2026-09-15-0822-bill-morgano-ai-continuous-validation-life-sciences

William (Bill) Gargano on AI and continuous validation in life sciences

Recorded 2026-09-15 at 0822 via Plaud. 42:15.

Summary

A conference session that opens with housekeeping from an unnamed host (Wi-Fi credentials, exhibitor bingo cards, lunch-and-learns at 10th Terrace, upcoming events in Puerto Rico, Amsterdam, and San Diego) and then becomes a lecture by William (Bill) Gargano of RCM on AI in life-sciences computer system validation. Gargano’s framing statistic: at a top-five client, employees using AI more than eight hours a day grew from 600 in 2024 to 40,000 in 2025 to nearly 110,000 through July 2026, and he says AI now drives 95 percent of his own office day. His core claim is that AI is a force multiplier, never a replacement: it extrapolates from data outside your company, so blindly applying its output is the user’s failure, and since it is easier to edit than create, the winners will be the people who get efficient at directing the edit. On CSA, he calls it a 40-year-old predicate rule and says implementing CSA on paper without changing the underlying process is a waste of time; validation has to shift from a one-time event to a continuous process, the way manufacturing moved to continuous manufacturing 20 years ago.

The central case study is a financial-services software vendor entering life sciences: RCM had them integrate Caterpillar and Cucumber for self-testing and regression-based testing and move to nightly releases of a GMP product; the setup has passed three FDA inspections with provable 99.9 percent test coverage, and an AI analysis layer added on top predicts failure points and where customers will configure the product. Gargano’s mantra is that validation is not a document-based process but a data-based one; he mocks the practice of wrapping vendor documentation after a two-day, three-person review of a three-million-line product, and notes Oracle IQ certificates are auto-generated templates with about twelve variables plugged in. The FDA’s number one focus is data integrity, specifically trend analysis across deviations and failures, followed by what the data says the organization should improve, which is exactly the work validation staff should be redeployed into Quality to do. He reads a real FDA warning letter citing AI agents used to generate specs, procedures, and master production control records without Quality Unit review and calls it a process failure rather than an AI failure, and explains AI drift as model performance changing as the model keeps learning. Numbers cited: about 80 percent of top-100 pharma use AI in validation; the largest pharma spent 3.8 billion dollars on its own LLM, reports about 60 percent confidence in its results, and laid off 60 percent of its validation team; a client with 27 global sites saves 2.4 million dollars per site per year after a 3 million dollar investment with zero headcount reduction; RCM’s standing offer is a 65 percent cost cut and throughput time more than halved within 18 months. He closes with the Claude em-dash tell and the trick of running Claude output through Copilot with “clean it up,” the tractor as the tool that displaced the most people in history, his wife’s 360-dollar AI toaster, and the advice to aim AI at the least consistent, most consultant-heavy process area, then have QA review what the engine produces.

Plaud AI notes

App-generated, verbatim from Plaud’s Summary tab.

Date: 2026-09-15 Location: [Insert Location] Instructor: William (Bill) Gargano

Summary

William (Bill) Gargano from RCM presents on the transformative role of Artificial Intelligence (AI) in the life sciences industry, focusing on its impact on Computer System Validation (CSV). The session opens with conference logistics covering Wi-Fi, lunch sessions, and upcoming events. Bill argues AI is not a job replacement but a force multiplier that enhances human capabilities and efficiency. He calls for a paradigm shift from static, document-heavy CSV to a dynamic, data-driven, continuous validation approach. Key themes include rapid AI adoption, “continuous validation,” and the FDA’s heightened focus on data integrity and trend analysis. Used correctly, AI can significantly reduce validation costs and timelines, freeing skilled personnel for higher-value quality work such as analyzing data trends. He cautions against unreviewed use of AI outputs, citing an FDA warning where AI-generated specs and procedures lacked Quality Unit oversight. The core message: AI should accelerate human expertise and help people perform at a higher level—not replace them.

Knowledge Points

1. Conference Logistics and Opening Remarks

• General Information
  • Wi-Fi network: “Redbird Guest”; password: “2025 event.”
  • Attendees should visit the exhibitor area, get bingo cards signed for a prize drawing on September 16, 2026, and stop by the membership table.
• Lunch and Learn Sessions
  • Held on the 10th floor at the restaurant 10th Terrace; look for two reserved tables.
• Post-Event Communication
  • End-of-day surveys and a presentation compendium will be emailed. The compendium will be fully updated by the end of the week (week of September 14, 2026) as presentations are received.
• Upcoming Events
  • Busy schedule ahead with events in Puerto Rico, Amsterdam, and San Diego.

