DIA 2026 · Jun 14–18 · Philadelphia
AuroraPrime · DIA 2026 strategic prep
AuroraPrime · DIA 2026 主舞台战略准备
Pharma Priorities药企优先级
What 20 top pharmas are actually buying — AI priorities, pain points, blockers, and what they say in public.
20 家头部药企真实购买的内容：AI 优先级、痛点、阻碍因素、公开表态。
Pharmas tracked
覆盖药企
6 Tier 1 / 14 Tier 2-3
6 家 Tier 1 / 14 家 Tier 2-3
Stated AI priorities
AI 优先级主题
Drug discovery still #1
药物发现仍是第一
Adoption blockers
采用阻碍
Hallucination + GxP top the list
幻觉与 GxP 列首位
Documented pain points
痛点条目
CSR critical-path most urgent
CSR 关键路径最紧迫
Stated AI priorities公开表态的 AI 优先级
Ranked by frequency in earnings calls, R&D days, and DIA/RAPS panels — and weighted by where the money is.
按财报会议、R&D Day、DIA / RAPS 论坛中的出现频率排序，并以投入金额加权。
AI-driven drug discovery / target ID / molecule design
AI 驱动药物发现 / 靶点 / 分子设计
Sanofi (plai), AstraZeneca (BenevolentAI), Lilly (OpenAI/NVIDIA), Novartis (Microsoft, Isomorphic), Bayer (Google)
Sanofi（plai）、AstraZeneca（BenevolentAI）、Lilly（OpenAI/NVIDIA）、Novartis（Microsoft、Isomorphic）、Bayer（Google）
GenAI for medical affairs, MLR, content production
GenAI：医学事务、MLR、内容生产
Pfizer 'Charlie' (2023→2024-25 expansion), Sanofi plai apps, J&J in-house, GSK MLR pilots
Pfizer「Charlie」（2023 上线，2024-25 扩展）、Sanofi plai 应用、J&J 自研、GSK MLR 试点
Agentic / 'copilot' AI for R&D operations
智能体 AI 助手：临床运营、PV、数据管理
Novartis 'AI factory' (Microsoft), AZ Evinova, Roche internal copilots, BI X digital lab
Novartis「AI factory」（Microsoft）、AstraZeneca Evinova、Roche 内部 copilots、BI X 数字实验室
Regulatory affairs / submissions / medical writing automation
监管事务 / 递交 / 医学写作自动化
Roche CSR pilots, GSK regulatory writing, Takeda RA roadmap, AbbVie submissions
Roche CSR 试点、GSK 监管写作、Takeda RA 路线图、AbbVie 递交
Pharmacovigilance / signal detection / safety narratives
药物警戒 / 信号检测 / 安全性叙述
Sanofi safety processing, J&J signal detection, Bayer ICSR triage, Pfizer adverse-event narratives
Sanofi 安全性数据处理、J&J 信号检测、Bayer ICSR 分诊、Pfizer 不良事件叙述
Documented pain points已记录的痛点
Why each pain matters at the board level, and the public signal you can cite from stage.
