Agentic AI Platform for Clinical Development and Commercial Operations in Pharmaceutical Drug Development

astrazeneca · Healthcare · 2024 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.2 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Pharmaceutical operations

AstraZeneca needed agentic AI systems for clinical development and commercial operations to accelerate drug-development and field workflows.

Constraints:

Clinical/regulatory domains
Global user base
Many data products
Commercial scale

Architecture

AstraZeneca's agentic AI platform spans clinical development and commercial operations. The Development Assistant connects regulated clinical, regulatory, patient-safety, and quality data products so authorized users can ask natural-language questions, inspect reasoning, trace answers back to source documents and data products, and rely on nine agents over sixteen data products, eight knowledge bases, and seven domains. The commercial AZ Brain/InsideIQ side composes 20-plus agents, models, documents, datasets, and guidelines to produce n=1 commercial recommendations, content workflows, and reimbursement dossier support.

Diagram confidence: explicit (topology described by the source).

flowchart LR
  development-assistant[Development Assistant] --> clinical-data-products[Clinical data products]
  clinical-data-products --> clinical-agent-set[Clinical agent set]
  clinical-agent-set --> knowledge-bases[Knowledge bases]
  clinical-agent-set --> source-provenance[Source provenance]
  source-provenance --> development-assistant
  az-brain-platform[AZ Brain platform] --> commercial-agent-suite[Commercial agent suite]
  commercial-agent-suite --> models-and-datasets[Models and datasets]

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Development Assistant`	Clinical-development assistant for natural-language portfolio queries.	`Clinical data products`, `Clinical agent set`, `Source provenance`
`Clinical data products`	Sixteen governed data products across clinical, patient safety, regulatory, and quality domains.	`Development Assistant`, `Clinical agent set`
`Clinical agent set`	Nine agents working across knowledge bases and domains.	`Clinical data products`, `Knowledge bases`, `Source provenance`
`Knowledge bases`	Eight knowledge bases used by the clinical assistant.	`Clinical agent set`
`Source provenance`	Reasoning, data, source-document, and data-product links for user verification.	`Development Assistant`
`AZ Brain platform`	Commercial intelligence platform for patient, HCP, content, and reimbursement workflows.	`Commercial agent suite`, `Models and datasets`
`Commercial agent suite`	20-plus agents that query datasets, models, documents, and guidelines.	`AZ Brain platform`, `Models and datasets`
`Models and datasets`	Predictive, classification, and generative models plus documents and guidelines.	`Commercial agent suite`

Data flow

Authorized users ask questions through Development Assistant or commercial assistants. Agents retrieve governed data products, knowledge bases, models, documents, and guidelines; the assistant shows reasoning, data, sources, and clickable provenance; commercial agents query cross-domain datasets and models to generate concrete recommendations or content/dossier artifacts for review.

3 supporting quotes

“It allows any user with the appropriate level of access to those data products, which is key when we're talking about regulated environments to ask any question they have in a very natural language kind of way.”transcript, stated

“You'll see from the screen that it shows the reasoning, it shows the data, it shows you where it's pulled that information from and provides all of the insights in a very simple, easy to understand way that you can then click through to every single source document”transcript, stated

“Again, it's a collection of about 20 plus agents that queries cross and interrogates all the different data sets”transcript, stated

Why this architecture

AstraZeneca needed governed multi-agent access because clinical and commercial workflows depend on siloed data, regulated access, and source verification. The platform connects previously separate products and models while preserving authorization and provenance, and the commercial side uses agent collections to turn many datasets and models into patient- or HCP-level recommendations.

