerging industry leaders are accelerating clinical development by scaling cross-functional, human-centric systems.

著者：Satty Chandrashekhar, Jackie Flanagan, Robbie Sandig, Enrique Pinuela, Fanny Perraudeau , Shikhar Kwatra, and Aarti Bagul

First published on 1月 07, 2026
min read


概要 Clinical development has become the bottleneck between discovery and patient outcomes. Trials take longer, costs are soaring, and complexity is overwhelming. AI can deliver higher, faster ROI in clinical development than in most other areas in pharma. To get there, the industry needs to move beyond isolated pilots to automated orchestration that reduces friction across stakeholders, accelerates timelines, and improves quality. Leading companies are making their data AI-ready, building orchestrated AI systems, and using AI to amplify human judgment, not replace it. Clinical development, the critical link between scientific discovery and real-world patient outcomes, is becoming a bottleneck in pharmaceutical innovation. AI is already compressing drug discovery cycles. But development isn’t keeping pace: R&D spending is rising while productivity remains flat. Trial timelines have grown by more than a third over the past decade. More than half of sites report bandwidth constraints and say complexity has increased since 2020. Without radical changes in trial design, execution, and governance, sponsors and contract research organizations (CROs) will watch discovery gains evaporate. Innovation will continue to outpace the ability to deliver it to patients. Written in conjunction with Written in conjunction with That’s not to say the industry hasn’t embraced AI. Large pharma companies are piloting AI-enabled recruitment and protocol design with start-ups, experimenting with internal “AI factories,” and modernizing their data foundations with hyperscalers. Biotech firms are testing niche vendors for safety review or recruitment; some are developing bespoke AI copilots for regulatory work. CROs and contract development and manufacturing organizations (CDMOs) are building proprietary AI trial platforms and investing in digital recruitment marketplaces. Many trial sites are testing AI-enabled add-ons or chatbots for patient engagement. The problem is fragmentation. Most AI efforts are narrow pilots resulting in disconnected tools, with only 28% of pilots making it into production on average, according to a survey from Bain & Company, Bessemer Venture Partners, and Amazon Web Services (AWS). Improvements are incremental and isolated. Clinical development is still held together by manual coordination across sponsors, CROs, investigators, and regulators. No clear leaders have emerged in the end-to-end transformation of clinical development. To smash the innovation bottleneck, next-generation leaders will need to build integrated, end-to-end AI applications that streamline core trial workflows, reduce friction among stakeholders, and meaningfully accelerate clinical development. Why clinical development is the right starting point for AI at scale Fragmented stakeholders and high complexity can make AI transformations seem daunting. But clinical development has a high concentration of immediate, high-value AI opportunities (see Figure 1). Figure 1 Clinical development AI use cases have higher and faster return on investment 出所 Bain analysis Why? Clinical development use cases combine: rich, structured data across clinical trial management systems (CTMS), electronic health records (EHR), and electronic patient-reported outcomes (ePRO) systems; repeatable, rule-based processes that can be codified for AI reasoning; direct P&L and net present value results from even modest cycle-time reductions—for example, getting from first patient in to database lock faster; and strong regulatory frameworks that reward explainability and traceability. Accelerating clinical development by 20% to 30% can translate to hundreds of millions in earlier sales per major asset for large sponsors, which scales to billions at the portfolio level thanks to earlier approval, faster market entry, and incremental revenue. And clinical development creates a scalable model for applying AI across R&D. Where AI can move the needle now The holy grail is reimagining the protocol design process. Today, it is fragmented, as teams tackle scientific, operational, and competitive considerations in sequence and in silos. That disconnect makes timely trade-offs between scientific rigor, cost, and speed difficult, especially under competitive pressure. AI can change that, integrating these dimensions much earlier. Using tools such as digital twins and advanced simulations, sponsors can assess feasibility, cost, timelines, and competitive benchmarks in real time, enabling more informed, holistic protocol design decisions. This level of precision is still aspirational: Only 13% of pharma companies are actively piloting AI-enabled protocol design, and very few, if any, have scaled solutions, according to a Bain, Bessemer Venture Partners, and AWS survey. However, among those few, 75% are satisfied with efficiency gains so far and expect, on average, four times ROI. Yet protocol design alone isn’t enough. Success also depends on operational excellence. There’s immense potential for sponsors and CROs alike to embed AI purposefully across portfolios, streamline site engagement, and enable teams with data-driven decision support. Across stages of trials, here’s how AI can change how work gets done. Winners will place their bets carefully, embed change at the right organizational layers, and differentiate through a superior site and employee experience, as well as data-driven study planning and execution. Several strategic paths are emerging. Horizontal: Commit fully to site-facing tools (e.g., using ChatGPT Enterprise in a HIPAA-compliant workspace), building AI literacy, improving site efficiency, and accelerating enrollment, ultimately becoming the sponsor of choice. Vertical: Embed AI throughout internal clinical operations to become the tech-first employer of choice with smarter, faster delivery. Upstream: Invest in reimagining protocol design—a high-risk, high-reward strategy for science-motivated innovators seeking to redefine trial design. End-to-end: Select a few lighthouse trials to test, learn, and scale for a more measured and focused path to enterprise-wide AI transformation. Why humans are still central Even as trial costs rise, focusing only on cost and efficiency misses a bigger opportunity: people. Bain research shows that a better investigator experience leads to faster trial recruitment and completion. Past transformations, like remote or risk-based monitoring, show that performance suffers when technology leapfrogs trust. Rather than automating away what matters most to the people involved, leading sponsors will start by developing an understanding of the realities of site monitors, coordinators, investigators, and patients. They will cocreate intuitive tools with users and build their confidence through measurable pilots that prove AI’s advantages. The future isn’t “autonomous trials” but a network of intelligent assistants supporting every stakeholder—sponsors, CROs, investigators, and regulators—in delivering faster, smarter, and more reliable trials. How to get started: a practical blueprint for an AI-powered clinical enterprise We’ve seen three common moves among organizations that are pulling ahead. 