每个痛点为何会进入董事会层面，以及可在主舞台引用的公开信号。
CSR drafting on the critical path of pivotal P3 readout
CSR 撰写处于关键 III 期递交关键路径
Why board cares: Each week of CSR delay = week of submission delay = NPV loss
董事会为何关心：CSR 每延误 1 周 = 递交延误 1 周 = NPV 损失
Signal: Multiple Top-10 pharmas have stated CSR-cycle-time targets at DIA
信号：多家 Top-10 药企已在 DIA 公开提出 CSR 周期时间目标
Medical writer supply constraint
医学写作人才短缺
Why board cares: Programs blocked not by science but by writer FTE availability
董事会为何关心：项目卡住不是因为科学问题，而是写作 FTE 不够
Signal: EMWA / AMWA shortage data 2023–2024
信号：EMWA / AMWA 2023–2024 人才短缺数据
Quality + consistency across CTD modules
CTD 模块间质量与一致性
Why board cares: Inconsistent description → regulator queries → review delays
董事会为何关心：描述不一致 → 监管发问 → 审评延误
Signal: FDA RTF / EMA Day-120 question patterns
信号：FDA RTF / EMA Day-120 问题模式
eCTD v4.0 / ICH M11 transition burden
eCTD v4.0 / ICH M11 过渡负担
Why board cares: Structural change to how protocols are authored
董事会为何关心：方案撰写方式发生结构性变化
Signal: FDA/EMA/PMDA roadmaps 2024–2026
信号：FDA / EMA / PMDA 2024–2026 实施路线图
Response-to-questions cycle time during agency review
审评期间答复周期时间
Why board cares: 30-day deadlines consume RA + writing FTE at peak load
董事会为何关心：30 天答复期在高峰期榨干 RA 与写作 FTE
Signal: Surfaces in DIA / RAPS panel Q&A every year
信号：每年都在 DIA / RAPS 论坛问答中浮现
PSUR / PBRER / DSUR / PADER periodic safety reporting
PSUR / PBRER / DSUR 周期性安全报告
Why board cares: High-volume, repeating, classic automation target
董事会为何关心：高量、重复、典型的自动化目标
Signal: Sanofi, Pfizer, J&J safety-narrative automation pilots
信号：Sanofi、Pfizer、J&J 的安全性叙述自动化试点
Multi-regional submissions
多区域递交（同一档案 × 不同区域内容）
Why board cares: Combinatorial explosion of variants
董事会为何关心：变体数量呈组合式爆炸
Signal: Top-20 pharmas file in 50–100+ jurisdictions
信号：Top-20 药企在 50–100+ 个司法辖区递交
Lay summaries (EU CTR Article 37)
通俗摘要（EU CTR 第 37 条）
Why board cares: Mandatory, recurring, multilingual, plain-language
董事会为何关心：强制、重复、多语种、通俗化
Signal: EU CTR full effect 2023; backlog reported
信号：EU CTR 于 2023 年全面生效；已出现积压
Investigator brochure (IB) updates at high cadence
研究者手册高频更新
Why board cares: Constant low-grade authoring burden
董事会为何关心：持续性的低强度撰写负担
Signal: Chronic, under-discussed
信号：长期存在但鲜被讨论
Adoption blockers (ranked)采用阻碍（排名）
What stops AI deals from closing inside pharma. Mention 5 of these on stage and you'll have the room.
阻碍 AI 交易在药企内部成交的因素。
Hallucination risk in regulated content
幻觉风险（监管内容）
Sophisticated framing: We need traceability of every claim to a structured source
高阶论述：我们需要把每条论断都可追溯到结构化来源
GxP / 21 CFR Part 11 / Annex 11 compliance
GxP / Part 11 / Annex 11 合规
Sophisticated framing: GAMP 5 SE (2022), CSA mindset, model qualification distinct from system qualification
高阶论述：GAMP 5 SE（2022）、CSA 思维，模型验证有别于系统验证
Validation cost & re-validation on model updates
验证成本 & 模型更新需重新验证
Sophisticated framing: Model version pinning + re-qualification protocol + risk-based scaling
高阶论述：模型版本锁定 + 再验证规程 + 基于风险的分级处理
Data residency, IP leakage, training-data concerns
数据驻留、IP 泄漏、训练数据担忧
Sophisticated framing: Tenant-isolated, BYOC, on-prem, no-train contractual guarantees
高阶论述：租户隔离、BYOC、本地部署、合同层面保证不用于训练
20-pharma AI signal log20 家药企 AI 信号档案
Filter by tier; sort by pharma name or date. Sources are from public filings + investor materials.