2. The Role and Adoption of AI

• AI as an Enhancer, Not a Replacement
  • AI’s purpose is to unlock unused potential and make people more effective. It provides a foundation that’s easier to refine than creating from scratch.
  • While technology evolution retires certain roles (e.g., telephone operators), the speaker believes AI elevates human performance rather than replacing people.
• Rapid AI Adoption in Business
  • At a top-five client, daily heavy AI users (>8 hours/day) rose from 600 in 2024 to 40,000 in 2025, and nearly 110,000 by July 2026—an exponential uptick.
  • The speaker uses AI to summarize the prior day and plan the current day, driving 95% of his daily office activity.
• Understanding and Using AI Output
  • AI extrapolates from large information sets to recommend actions; applying outputs without human review is a critical mistake.
  • Always read and review AI-generated content before use. Each AI engine (Claude, ChatGPT, Copilot) has distinctive patterns (e.g., Claude’s em dashes) that can reveal its use.

3. The Paradigm Shift in Validation

• From Event-Based CSV to Continuous Validation
  • Treating CSV as a one-time event is inefficient. Implementing CSA without changing the underlying process wastes effort.
  • The future is a continuous model, akin to manufacturing’s shift to continuous production two decades ago.
  • Integrating development, testing, and validation into one ongoing process enables nightly releases demonstrably tested to 99.9%—a standard that has passed FDA inspection three times.
• Validation as a Data-Driven Process
  • Validation must shift from document-based to data-based. Knowing your data is fundamental to adopting AI.
  • The move from “prove-it-on-paper” (CSV) to “prove-it-works” (CSA/AI era) prioritizes critical thinking and risk-based testing depth, with continuous monitoring.
• FDA Focus on Data Integrity and Trend Analysis
  • Data integrity is the FDA’s top priority; auditors specifically look for trend analyses across deviations and failures.
  • The next question is what the data reveals about improvement opportunities. Quality teams should use this to drive corrective actions and enhancements.
• Real-World AI Citation
  • The FDA issued a warning letter for “inappropriate use of artificial intelligence in pharmaceutical manufacturing.”
  • The issue was not technical AI failure but the use of AI agents to create specs, procedures, and master production records without adequate Quality Unit review— a process failure.
• AI Drift
  • AI drift is performance change over time as models learn from new data. As learning continues, outputs can degrade or differ from prior responses.

4. Financial and Operational Impact of AI

• Return on Investment (ROI)
  • Automating validation (CSV, equipment, process) can cut costs by 65% and reduce throughput time by more than half within 18 months.
  • One client with 27 global sites invested about $3millioninAIforvalidationandnowsaves$2.4 million per site annually, achieving an 18-month site-level payback without staff reductions.
  • 42% of pharma AI initiatives miss ROI targets, typically due to poor implementation strategy—not technology—especially when Quality and Compliance aren’t engaged early.
• Resource Reallocation
  • Efficiency gains need not drive headcount cuts. Redeploy talented staff from repetitive validation tasks to Quality functions to analyze trends and drive improvements.
• Identifying Areas for AI Implementation
  • Start where variability and inconsistency are highest or where consultant outputs vary widely. Use AI to standardize data and streamline processes.

Questions

• [Insert Question/Confusion]

Assignments

• 1. Review internal validation processes to identify inefficiencies and opportunities to shift from event-based to continuous, data-driven validation.
• 2. Identify the least consistent areas (e.g., deviation management, root cause analysis) as initial targets for AI implementation.
• 3. Involve Quality and Compliance at the outset of AI tool selection and implementation planning.
• 4. Establish a formal review and verification process for all AI-generated content and outputs prior to publication or use.

Related Photos

Transcript

[00:14] Speaker 1: Great. Good morning, everyone. Thank you all for being here. I’m going to make this quick so we can go ahead and jump in and get started. Quick thank you to all of our sponsors for being here today. Everybody, make sure they stop in the exhibitor area, get your bingo cards. We’ll have some great prizes to wrap off tomorrow, so make sure you get all those signatures. A couple of quick updates: the Wi-Fi password, so the network is under Redbird Guest, and the password is 2025 event. We’ll have a couple of networking breaks today, so those are good times to make sure you’re mingling, get your bingo card signed, take a moment, get some coffee. We will have some great luncheon learnings. So we’ll have some lunch and learns today. They will take place on the 10th floor in the restaurant called 10th Terrace. When you get to the lobby, hop in the elevator. It’s very nice up there. Great view. We’ll have some good lunch. There will be two tables reserved for the lunch and learn sessions. Just keep an eye out for those if that’s something you’d like to join. Also keep an eye on your emails. We’ll have the end of day surveys and the compendium sent out. Please be patient with the compendium. We have to update it as we get the PowerPoint presentations, so it will be fully updated by the end of the week. You can click on the session titles in order to access the files. Make sure you give Ryan a quick visit over at our membership table. We’re going to be in that corner. We do a little high top over there, but we’ve got some fun hats to check out, as well as some other fun things. Come stop by, get some more information about that. And then here’s just a brief look at some other things we have going on for this year. We are hitting our busy point of the year, so there’s lots of things to choose from. We’ll be in Puerto Rico, we’ll be in Amsterdam, we’ll be back in San Diego. So it’s going to be a busy rest of the year. And that is my spiel. If anybody has any questions about anything, please come visit us at the registration table. But with that, I will turn it over to William (Bill) Gargano from RCM.

