Applied AI
How RF Consulting uses AI, and the controls around it
AI is part of how I deliver every engagement, and part of what I help commercial teams put into practice. This page covers the environment I run for sensitive work, the tools I use day to day, and where AI earns its place in a commercial organization.
Restricted work
A secure environment for material nonpublic information
Launch work runs into information that can move a share price: trial readouts, FDA correspondence, financing and partnership terms. Consumer AI tools are the wrong place for it. Restricted work runs only on open-source large language models, in one of two places.
- On my own encrypted hardware, where prompts, documents, and outputs never leave the machine.
- On frontier-level open-source models hosted by trusted providers with compute primarily in the U.S. Prompts and responses are never logged or used for training, and the providers keep no data.
The same controls apply in both places:
- Local model files come from the publishers’ official distribution channels and run with local inference software.
- During Restricted work, models run without plug-ins or web search, and nothing goes to a service that logs, retains, or trains on the data.
- Prompts, outputs, and document indexes are stored in the client’s Restricted folder on the encrypted disk and deleted at engagement close.
- When a general cloud tool such as Claude or ChatGPT helps with structure, such as a model shell or a slide framework, Restricted material is abstracted first: names removed and figures replaced with placeholders, so the text reveals nothing if published.
I also help client teams set up the same kind of environment for their own analysts, including model selection, hardware sizing, access controls, and the written policy a compliance team will ask for.
Tools I use and what each is trusted with
Every tool has a defined data tier. Model training is switched off on every account used for client work, and each client has its own workspace in each tool.
| Tool | Used for | Client information allowed |
|---|---|---|
| Local open-source models | Analysis of Restricted material, and all work for clients that elect local-only processing | All tiers, including MNPI |
| Hosted open-source models | Analysis of Restricted material that needs a frontier-level model, through trusted providers with compute primarily in the U.S., no logging or training on prompts or responses, and zero data retention | All tiers, including MNPI |
| Claude and ChatGPT | Analysis, modeling, drafting, document and slide production | Public and Client Confidential |
| Granola | Meeting notes, only with every participant’s consent | Public and Client Confidential |
| Google Gemini | Public research and general drafting | Public only |
| Microsoft 365 | File storage and email | All tiers, Restricted in a labeled folder |
Where AI helps a commercial organization today
The useful applications are specific and close to the data a commercial team already owns. These are the areas I work on with clients.
Market and payer intelligence
Literature and competitor synthesis, payer policy monitoring, and first drafts of market sizing that an analyst then checks against the data.
Forecasting and gross-to-net
Scenario building, assumption documentation, and variance explanations, with the model logic kept in tools finance can audit.
Incentive compensation and reporting
Plan modeling, payout quality checks, and automated performance reporting built on the commercial data warehouse.
Field and omnichannel
Targeting and call-planning support, pre-call briefs, and channel analytics, with promotional content kept inside the client’s medical, legal, and regulatory review.
Data infrastructure
Getting syndicated, specialty pharmacy, and claims data into a warehouse such as Snowflake in a shape that AI tools and BI dashboards can both use.
Governance and policy
Data tiers, approved tools, MNPI controls, and contract language, so legal and compliance can approve AI use instead of blocking it.
How I use it in client work
Research and synthesis
Literature, label and policy review, competitor reviews.
Analysis and modeling
Forecast shells, GTN scenarios, data checks.
Production
Board materials, slide decks, plans, and SOPs built from the analysis.
Meeting capture
Notes and action items, with consent at the start of every call.
Review standard
All AI output is treated as a draft. I review every deliverable and check figures, citations, and policy statements against the source before it reaches a client. Clients can ask which parts of a deliverable were AI-assisted, and I will tell them.
The same controls inside a client’s operating model
The same governance applies when AI-enabled workflows are designed into a client’s operating model, as they are in the AI-First and AI-Enhanced alternatives of the RF Launch Operating Model Framework.
Putting AI to work in your commercial team
Whether you need a governance policy your compliance team will sign, a secure environment for sensitive analysis, or a roadmap for the commercial function, start with a conversation.