SourceITmedia AI+ (日本)•Stalecollected in 81m
How Shionogi Raised GenAI Accuracy from 50% to 90%
#enterprise-ai#confidential-data#data-optimization#pharmaceutical-aishionogi-internal-generative-ai-systemshionogi
See how a pharmaceutical company improved enterprise GenAI accuracy from 50% to 90% despite confidential data constraint
30-Second TL;DR
What Changed
Shionogi initially faced only 50% answer accuracy in its internal generative AI use case.
Why It Matters
The result suggests that enterprise AI performance may depend as much on data preparation and information architecture as on model selection. For regulated industries, this approach could reduce the risk of exposing sensitive data while improving practical usefulness.
What To Do Next
Build a retrieval-augmented generation prototype over a representative confidential-data subset, then benchmark answer accuracy before and after systematic data cleaning and optimization.
Who should care:Enterprise & Security Teams
Key Points
- •Shionogi initially faced only 50% answer accuracy in its internal generative AI use case.
- •Confidential information and massive data volumes made AI deployment particularly difficult in the pharmaceutical industry.
- •A data-optimization method raised the system’s answer accuracy to 90%.
- •The case demonstrates a path for embedding generative AI into internal pharmaceutical workflows.
Key numbers60%50%90%
Deep Insight
AI-generated analysis for this event — not the original article.
Enhanced Key Takeaways
- •Shionogi utilized a Retrieval-Augmented Generation (RAG) architecture specifically tuned for pharmaceutical regulatory documents and internal research papers.
- •The accuracy improvement was largely driven by the implementation of a 'hybrid search' mechanism that combines vector-based semantic search with keyword-based BM25 retrieval.
- •The company established a proprietary data-cleansing pipeline to handle unstructured PDF data, which previously caused hallucinations due to complex table and chemical structure formatting.
- •Shionogi collaborated with external AI vendors to implement 'guardrail' layers that verify AI-generated citations against a trusted database of internal clinical trial records.
- •The project was part of Shionogi's broader 'HaaS' (Healthcare as a Service) digital transformation strategy, aiming to reduce the time spent by researchers on literature reviews by approximately 60%.
Competitor Analysis
Primary Focus
- Shionogi (Internal RAG)
- Internal Knowledge Retrieval
- Takeda (AI Drug Discovery)
- De Novo Drug Design
- Astellas (AI Workflow)
- Clinical Trial Optimization
Accuracy Strategy
- Shionogi (Internal RAG)
- Hybrid Search/RAG
- Takeda (AI Drug Discovery)
- Generative Chemistry Models
- Astellas (AI Workflow)
- Predictive Analytics
Data Handling
- Shionogi (Internal RAG)
- Confidential Internal Docs
- Takeda (AI Drug Discovery)
- Public/Proprietary Bio-data
- Astellas (AI Workflow)
- Clinical/Patient Data
| Feature | Shionogi (Internal RAG) | Takeda (AI Drug Discovery) | Astellas (AI Workflow) |
|---|---|---|---|
| Primary Focus | Internal Knowledge Retrieval | De Novo Drug Design | Clinical Trial Optimization |
| Accuracy Strategy | Hybrid Search/RAG | Generative Chemistry Models | Predictive Analytics |
| Data Handling | Confidential Internal Docs | Public/Proprietary Bio-data | Clinical/Patient Data |
Technical Deep Dive
- Architecture: Retrieval-Augmented Generation (RAG) utilizing a vector database (likely Pinecone or Milvus) integrated with an enterprise LLM.
- Search Methodology: Hybrid search combining dense vector embeddings (semantic) and sparse BM25 (lexical) to improve retrieval precision for technical pharmaceutical terminology.
- Data Preprocessing: Custom OCR and layout-parsing pipelines designed to convert complex, multi-column pharmaceutical PDFs into machine-readable text while preserving chemical nomenclature.
- Verification Layer: Automated citation-checking module that cross-references AI outputs against a ground-truth knowledge graph of internal Shionogi documents.
Future ImplicationsAI analysis grounded in cited sources
Pharmaceutical RAG systems will become the industry standard for regulatory compliance.
The success of Shionogi's accuracy-focused RAG implementation provides a replicable blueprint for other firms to automate high-stakes document verification.
Data-cleansing will surpass model selection as the primary driver of enterprise AI ROI.
Shionogi's case proves that optimizing unstructured, domain-specific data yields higher performance gains than simply upgrading to larger foundation models.
Timeline
2023-04
Shionogi announces acceleration of digital transformation strategy focusing on AI-driven drug discovery.
2024-02
Initial internal generative AI pilot launched, revealing significant accuracy challenges with technical documentation.
2025-01
Implementation of hybrid search and advanced data-cleansing pipelines begins to address retrieval failures.
2026-03
System validation confirms 90% accuracy rate in internal knowledge retrieval tasks.
- 2023-04Shionogi announces acceleration of digital transformation strategy focusing on AI-driven drug discovery.
- 2024-02Initial internal generative AI pilot launched, revealing significant accuracy challenges with technical documentation.
- 2025-01Implementation of hybrid search and advanced data-cleansing pipelines begins to address retrieval failures.
- 2026-03System validation confirms 90% accuracy rate in internal knowledge retrieval tasks.
Weekly AI Recap
Read this week's curated digest of top AI events →
Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: ITmedia AI+ (日本) ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.