[03:06] Speaker 2: Do I need to use a microphone? Can everybody hear me? Use the microphone. The one thing I’ve never been told is that people can’t hear me when I talk. Good morning, and we’re here to talk about basketball. Connection is unstable. That’s AI. There you go. Anybody ever hear me speak before? The vendors now. So I always start off with this one. In 2024, I actually did this conference, and there were about 200 validation people in the room. And the first thing I said is, “Go out.” In 2025, there were only about 25 validation people in the room because they all want a new job. Not today. The reality, though, is our landscape was changing virtually every day. I can tell you from a fact, and this actual set of numbers. This is a very large client, top five. In 2024, they actually kept metrics of this business going across their network. 600 people in their company used AI more than eight hours a day. In 2025, that number went up to 40,000. Through July of 2026, that number is close to 110,000. So when people say we don’t really use AI, I will just ask this: When’s the last time you logged into AI? Today, how many of you logged in this morning? I get woken up by AI, and I’m not exaggerating. I actually have my clock come on in the morning and summarize my day yesterday, and it tells me in my activity what I need to do to complete, what’s open, and what I need to do today. It literally drives 95% of my activity in the day-to-day office. It has accelerated my ability. Just yell it. So anyway, it literally drives about 95% of what I do. And what it does is make me more effective. So when we talk about anything in life sciences, anything—I don’t care what it is—are we looking to replace people? No. What we’re looking to do is harness what we don’t use. How do we make ourselves more effective? How many articles are there a day? AI is going to replace this. Going to replace that. Going to do this. Going to do that. Going to replace this. That’s evolution. Let’s be real. How many people still remember telephone operators? Anybody ever remember the old origins? Have to take a wire plug in. I’m that old. That actually remember that. Some of you are probably going. What the hell is he talking about? Realistically, we used to do that. When’s the last time you talked to your stopper? There’s certain activities that just don’t happen anymore. So think about this. When’s the last time that you were comfortable enough to allow someone else to. Dictate the root cause of your deviation that was not in your company. Anybody ever do that? Of course not. That would be probably one of the most insincere statements I could make. Especially if you’re in QA, everybody would be looking at me going, “What are you talking about?” And I apologize for my language from New Jersey, so we always talk. But realistically, that is exactly what AI does. AI is not in your company. What’s AI doing? AI is going out, looking at past performance, readily available information that’s sitting in some database somewhere that’s accessible. It’s extrapolating. It’s determining. It’s going out and saying, based upon the twenty-seven thousand different things that I’m doing here, this is what you should do. That’s basically what AI does at the end of the day. I’ve simplified it a whole lot. So for those of you that are the AI gurus in the room, yes, don’t chastise me later for just oversimplifying AI into one sentence. But realistically, that’s exactly what happened. If you take the answer and you just blindly apply it, shame on you. Because that’s the false of AI. It does not replace people. It enhances people. It gives you a foundation. How many people have ever said this statement? It’s easier to edit than create. Ever use that anywhere? Then what the heck is AI? It’s edit, not create. That’s what it does. So those that get more efficient with how to create the foundation of the edit are the ones that will succeed. My kids are, as you can tell, I don’t talk to the slides a whole lot. Not one slide’s worth it. So Brittany said, “Oh my God, you have a lot of slides.” I said, “Don’t worry about it. I don’t talk to them anyway.” My kids are 27 and 22. One’s a lawyer; the other one’s a theoretical physicist. Don’t ask me what the heck he does. I have no idea. But what was wild is that we’re in the car, and I ask him a question, and on my screen pops up this message, and it’s my kid from the back seat answering me on a test. And I said, “What’s the matter? You can’t talk to me?” Oh, it’s faster for me to type. I said, “Really?” He goes, “Yeah, I don’t have to look at you. I just kind of…” I said, “You have a mouth, don’t you?” Well, I don’t like to talk. So what does he do? He talks to AI every single day, probably ten to twelve hours a day. And the stuff that he comes up with is mind blowing to me, absolutely mind blowing. He’s working on his PhD now to combat mine in time distortion, and we argue about it all the time. And some of the stuff that he’s come up with, I said, “You didn’t even study that. Where’d that come from?” He says, “Oh, I was on Facebook.” I was on this. I was on that. I took this. I did this. And I said, okay. And I said, now how do you prove that? He goes, you want me to prove it too? I said, well, that’s isn’t that the human element? So when we say harness the power, this is about harnessing the human power, not AI. I do not ever believe that AI will