1. Make data AI-ready. Siloed, inconsistent, and poorly labeled trial data is a speed bump for scalable AI. Only about 20% of clinical trial data is analyzed while the rest is stored, inaccessible to AI. And 47% of pharmaceutical companies cite data readiness as a top obstacle in moving from pilot to scale, based on Bain’s survey with Bessemer and AWS. Industry leaders will connect and standardize operational and clinical data across CTMS, EHR, ePRO, and other platforms so that AI can reason across it in real time. They will establish shared ontologies, add missing metadata, and create a governed path for the clean entry of new data. 2. Orchestrate development as one AI-powered system. Successful organizations will move beyond standalone pilots to orchestrated AI systems that can plan and execute end-to-end workflows with escalation to humans as necessary. Tools like AgentKit from OpenAI can help teams test and build smart workflows without deep technical skills. With AgentKit’s Visual Agent Builder, for instance, teams can quickly create and refine automated processes using a simple interface, then export them as ready-to-use code for any cloud environment. A few principles that can guide teams as they build orchestrated systems: Start experimenting with simple setups, then extend them with deeper customization and integration to pave the way for a smooth transition into reliable, enterprise-grade systems. Build workflows on shared standards so specialized agents—such as those for protocol review or quality monitoring—can plug into the same system without creating new silos. This ensures consistency as capabilities grow. Bake in testing and monitoring to ensure automated agents follow protocols, handle exceptions correctly, and deliver consistent results. Reliability is not optional in regulated environments. Evaluation loops are especially critical for complex tasks requiring human oversight, like site activation or document review. The goal is a coordinated network of agents that reduce manual coordination, improving speed, accuracy, and reliability—while still keeping humans firmly in control. 3. Redesign the operating model for AI-augmented workflows. An organization can only scale AI as fast as it builds trust. As emerging leaders rethink their operating models for AI, including role redesign, they are focusing on the evolution of human-in-the-loop accountability. A human-in-the-loop role is not a single “checker” but a defined set of decision rights embedded in workflows to complement AI outputs, manage risk, and accelerate feedback loops that continuously improve model performance. How does this work in practice? As AI enables teams to shift to supervising, validating, and guiding AI outputs, roles will change. A clear RAPID® framework can ensure explicit human-in-the-loop workflows by clarifying Recommend, Agree, Perform, Input, and Decide responsibilities for reviewers, model stewards, and data curators. Updated standard operating procedures (SOPs) lay out clear escalation paths, quality thresholds, and audit requirements. For each workflow’s accountable AI owner, training and playbooks clarify when to override, retrain, or escalate, while governance ensures traceability and accountability regardless of team size. The result is a thinner execution layer and a stronger judgment layer with teams that make faster, safer decisions across the clinical life cycle. As agentic systems take on more routine coordination, data harmonization, and quality checks, it will allow individuals, supported by consistent guardrails and shared evaluative tools, to oversee a broader set of tasks. The shift doesn’t diminish human involvement; instead, it empowers humans to exercise the judgment, supervision, and accountability that will guide and refine these systems over time. Five actions to get started AI-enabled clinical development is not a distant future state. It is here, for the organizations that are willing to move beyond point tools toward orchestrated, human-centered systems. Here’s what leaders can do now: Set an enterprise mandate and pick your strategic lane. Bring together your CEO, CIO, and R&D leaders to explicitly decide where you want to win first: site-facing enablement (“horizontal”), internal operations transformation (“vertical”), protocol reinvention (“upstream”), or a few lighthouse trials (“end to end”). Tie your choice to cycle-time and quality outcomes. Empower a single accountable owner to lead it. Start with two or three high-value workflows. Choose use cases with clear pain points, strong data signals, and near-term ROI. Make your data AI-ready in a focused sprint. Don’t aim for “perfect data.” Focus on connected, standardized, usable data needed for your chosen workflows. This is the flywheel that makes orchestration possible. Build orchestrated AI systems with humans in the loop. Use low-code prototyping to get workflows into clinicians’ and operators’ hands quickly. Design guardrails, traceability, and review paths into every step. Redesign roles and incentives. Update SOPs with clear human-AI decision rights and establish AI owners for each workflow. Pair training with real use cases so AI literacy grows in context, not theory. The organizations that act today, moving forward with coordinated orchestration, will set the new pace of clinical development, bringing innovation to patients faster. OpenAI OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity. Founded in 2015, OpenAI develops cutting-edge AI technologies, including the GPT series of language models, and partners with organizations to integrate AI capabilities into real-world applications responsibly. OpenAI is committed to building safe, ethical AI systems and fostering transparency, safety, and alignment across the global AI ecosystem.