按 Tier 筛选，按药企名或日期排序。数据来源于公开披露与投资者材料。
Pharma药企↕ | Tier梯队▲ | Signal信号 |
|---|---|---|
| Pfizer | Tier 1 | 'Charlie' GenAI content platform (Publicis); expanded to other content domains「Charlie」GenAI 内容平台（与 Publicis 合作）；已扩展到其他内容领域 |
| Sanofi | Tier 1 | 'plai' platform; 'first pharma powered by AI at scale' (Hudson)「plai」平台；CEO Hudson 提出「首个 AI 大规模赋能的药企」 |
| Novartis | Tier 1 | Microsoft 'AI factory'; Isomorphic Labs collab与 Microsoft 共建「AI factory」；与 Isomorphic Labs 合作 |
| AstraZeneca | Tier 1 | Evinova digital health spinout; BenevolentAI partnershipEvinova 数字健康分拆；与 BenevolentAI 合作 |
| Eli Lilly | Tier 1 | OpenAI AMR partnership; NVIDIA BioNeMo; Catalyze360与 OpenAI 抗菌耐药合作；NVIDIA BioNeMo；Catalyze360 |
| Novo Nordisk | Tier 1 | NVIDIA Denmark 'Gefion' supercomputer (~$100M)丹麦「Gefion」超算项目（NVIDIA 合作，约 $100M） |
| J&J | Tier 2 | Multiple internal GenAI tools across R&D, MA, PV; NVIDIA Ethicon在 R&D、MA、PV 多领域部署内部 GenAI 工具；Ethicon × NVIDIA |
| Roche | Tier 2 | Genentech ML pubs; internal RA pilots; NAVIFYGenentech ML 论文产出；内部 RA 试点；NAVIFY 平台 |
| Merck & Co. | Tier 2 | Moderna mRNA AI; internal GenAI clinical ops; AWSModerna mRNA AI；内部临床运营 GenAI；AWS 合作 |
| Bristol Myers Squibb | Tier 2 | AI target ID; internal GenAI; ExscientiaAI 靶点识别；内部 GenAI；与 Exscientia 合作 |
| GSK | Tier 2 | 'Jules' / GenAI assistants; Tempus AI oncology; Cerebras「Jules」等 GenAI 助手；Tempus AI 肿瘤合作；Cerebras |
| Bayer | Tier 2 | Google Cloud; 'MyData'; AI in radiologyGoogle Cloud 合作；「MyData」；放射学 AI |
| Amgen | Tier 2 | 'ATOMIC' AI for drug design; clinical-ops GenAI「ATOMIC」AI 用于药物设计；临床运营 GenAI |
| Boehringer Ingelheim | Tier 2 | 'BI X' digital lab; IBM partnership; internal GenAI「BI X」数字实验室；与 IBM 合作；内部 GenAI |
| Takeda | Tier 2 | Digital roadmap; AI in discovery + clin-ops; RA digital数字化路线图；发现 + 临床运营的 AI；RA 数字化 |
| AbbVie | Tier 3 | Internal AI for clinical ops + submissions; vendor pilots内部用于临床运营 + 递交的 AI；外部厂商试点 |
| Gilead | Tier 3 | Lower public AI profile; pilot disclosures公开 AI 形象较低调；披露过若干试点 |
| Daiichi Sankyo | Tier 3 | Lower English-language profile; JP-market DT announcements英文渠道曝光少；以日本市场为主的数字化转型公告 |
| Astellas | Tier 3 | 'Rx+' digital strategy; AI in discovery「Rx+」数字战略；药物发现 AI |
| Biogen | Tier 3 | Cloud + AI vendor partnerships; internal pilots云 + AI 厂商合作；内部试点 |
ROI / budget referenceROI / 预算参考
Numbers to anchor the cost-justification conversation in dollars-of-NPV, not hours-of-FTE.
