52 Datasources and AI playbook for Reg Intel from RAPS Workshop

For those who attended the RAPS Global Regulatory Intelligence pre-conference workshop, “Locating and Analyzing Regulatory Precedent; Incorporating Research and Application Techniques in Strategic Regulatory Decision Making,” the key takeaway was clear: the era of manual, brute-force regulatory intelligence is over. The full-day, immersive session, led by experts from Temple University, Pfizer, Genmab, and BeOne Medicines, took attendees on a journey from the painstaking reality of manual research to the strategic imperative of AI adoption.

We were in the room, and we synthesized the day’s key learnings into a 5-phase playbook specifically for pharma and medical device regulatory intelligence leaders. This is the strategic roadmap that was laid out, from foundational principles to make informed decisions when using or buying AI solutions.

The AI Adoption Playbook (From the Workshop)

Phase 1: Acknowledge the Manual Pain Point

The workshop began by grounding everyone in a shared reality: the regulatory workload is exploding, but your time isn’t. The initial hands-on exercises were a powerful reminder of the tedium and inefficiency of manual precedent research. Spending hours with advanced search operators on FDA.gov, cross-referencing databases, and manually compiling tables is no longer a sustainable model for a strategic Regulatory Intelligence function.

This manual baseline served as the problem statement for the entire day: How can we move from being information janitors to strategic advisors?

Phase 2: Master the Prompt (The Human-AI Interface)

The first leap forward came with a deep dive into prompt engineering. The core message was that AI is not a search engine; it’s a tool that requires skillful operation. Vague, open-ended prompts lead to generic, unreliable results. The workshop provided a masterclass in structured prompting, introducing frameworks like ROLE (Role, Output, Limitations, Extra Context) to ensure outputs are targeted, accurate, and useful.

Role: I’m the regulatory intelligence specialist for [company] with focus on [devices / drug / indication / therapeutic areas]

Output: I need to write an executive summary of the new EU standards’ impact on product [product name] to senior leaders in meeting memo/5 slides/excell with bullet points/full sentences

Limitations: keep the tone high-level strategic in layman terms to make regulatory easy to understand, look at section [x] and leave out [y]

Extra Context: plug into extra datasources (NyquistAI Data API), upload latest competitor drug/device list

Note: most enterprise LLM such Microsoft Copilot, it is a fixed model trained with the past data, they normally don’t have the latest regulatory intelligence

If you are experimenting with personal access, please keep your information private.

The key principle: Clarity + Context + Control. By defining the AI’s role (“You are a regulatory policy analyst”), specifying the output format, and providing context, you transform the AI from a novelty into a force multiplier for your team’s expertise.

• Ask and Refine AI Proposed Steps: “How could an AI get from [X] to [Y]?”
• Confirm Understanding In Prompt: “Your Task is to do [X]. Do you understand?”
• Offer Context: “Do you need any additional information to do this effectively?”

Phase 3: Mitigate the Risks (Hallucinations & Human Oversight)

With the power of AI established, the facilitators delivered a critical warning: AI models are designed to be persuasive, not necessarily accurate. They can and do “hallucinate,” a persistent problem that even advanced training has not eliminated. The workshop presented a case study where an AI’s initial summary of a Health Authority review was shallow and missed critical context, demonstrating that without expert human oversight, AI-generated insights can be dangerously misleading.

Your playbook must therefore have a non-negotiable rule: AI handles the heavy lifting of information gathering, but humans must always perform the final analysis, verification, and decision-making.

Phase 4: Build Your Capability (The Build vs. Buy Decision)

The afternoon session pivoted from using public AI tools to building your own. The workshop detailed how to create Custom GPTs and in-house RAG (Retrieval-Augmented Generation) chatbots built on your own regulatory intelligence library. This revealed the true strategic choice facing Regulatory Intelligence leaders.

A successful AI project requires a team of four distinct experts:

• 1. The Domain Expert (who understands the problem)
• 2. The Data Engineer (who collects, moves, and harmonizes data)
• 3. The Data Scientist (who selects and trains the model)
• 4. The Deployment Expert (who designs the output and automation)
• 5. The Culture/Change Expert (who empower the adoption of AI - our NyquistAI take)

Most Regulatory Intelligence teams only have the first. This means you must either partner with internal IT to build this team from scratch or procure a purpose-built platform that has already made this investment.

When asked about the decision of “Build vs. Buy”, the experts from Temple University, Pfizer, Genmab, and BeOne Medicines said the development of AI is so fast, they would suggest to select and buy and always keep your options open to switch. One expert shared that the actual build of their internal RAG shouldn’t take that long but the internal stakeholder mapping, alignment and execution took more than 10 months.

Phase 5: Start with a “Quick Win”

The final, actionable advice was to avoid boiling the ocean. The path to successful AI adoption starts with a “quick win” project: one that is low in technical complexity but high in business impact.

The perfect example provided was building a RAG chatbot to search your internal regulatory library. It’s high-impact because it’s accessible to everyone and generates measurable usage metrics. It’s low-complexity _if_ you’ve done the foundational work of data prep. As one facilitator stressed, “Data labeling is where most initial AI projects get derailed.”

However, we understand that medical devices and IVD could be quite different given the data quality is different. The medical device indication for us and summaries are unstructured data which add additional challenges for companies to build their own AI tools.

What This Means for You

The workshop provided a clear mandate: Regulatory Intelligence leaders must evolve from managing information to managing systems of intelligence. The “build vs. buy” decision is now at the center of that strategic evolution.

At NyquistAI, we have served over 200 global organizations over six years building the platform that the workshop taught you how to design. Our life-sciences-trained AI, curated global data, and secure knowledge management tools are engineered to provide the capabilities of a custom-built intelligent system, without requiring you to become a data scientist.

If the playbook from the RAPS workshop resonates with the challenges you’re facing, we would be happy to show you how our platform provides a ready-made solution for each of these five phases. Book a no-obligation discovery call with our team to discuss your specific AI adoption roadmap.

We summarized a list of both free and paid Regulatory Intelligence datasources and tools for you to take advantage of.