We’re putting $5.5M behind research and development of smaller, specialized AI models for real-world deployment — Read our manifesto →
    All posts

    [Case Study] 750 GPTs in 60 Days: Moderna's Distributed AI Adoption Model

    How a biopharma company combined a secure enterprise assistant, employee-built tools, and scientific workflows to scale adoption without one central backlog.

    Conscious Engines

    Industry: Biotechnology and pharmaceuticals
    Organization: Moderna
    Use case: Distributed employee AI and scientific workflow support
    Evidence basis: OpenAI customer case using Moderna-reported adoption data
    Disclosure: This is an independent analysis by Conscious Engines. The source is published by the model provider and emphasizes adoption rather than audited financial return.

    1. Outcome at a Glance

    Moderna employees created 750 custom GPTs within two months. The company reported that 40% of weekly active users had created a GPT, with an average of 120 conversations per user per week. Its mChat assistant reached more than 80% of employees.

    Key Outcomes

    750 in 60 days

    Custom GPTs created

    Creation scale reported in the cited case.

    40%

    Weekly active users who created a GPT

    Distributed building behavior reported in the cited case.

    More than 80% of employees

    mChat reach

    Broad adoption reported in the cited case.

    About 2,000 weekly participants

    Internal AI forum

    Community scale reported in the cited case.

    MeasureReported resultEvidence note
    Custom GPTs created750 in 60 daysCreation scale
    Weekly active users who created a GPT40%Distributed building behavior
    Average conversations120 per user per weekUsage intensity
    mChat reachMore than 80% of employeesBroad adoption
    Internal AI forumAbout 2,000 weekly participantsCommunity scale

    The figures demonstrate unusually fast adoption. They do not establish that all 750 GPTs were active, unique, safe, or economically valuable. The strongest evidence is behavioral: employees used AI frequently and many built task-specific tools rather than waiting for a central team.

    2. The Operational Problem

    Biopharma work spans research, clinical development, manufacturing, regulatory, legal, commercial, and corporate functions. A central AI team cannot discover and build every useful workflow. Domain experts know where repetitive synthesis, drafting, analysis, and search work occurs, but may not be software developers.

    Opening unrestricted AI access would create another problem. Users could expose confidential data, produce unsupported scientific claims, duplicate tools, or rely on unvalidated outputs in regulated workflows.

    The challenge was to create a secure environment where employees could experiment and encode task knowledge, while keeping higher-risk applications under stronger engineering and governance.

    3. What Was Built

    Moderna combined mChat, enterprise ChatGPT access, custom GPT creation, and more specialized applications. One highlighted workflow, Dose ID, helped teams analyze and visualize clinical data to validate dose choices.

    System at a Glance

    Enterprise assistant

    General drafting, analysis, and knowledge work.

    Custom GPT builder

    Let domain experts encode instructions and reference material.

    Community

    Share examples and build user capability.

    Specialized applications

    Support higher-value scientific workflows such as Dose ID.

    Governance

    Apply data, access, and intended-use controls.

    LayerRole
    Enterprise assistantGeneral drafting, analysis, and knowledge work
    Custom GPT builderLet domain experts encode instructions and reference material
    CommunityShare examples and build user capability
    Specialized applicationsSupport higher-value scientific workflows such as Dose ID
    GovernanceApply data, access, and intended-use controls

    The custom GPT layer reduces the central backlog but does not remove the need for production engineering. A useful prototype may still need validated data connectors, role-based access, evaluation, monitoring, and a stable user interface.

    The correct portfolio separates personal productivity, team tools, operational systems, and regulated decision support. Each tier receives a different review and release process.

    4. How It Reached Production

    Moderna paired technology access with a strong internal adoption program.

    Make creation accessible. Domain experts can encode instructions and examples without waiting for full application development.

    Build a community. About 2,000 weekly forum participants created a channel for use cases, training, and peer learning.

    Promote valuable prototypes. High-use tools can move from individual experimentation into managed enterprise products.

    Tier by risk. A writing assistant and a clinical decision-support workflow cannot share the same approval requirements.

    Measure utility, not creation. Track active users, accepted outputs, cycle time, error, reuse, and outcome value. A count of 750 is a discovery signal, not a value total.

    The production funnel should retire duplicate or unused tools, consolidate common capabilities, and add stronger controls to workflows that affect regulated records or scientific decisions.

    5. What Pharma Leaders Should Take Away

    Moderna's case is a strong model for demand discovery. Employees created hundreds of task-specific assistants quickly, showing where language friction existed across the enterprise.

    That demand map should determine what deserves a bespoke production model. High-volume or high-value workflows can move to private RAG, domain SLMs, specialized speech, validated structured output, and integrated applications. Low-risk personal tasks can remain on a general assistant.

    The target metric is value per active production workflow, not the number of bots created. Distributed experimentation finds the opportunities. Bespoke engineering turns the best of them into dependable enterprise systems.

    Sources