概要

Clinical development has become the bottleneck between discovery and patient outcomes. Trials take longer, costs are soaring, and complexity is overwhelming.
AI can deliver higher, faster ROI in clinical development than in most other areas in pharma.
To get there, the industry needs to move beyond isolated pilots to automated orchestration that reduces friction across stakeholders, accelerates timelines, and improves quality.
Leading companies are making their data AI-ready, building orchestrated AI systems, and using AI to amplify human judgment, not replace it.

Clinical development, the critical link between scientific discovery and real-world patient outcomes, is becoming a bottleneck in pharmaceutical innovation.

AI is already compressing drug discovery cycles. But development isn’t keeping pace: R&D spending is rising while productivity remains flat. Trial timelines have grown by more than a third over the past decade. More than half of sites report bandwidth constraints and say complexity has increased since 2020.

Without radical changes in trial design, execution, and governance, sponsors and contract research organizations (CROs) will watch discovery gains evaporate. Innovation will continue to outpace the ability to deliver it to patients.

Written in conjunction with

That’s not to say the industry hasn’t embraced AI. Large pharma companies are piloting AI-enabled recruitment and protocol design with start-ups, experimenting with internal “AI factories,” and modernizing their data foundations with hyperscalers. Biotech firms are testing niche vendors for safety review or recruitment; some are developing bespoke AI copilots for regulatory work. CROs and contract development and manufacturing organizations (CDMOs) are building proprietary AI trial platforms and investing in digital recruitment marketplaces. Many trial sites are testing AI-enabled add-ons or chatbots for patient engagement.

The problem is fragmentation. Most AI efforts are narrow pilots resulting in disconnected tools, with only 28% of pilots making it into production on average, according to a survey from Bain & Company, Bessemer Venture Partners, and Amazon Web Services (AWS). Improvements are incremental and isolated. Clinical development is still held together by manual coordination across sponsors, CROs, investigators, and regulators. No clear leaders have emerged in the end-to-end transformation of clinical development.

