Start where the documents pile up. Pick the workflow with high document volume and a human already verifying every output.
Run the boundary test. If an output enters a submission, batch record, safety file or trial record, it waits for the GxP gate: validation plus a compliance owner's sign-off.
Test "GxP-ready" on your own work. Believe the claim only after your reviewers grade the AI's output on your own documents.
Measure from day one. Baseline the workflow's current hours and track whether the output gets used.
A free lesson from AI in life sciences and pharma, one of its 10 lessons. Read another free lesson
Follow the paper
Dalila Masic, Regional Medical Director for a Type 1 diabetes team at Sanofi, found in a baseline survey of her team that over 50% of their week went to admin work. That is where the first AI prize sits: medical and regulatory writing, literature and data watch, safety and quality intake. Start your search in the work that eats your team's week.
Her ladder for medical affairs AI has five levels:
Consume
Create
Augment
Orchestrate
Transform
The first-workflow checklist
Volume and error cost pick the workflow. Masic's ladder for medical affairs AI runs consume, create, augment, orchestrate, transform. She says start at consume, summarizing a guideline update or the literature, or helping answer a medical information request. Virginia Chachati, a UK registered pharmacist and medical writer who founded the Write Clinic, says AI helps her structure documents and pull information together. Her limit: "ultimately you have to check it as a human." Pick the workflow with high document volume and a human already verifying every output.
Run the boundary test before go-live. Does the output enter a submission, a batch record, a safety file or a trial record? Then it waits for the GxP gate: validation plus a compliance owner's sign-off (lesson 6). Darshan Kulkarni, a principal attorney at the Kulkarni Law Firm, says courts have held a company responsible for what its AI says. In medical affairs and promotion, an AI that promotes the product or gives out problematic information is the company's liability, not the toolmaker's. Inputs count too. In a recent decision Kulkarni cites, the other side argued that running a client's thoughts through a free AI tool was itself a disclosure. In life sciences, he says, a disclosure can trip the patent on-sale or publication bar. He has seen trial sites upload patient data and sponsor IP into AI tools while ignoring legal advice. The harm surfaces years later, sometimes as an invalid patent. Run your test on inputs as well as outputs, inside the tool your company has approved.
Test a "GxP-ready" claim on your own work. Nataliya Andreychuk, who works in pharma content production, says big vendors now bring agentic AI to review promotional and medical materials. They claim to cut human effort on medical and regulatory review by 70%. Kulkarni runs AI on 50-state licensing reviews. The citations, he says, are often wrong but land "within the same zip code." Because the tool never gives him a complete answer, he still has to check 100% of the output. Before you sign, run it on your own documents. Let your reviewers grade the output. Course 7's lesson "vendor-real-ai-or-relabel" has the fuller test.
Measure from day one, and measure use. Baseline the hours the workflow takes now. Set one target you can count. Andreychuk says 77% of pharma's approved digital assets from 2025 were never used after medical-legal review. Count whether yours get used. Course 1's lesson "baseline-and-bar-before-pilot" has the method.
Where the pay sits
Stop
Where the pay actually sits
Bench
Elias Zerhouni, Vice Chairman and President of OPKO Health, says that in his company's antibody work, AI computes molecule configurations before any wet-lab experiment. Wet-lab experiments are slower and costlier, and Zerhouni puts the reduction at 50% to 70%.
Trial
Zerhouni says AI cannot predict a molecule's behavior in humans. The clinical trial remains the point of truth. Trial records wait for the gate (lesson 6).
Submission
First drafts and submission admin pay now; anything bound for the file waits for validation and the compliance owner's sign-off (lesson 6).
Plant
Quality intake pays first; the batch record waits for the gate (lesson 6).
Safety
Intake pays first; the safety file waits for the gate (lesson 6).
Market
Promotional drafting pays only with experts in the loop. Kausar Riaz Ahmed, Senior Director, Head of AI, Precision Oncology at AstraZeneca, says why. An unsupervised model drafting promotional content could drift toward efficacy and away from adverse effects, unbalancing the required benefit-risk profile.
Exercise (20 minutes): fill in the first-workflow card
Pick one workflow you are considering for AI, the one with the most documents and a person already checking every output, such as the literature watch or the safety intake queue. Fill in the card for it; you won't know every line yourself. Every line you can't fill gets an owner and a date before you close the card. The card download includes a filled-in example of a workflow that isn't ready yet, and why.
First AI workflow card (PDF)
A blank card and a filled-in example, one page each.
Show the three prompts: fill the card, test it, turn it into a plan
Use AI to fill in and test your card
These three prompts go into your company's approved AI assistant. Use the first to fill in the card one row at a time; it won't guess a number for you. Use the second when the card is done, to hear the case against starting yet. The third asks about your part of life sciences and your workflow, and turns the lesson into a 30-day plan, with a checklist, rubric or brief for your boss if you want one. Use placeholders for company, system and people names; the help doesn't depend on them.
Prompt 1: Fill in my card with me
Paste this, then add your workflow in one or two sentences and the card below.
Prompt
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X) and round any figures. The help you get does not depend on them.
You are helping me fill in a first-workflow card for one AI workflow in a life-sciences company. The card is pasted at the end.
Rules:
- Ask me about one row at a time, in order, and wait for my answer before moving on.