replace humans. I think it will take us to another level. I think it will push us beyond what we even know is feasible today, because we can’t think in the abstract as well as AI can. The things that we want to talk about. Let’s see if I get this to work. Lindsay, how do I get this to work? Or Randy, how do I get this to work? This is not working. There we go. I got it. We can sit here and talk. How many people here are using CSA every day in your company? Not that many. Wow. How many are planning to go to it? How many don’t care? First off, I. CSA has been around for. Forty years. Just nobody knew that. It’s a predicate rule. It’s based on exactly everything we’ve had before. Here’s the big difference. If you just implement CSA on its own on paper, don’t bother. Absolute waste of time. How many people want to kill me right now? Especially the folks that helped write a CSD. It’s a waste of time. Why? There isn’t enough game. Everybody will talk about it. It’ll say this. It’ll do this. But here’s the one thing I will tell you: the entire paradigm of validation and the way that we think has to change. Five years from now, we won’t do it. Mark my words. We will not do it. The words will change the way we process, the way we do things. Completely different. Why? Validation CSA will trigger an event. Buy an application, a piece of software. What do you got to put in your plans? Twenty, thirty percent for validation documentation. Not. And management sits back and goes, “Well? We’re late for this, for that. We’re going to cut this out. We’re going to change this. We’re not going to do this.” And like every other place I’ve ever consulted at, Bill, how can we still get passed by the FDA, but we only have? Instead of giving you six months, we’re going to give you six weeks. So my comment is. Why are we treating it like an event? Manufacturing, and there’s a slide later somewhere in this deck that said 20 years ago, manufacturing completely changed the way they did business. How? They went to continuous manufacturing. So why do we think validation in a single event, as opposed to continuous? That’s what AI is going to afford us. We worked last year with a software vendor. The software vendor had never been in the life sciences market before. They had only worked with financial services. They came to us and said, “We’re selling this product now to the life sciences.” And they came to us and asked us if we were a GXP. We all looked at them and said, “What the hell are you talking about?” We don’t know what that is. So how do we get there? They came to us and said, “Can you help us get pre-validated?” They actually had the right words. Somebody had educated them, so they came to us and said, “Hey, how can you get pre-validated?” The really beauty part of the thing was, we said, “We can’t do it.” A consulting firm that actually said they can’t do something, which is not usually right. Consultants always say yes to everything. Anyway, we told them we can’t do it. We said, “Here’s what we can do for you. We’re going to change the way you function. We’re going to change the way you operate. We’re going to integrate your development, your testing, and your validation all in one process.” We had them integrate Caterpillar, Cucumber, a couple other things into their toolset. For self-testing and regression-based testing, we got them to think about continuous development. Instead of releasing the product once a month or once a quarter, they release it every night. Every single night, they release an upgrade. You’re the cod consumer, and it’s a GMP solution. Every single one of you here is going. How the hell do you validate that? How do you do that? I can tell you for a fact that it’s passed the FDA three times, not once, not twice, three times. Because we can prove that we test the software ninety-nine point nine percent. Of anything that’s done. Now I’ll bet you a dollar. I’ll bet you more than that. That takes SAP. Take I don’t care what product is. I shouldn’t pick on products. That’s not the point. None of them can prove that they’ve tested the software at that level of detail, and they can show you proven results that are there. That’s continuous validation. You layer on top of that AI analysis, which we did, to be able to demonstrate potential failure points or where the customer is more likely to configure the product. It changes the entire landscape of how you look at validation. One of our partners is here. I’m going to give a shout out to Brian Dennis, who’s sitting in the corner. Brian is the founder of a company called Square, and one of the things, and I’ll let him talk about the product later and all that kind of stuff. But there is one thing that I took from Brian from the day one that I met him and we started working together with their product was again, and I think the reason we hit it off is because there’s one single statement. Validation isn’t an event. Validation is data. It’s not a document-based process anymore. It’s a data-based process now. So you have to think in data concepts as opposed to thinking about I got to produce this document. So when you look at AI and anybody, I know Randy will sit here and Ray is the head of our delivery practice at RCM, but he also happens to be our data guru. And the very first thing that he will tell you, if you want to move it into AI, is you have to know your data. Data is your foundation. So