To smash the innovation bottleneck, next-generation leaders will need to build integrated, end-to-end AI applications that streamline core trial workflows, reduce friction among stakeholders, and meaningfully accelerate clinical development.

Why clinical development is the right starting point for AI at scale

Fragmented stakeholders and high complexity can make AI transformations seem daunting. But clinical development has a high concentration of immediate, high-value AI opportunities (see Figure 1).

Clinical development AI use cases have higher and faster return on investment

出所 Bain analysis

Why? Clinical development use cases combine:

rich, structured data across clinical trial management systems (CTMS), electronic health records (EHR), and electronic patient-reported outcomes (ePRO) systems;
repeatable, rule-based processes that can be codified for AI reasoning;
direct P&L and net present value results from even modest cycle-time reductions—for example, getting from first patient in to database lock faster; and
strong regulatory frameworks that reward explainability and traceability.

Accelerating clinical development by 20% to 30% can translate to hundreds of millions in earlier sales per major asset for large sponsors, which scales to billions at the portfolio level thanks to earlier approval, faster market entry, and incremental revenue. And clinical development creates a scalable model for applying AI across R&D.

Where AI can move the needle now

The holy grail is reimagining the protocol design process. Today, it is fragmented, as teams tackle scientific, operational, and competitive considerations in sequence and in silos. That disconnect makes timely trade-offs between scientific rigor, cost, and speed difficult, especially under competitive pressure. AI can change that, integrating these dimensions much earlier. Using tools such as digital twins and advanced simulations, sponsors can assess feasibility, cost, timelines, and competitive benchmarks in real time, enabling more informed, holistic protocol design decisions.

This level of precision is still aspirational: Only 13% of pharma companies are actively piloting AI-enabled protocol design, and very few, if any, have scaled solutions, according to a Bain, Bessemer Venture Partners, and AWS survey. However, among those few, 75% are satisfied with efficiency gains so far and expect, on average, four times ROI.

Yet protocol design alone isn’t enough. Success also depends on operational excellence. There’s immense potential for sponsors and CROs alike to embed AI purposefully across portfolios, streamline site engagement, and enable teams with data-driven decision support.

Across stages of trials, here’s how AI can change how work gets done.

Winners will place their bets carefully, embed change at the right organizational layers, and differentiate through a superior site and employee experience, as well as data-driven study planning and execution. Several strategic paths are emerging.

Horizontal: Commit fully to site-facing tools (e.g., using ChatGPT Enterprise in a HIPAA-compliant workspace), building AI literacy, improving site efficiency, and accelerating enrollment, ultimately becoming the sponsor of choice.
Vertical: Embed AI throughout internal clinical operations to become the tech-first employer of choice with smarter, faster delivery.
Upstream: Invest in reimagining protocol design—a high-risk, high-reward strategy for science-motivated innovators seeking to redefine trial design.
End-to-end: Select a few lighthouse trials to test, learn, and scale for a more measured and focused path to enterprise-wide AI transformation.

Why humans are still central

Even as trial costs rise, focusing only on cost and efficiency misses a bigger opportunity: people. Bain research shows that a better investigator experience leads to faster trial recruitment and completion.

Past transformations, like remote or risk-based monitoring, show that performance suffers when technology leapfrogs trust. Rather than automating away what matters most to the people involved, leading sponsors will start by developing an understanding of the realities of site monitors, coordinators, investigators, and patients. They will cocreate intuitive tools with users and build their confidence through measurable pilots that prove AI’s advantages.

The future isn’t “autonomous trials” but a network of intelligent assistants supporting every stakeholder—sponsors, CROs, investigators, and regulators—in delivering faster, smarter, and more reliable trials.

How to get started: a practical blueprint for an AI-powered clinical enterprise

We’ve seen three common moves among organizations that are pulling ahead.

1. Make data AI-ready. Siloed, inconsistent, and poorly labeled trial data is a speed bump for scalable AI. Only about 20% of clinical trial data is analyzed while the rest is stored, inaccessible to AI. And 47% of pharmaceutical companies cite data readiness as a top obstacle in moving from pilot to scale, based on Bain’s survey with Bessemer and AWS.

Industry leaders will connect and standardize operational and clinical data across CTMS, EHR, ePRO, and other platforms so that AI can reason across it in real time. They will establish shared ontologies, add missing metadata, and create a governed path for the clean entry of new data.