3 supporting quotes

“the complexity of the work that we do across the globe can lead to disparate systems. It can lead to siloed data. And what that means is it can be really difficult to answer questions.”transcript, implied

“Now the way that we've done that in the past is manual. It's hours and hours of work. It's pulling data from here, from there, a little bit of data here, transforming that data, analyzing that data, getting some insights.”transcript, stated

“single source document, source, data, product that we have. So you can go back and you can check, is this accurate? Is this telling me the right information? Oh, that's a really interesting insight. I'm not sure I believe that. Let me go and understand that. Or you know what, I found out something”transcript, stated

Models used: generative AI models · predictive models · classification models

Scale

The Development Assistant moved from proof of concept to MVP in six weeks.
“In six weeks, we went from a proof of concept to an MVP. Reaching more than a thousand users across 21 different countries.”transcript, stated
It reached more than 1,000 users across 21 countries.
“Reaching more than a thousand users across 21 different countries. This is a truly global technology.”transcript, stated
Commercial users engaging with the assistant generated two times more scripts.
“People who use this and engage with it generate two times more scripts. That's the contributing to the revenues plus reaching the patients.”transcript, stated

Alternatives considered (roads not taken)

Manual cross-system clinical research

Rejected because: Manual data pulling across silos took hours and slowed trial-design questions.

“It's pulling data from here, from there, a little bit of data here, transforming that data, analyzing that data, getting some insights.”transcript, implied

Analytics-oriented data products unchanged for agents

Rejected because: AstraZeneca says agent consumption needs different patterns than human analytics interfaces.

“A lot of these data products have been traditionally designed with analytics and human interaction in mind agents are not the same. You know, agents do require specific patterns and specific assets to be organized in a certain way and be before you know it.”transcript, stated

Patterns

`multi-agent-orchestration` (pharma-platform)

Clinical development use cases are served by multiple coordinated agents.

“that clinical patient safety, regulatory, and quality domain. There are nine agents working together. So it's truly a multi-agent system.”transcript, stated

`tool-use-orchestration` (data-products)

Agents are packaged with access to governed clinical data products.

“It allows any user with the appropriate level of access to those data products, which is key when we're talking”transcript, stated

Tradeoffs

Invest in production-grade agent infrastructure rather than one-off prototypes.

Chose Secure runtimes, identity, memory, marketplace patterns, and maintained healthcare/life-science starters over Letting each team chase fast MVPs with separate frameworks and infrastructure.

Rationale: The platform had to scale agents in regulated clinical and commercial settings where developers face a learning curve and governance requirements.

“So even though you know many of you in the room may be familiar with what AgentCore is, it's still a learning curve. You know, we continuously launch so many new features”transcript, stated

How to expose clinical data to agents.

Chose governed data products with source provenance over manual extraction from disconnected systems.

Rationale: Regulated users need natural-language access while still being able to trace results to the underlying source.

“provides all of the insights in a very simple, easy to understand way that you can then click through to every single source document”transcript, stated

Gotchas (what bit them)

Agentic AI projects hit walls when data products are not designed for agents.

Root cause: Many existing data assets were built for analytics and human interaction rather than agent consumption.

Resolution: AstraZeneca and AWS emphasized agent-ready data foundations before scaling workflow automation.

Clinical research questions were slowed by siloed global trial data.

Root cause: Clinical, regulatory, patient-safety, and quality data lived across disparate systems and required manual transformation.

Resolution: The Development Assistant connected data products and agents so authorized users could query the portfolio in natural language.

“It can lead to siloed data. And what that means is it can be really difficult to answer questions. So for example, our clinical researchers may be thinking”transcript, stated

“Now the way that we've done that in the past is manual. It's hours and hours of work. It's pulling data from here, from there, a little bit of data here”transcript, stated

Evaluation

The case reports workflow acceleration, prescription lift, and authoring time reduction.

Metrics: time-to-market · prescriptions · authoring timeline

“g 20 new medicines by 2030. The company built two major production systems: a Development Assistant serving over 1,000 users across 21 countries that integrates 16 data products with 9 agents to enable natural language queries across clinical trials, regulatory submissions, patient safety, and quali”summary, stated

Outcomes

The summary reports Development Assistant served 1,000+ users across 21 countries.

Commercial assistant use is linked to increased prescriptions in the summary.

“The implementation reduced time-to-market for various workflows from months to weeks, with field teams using the commercial assistant generating 2x more prescriptions, and reimbursement dossier authoring timelines dramatically shortened through automated agent workflows.”summary, stated