把成本论证锚定在 NPV 美元，而非 FTE 工时。
Top-20 pharma R&D budget
Top-20 药企 R&D 预算
$8–14B/yr each
Largest 5 (Pfizer, J&J, Roche, Merck, Lilly); ~$200B+ industry
前 5 大（Pfizer、J&J、Roche、Merck、Lilly）；行业整体 $200B+
AI / digital transformation budget
AI / 数字化转型预算
1–3% of R&D
$80M–$400M/yr at Top-10 scale (BCG / McKinsey 2024)
Top-10 体量下每年约 $80M–$400M（BCG / McKinsey 2024）
Industry GenAI ROI consensus
行业 GenAI ROI 共识
30–50% time reduction
First-draft authoring (DIA / RAPS panels)
首稿撰写场景（DIA / RAPS 论坛）
AuroraPrime CSR/protocol benchmark
AuroraPrime CSR / 方案基准
90% first-draft time reduction
Internal benchmark (alphalifesci.com)
内部基准（alphalifesci.com）
AuroraPrime safety narratives
AuroraPrime 安全性叙述
95% first-draft time reduction
Internal benchmark
内部基准
Novo Nordisk 'Gefion' compute
Novo Nordisk「Gefion」算力投入
~$100M
NVIDIA + Foundation + EIFO, Nov 2024
NVIDIA + Foundation + EIFO，2024 年 11 月
Voice of customer客户声音
"We have 50 pilots and 0 in production. Talk to me when you have inspection-ready production deployments."
"我们有 50 个试点，0 个进入生产。等你做出能通过监管检查的生产部署，我们再谈。"
— Pharma RA leader (DIA panel)
—— 某药企 RA 负责人（DIA 论坛）
Show me production, not pilots
拿出生产案例，不要试点
"The model is the easy part. The validation envelope is 80% of the project."
"模型本身是简单部分；验证体系才占整个项目 80%。"
— Pharma quality lead (ISPE GAMP)
—— 某药企质量负责人（ISPE GAMP）
Validation is the work
验证才是真正的工作量
"Every claim in the document must trace to a source artifact. No exceptions."
"文档里的每一条论断都必须可追溯到源头物料，没有例外。"
— Pharma RA panelist (RAPS)
—— 某药企 RA 嘉宾（RAPS）
We need traceability, not magic
我们要的是可追溯，不是魔法
"We are not going to be the FDA's first AI-failure case study."
"我们绝不愿意成为 FDA 第一个 AI 失败的案例。"
— Top-10 pharma RA head (off-the-record)
—— Top-10 药企 RA 负责人（私下交流）
Risk aversion is rational
风险规避是理性的
Recurring themes (DIA / RAPS)DIA / RAPS 反复出现的主题
Show me production, not pilots
拿出生产案例，不要试点
"We have 50 pilots and 0 in production"
"我们有 50 个试点，0 个进入生产"
Heard at: Every panel
听到于：几乎所有论坛
Validation is the work
验证才是真正的工作量
"The validation envelope is 80% of the project"
"验证体系占整个项目的 80%"
Heard at: DIA quality + RA tracks; ISPE GAMP
听到于：DIA 质量 + RA 议题；ISPE GAMP
We need traceability, not magic
我们要的是可追溯，不是魔法
"Every claim must trace to a source artifact"
"每条论断都必须可追溯到源头物料"
Heard at: RAPS panels
听到于：RAPS 论坛
Vault is the system of record
Vault 才是记录系统
"Anything that doesn't round-trip with Vault has an adoption ceiling"
"凡是不能与 Vault 双向同步的方案，采用率都有天花板"
Heard at: Veeva Summit + DIA
听到于：Veeva Summit 与 DIA
ICH M11 changes everything for protocols
ICH M11 对方案撰写是颠覆性的
"M11 makes protocols structurable"
"M11 让方案变得可结构化"
Heard at: DIA + TransCelerate
听到于：DIA 与 TransCelerate
Regulatory response drafting is highest-leverage AI use case
监管答复撰写是杠杆最高的 AI 用例
"Time-pressured, well-bounded, repeatable, high-value"
"时间压力大、范围明确、可重复、价值高"
Heard at: RAPS, DIA RA tracks
听到于：RAPS、DIA RA 议题
AI literacy in RA staff is a gap
RA 团队的 AI 素养是短板
"Hiring + training implications"
"对招聘与培训提出新要求"
Heard at: RAPS leadership panels
听到于：RAPS 高层论坛