2. Orchestrate development as one AI-powered system. Successful organizations will move beyond standalone pilots to orchestrated AI systems that can plan and execute end-to-end workflows with escalation to humans as necessary. Tools like AgentKit from OpenAI can help teams test and build smart workflows without deep technical skills. With AgentKit’s Visual Agent Builder, for instance, teams can quickly create and refine automated processes using a simple interface, then export them as ready-to-use code for any cloud environment.

A few principles that can guide teams as they build orchestrated systems:

Start experimenting with simple setups, then extend them with deeper customization and integration to pave the way for a smooth transition into reliable, enterprise-grade systems.
Build workflows on shared standards so specialized agents—such as those for protocol review or quality monitoring—can plug into the same system without creating new silos. This ensures consistency as capabilities grow.
Bake in testing and monitoring to ensure automated agents follow protocols, handle exceptions correctly, and deliver consistent results. Reliability is not optional in regulated environments. Evaluation loops are especially critical for complex tasks requiring human oversight, like site activation or document review.

The goal is a coordinated network of agents that reduce manual coordination, improving speed, accuracy, and reliability—while still keeping humans firmly in control.

3. Redesign the operating model for AI-augmented workflows. An organization can only scale AI as fast as it builds trust. As emerging leaders rethink their operating models for AI, including role redesign, they are focusing on the evolution of human-in-the-loop accountability. A human-in-the-loop role is not a single “checker” but a defined set of decision rights embedded in workflows to complement AI outputs, manage risk, and accelerate feedback loops that continuously improve model performance.

How does this work in practice? As AI enables teams to shift to supervising, validating, and guiding AI outputs, roles will change.

A clear RAPID® framework can ensure explicit human-in-the-loop workflows by clarifying Recommend, Agree, Perform, Input, and Decide responsibilities for reviewers, model stewards, and data curators. Updated standard operating procedures (SOPs) lay out clear escalation paths, quality thresholds, and audit requirements.

For each workflow’s accountable AI owner, training and playbooks clarify when to override, retrain, or escalate, while governance ensures traceability and accountability regardless of team size. The result is a thinner execution layer and a stronger judgment layer with teams that make faster, safer decisions across the clinical life cycle.

As agentic systems take on more routine coordination, data harmonization, and quality checks, it will allow individuals, supported by consistent guardrails and shared evaluative tools, to oversee a broader set of tasks. The shift doesn’t diminish human involvement; instead, it empowers humans to exercise the judgment, supervision, and accountability that will guide and refine these systems over time.

Five actions to get started

AI-enabled clinical development is not a distant future state. It is here, for the organizations that are willing to move beyond point tools toward orchestrated, human-centered systems. Here’s what leaders can do now:

Set an enterprise mandate and pick your strategic lane. Bring together your CEO, CIO, and R&D leaders to explicitly decide where you want to win first: site-facing enablement (“horizontal”), internal operations transformation (“vertical”), protocol reinvention (“upstream”), or a few lighthouse trials (“end to end”). Tie your choice to cycle-time and quality outcomes. Empower a single accountable owner to lead it.
Start with two or three high-value workflows. Choose use cases with clear pain points, strong data signals, and near-term ROI.
Make your data AI-ready in a focused sprint. Don’t aim for “perfect data.” Focus on connected, standardized, usable data needed for your chosen workflows. This is the flywheel that makes orchestration possible.
Build orchestrated AI systems with humans in the loop. Use low-code prototyping to get workflows into clinicians’ and operators’ hands quickly. Design guardrails, traceability, and review paths into every step.
Redesign roles and incentives. Update SOPs with clear human-AI decision rights and establish AI owners for each workflow. Pair training with real use cases so AI literacy grows in context, not theory.

The organizations that act today, moving forward with coordinated orchestration, will set the new pace of clinical development, bringing innovation to patients faster.

OpenAI

OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity. Founded in 2015, OpenAI develops cutting-edge AI technologies, including the GPT series of language models, and partners with organizations to integrate AI capabilities into real-world applications responsibly. OpenAI is committed to building safe, ethical AI systems and fostering transparency, safety, and alignment across the global AI ecosystem.