- Write down only what I tell you. Never suggest, estimate or invent a number, a name, a yes/no answer or a decision for me, even if I ask you to guess. If I ask, say that this row needs a real answer and ask who could give it.
- If I don't know an answer, write "Not known" and ask who should find out and by when. Put that in the Owner / by when column.
- After the last row, show the whole card as a table with three columns: Question, Answer, Owner / by when. Then list every row marked "Not known" with its owner, and remind me that any unanswered row means the workflow is not ready to start.
My workflow in one or two sentences: [DESCRIBE YOUR WORKFLOW]
The card:
[PASTE THE CARD]
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
Prompt 2: Stress-test my card
Paste this, then your filled card.
Prompt
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X) and round any figures. The review does not depend on them.
Below is a first-workflow card I filled in for one AI workflow. Act as a skeptical reviewer at the decision meeting and argue the case for NOT YET.
For each row:
- Say whether the answer is specific enough to act on (a number, a named role, a date) or too vague.
- Point out where answers contradict each other, for example a workflow marked as staying clear of regulated records against an output or an input that crosses the boundary.
- Say what evidence you would ask for before accepting it, especially on the boundary test and a vendor's "GxP-ready" claim.
Do not rewrite my answers and do not invent facts about my workflow. If a row is blank or says "Not known", treat it as not ready.
Finish with: your recommendation (GO FIRST, NOT YET or RE-TEST), the three rows that most need work, and one question I should ask my team on Monday.
My card:
[PASTE YOUR FILLED CARD]
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
To paste the card into prompt 1, copy its rows from the PDF.
Prompt 3: Make it yours
Paste this and answer its four questions.
Prompt
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X, Supplier B).
You are helping a leader in a life-sciences company apply one lesson to their own situation. The lesson's four decisions are below. Use only them and what I tell you. Never invent facts, numbers, vendors, laws or results about my organization or my industry. Where a number is needed and I have not given one, say what to measure and who should measure it, instead of guessing.
First, ask me these four questions, one at a time, and wait for each answer:
1. My part of life sciences (for example medical affairs, regulatory, quality, safety or commercial), and roughly how many people work in my organization.
2. The workflow I have in mind for AI, in one or two sentences.
3. Where it stands today: an idea, a pilot running, or about to go live.
4. What worries me most about it.
Then give me the following. Keep it short and practical:
A. My plan. For each of the four decisions: one action written for my workflow, the role that should own it (a role, not a name), and what should be done within 30 days. If my part of the industry makes a decision harder, for example a regulated record, promotional review or patient data, say which decision and why in one sentence.
B. Three questions to ask my team on Monday, specific to my workflow.
C. Ask me which one I want: a checklist for my workflow, a rubric to grade the output on our own documents, or a one-page brief for my boss. Then write it, and finish with a prompt I can paste later to update it.
Mark anything that rests on an assumption with "(check)".
The four decisions:
1. Start where the documents pile up: pick the workflow with high document volume and a human already verifying every output.
2. Run the boundary test: if an output enters a submission, batch record, safety file or trial record, it waits for the GxP gate: validation plus a compliance owner's sign-off.
3. Test "GxP-ready" on your own work: believe the claim only after your reviewers grade the AI's output on your own documents.
4. Measure from day one: baseline the workflow's current hours and track whether the output gets used.
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
Your company's assistant may answer differently, so read what it writes before you use it.
Check yourself
One budget line, two candidates: literature summaries a reviewer already checks, and first drafts of submission sections. Which goes first?
Show the answer
The summaries, once the boundary test clears them. A human verifies every output; the submission drafts wait for the gate.
A vendor says its reviewing AI is "GxP-ready" and cuts human effort on medical and regulatory review by 70%. What must happen first?
Show the answer
Run it on your own documents; let your reviewers grade the output.
A trial site wants to upload patient data to a public AI assistant. What do you do?
Show the answer
Stop it, and keep the work in the tool your company has approved. The harm can surface years later.
Sources
Dalila Masic, Regional Medical Director, Type 1 diabetes team, Sanofi, on OnPoint: The Medical Excellence Podcast (2026): "25. Dalila Masic on the Five Levels of AI for Medical Affairs Teams". https://exa.ai/library/podcast/68mvd6dq02s/episode/zfjsgjs3vkp
Kausar Riaz Ahmed, Senior Director, Head of AI, Precision Oncology, AstraZeneca, on From Spaghetti to Growth (2026): "Former FDA Regulator & Pharma AI Exec REVEALS: 'Patients Trust AI More Than Doctors'". https://exa.ai/library/podcast/627g89dmxwj/episode/gljs6b9cn27 Produced by Disruptive AI.
Your CEO wants the first AI project to answer customer calls because a competitor launched one. Which question actually picks the first workflow?
A quality team wants an AI to draft intake notes that will be attached to the batch record. What must happen before go-live?
A licensing-review tool impresses your team in a demo, but its citations are often slightly off. What must happen before you sign?
An AI now drafts your team's responses to medical information requests, and reviewers approve them quickly. Before calling the pilot a success, what do you count?
Where AI pays first in life sciences, and where it must wait
Ask about this lesson, or how it applies to your team. Every answer cites the lessons it draws on.
Try asking
Free preview: 3 questions a day, answered from this lesson. Follow-ups remember your earlier questions on this visit. Don't paste confidential details.