when you start looking at all of these regulations, when you start looking at whatever the process is going to be, how you’re going to do with it, it changes your paradigm. About what you do with it, it is a new way to think. It is a new way to work. It is a new way in which you are going to find results. People think that CSA does one really powerful thing for us. It allows us to use vendor documentation more consistently within our organization to cut down the amount of time that we need to validate. If the vendor has tested it and it looks good, and we go through and we review it, we’re going to put a wrapper around it and we’re going to say it’s validated. Anybody ever do that? How much time did you spend reviewing the documentation? A day? Two days, three people looked at it, signed off on it, and said, ‘Hey, this is okay.’ The product only took three years to develop. There’s only about three million lines of code, and there’s only this. But you reviewed all the documentation in two days. Is there something wrong with this picture? Does that logically make any sense? They’re X, Y, Z. Everybody owns it. It must be okay. Oracle. Anybody go on site to qualify Oracle as a vendor anymore? If you even propose that, you get fired. They send you a packet. Here’s the packet. Who even opens it and looks at it? Wait a minute, come on, somebody beat, somebody fess up. Anybody ever open it? No. I’m telling you, I’ve watched it. You did? Cool. Did you find any errors in it? Wasn’t a blank document. I appreciate the honesty. That was really cool. That was great. I couldn’t answer that one better. That was really cool. Here’s the thing. They’re big, and it must be okay. So you get an IQ from Oracle because they installed it. Anybody ever read the certificate to actually see what they actually did to find out what they really exercised? You’d be amazed if you actually read it in detail because it’s all AI generated. It’s an automatic template that comes out. They plug in like twelve variables and it spits it out. Because I can tell you, I can take it to the next client, and it’s exactly the same. They change the numbers to protect the innocent. So when we think about this, what’s the real value of what we’re doing here? If you need to look at what data you are ultimately producing, I’ve done a lot of research trying to figure out quantifiable numbers for people. There are. We’ll reverse this for a second. When we walk into a client and we say to them, “Hey, we’re going to automate your validation piece.” So when we say that, we’re talking about CSV, we’re talking about equipment quality, we’re talking about process value. So we take all three of them dynamically at the same time. We tell a client within eighteen months we’ll cut your costs by sixty-five percent. We’ll reduce your throughput time by more than half. That’s a minimum. We’ll almost guarantee that out of the gate. And the client will look at us and go, “How many heads is that?” And I said, “None.” And they go what? I said, well, the people that are doing the work are they really bright? Yeah. I said, do you have other work that they can do? Yeah. And they said, your quality department is already cut to the bones and it’s not big enough. So guess what? Take the really bright people, redeploy them, and put them in quality because that’s where you need to spend the time. So it’s interesting when you start thinking about what things could cost. Here’s the other thing. We built a process that automates. That’s awesome. What it doesn’t do is anticipate. What else should it be doing? Who needs to prompt that? What’s the number one thing that the FDA looks for today? Data integrity. Go one step further because you’re on the right path. What’s the exact words that they look for now in data integrity? Trend analysis across the board on deviations and failures. It’s the number one thing that they look for in every single company across the board. The second thing they ask is what anybody knows the second question. The second question is this: what is the data telling you? They don’t ask it that way. They’re looking for you to be able to explain to them if you’ve done your trend analysis. What is that information telling you about your organization and what you should be focusing on? To improve, correct, fix, eradicate, or whatever words you want to use. Who does that? Quality. So can we shorten the CSV cycle? We can put 50 slides up here and show you 20 or two different products. Some good, some of them bad. All of them say. Every vendor out there has AI in their slot, on their ban. Every single vendor. Two years ago, one had. Every single vendor now has it. Every single QMS product. You guys all have different ones, I’m sure. Which one of them doesn’t have AI now? They all say they’re AI. They didn’t rearchitect their product, but they’re all AI based. Magic. Hell, that happened, but whatever—it doesn’t matter. But they’re all there. Why? We slap some engine on the top of it, and all magically, I can go ahead and say, “Hey, guess what? Rewrite this for me, so it doesn’t sound stupid.” Do you ever send out an email anymore and don’t run it through Copilot or ChatGPT? Hey, rewrite this so it’s formal, so it sounds good, so I don’t sound like a moron writing to my boss. Try that one time. It’s really cool what it comes back with. I actually wrote that one day and said, “I don’t want to sound like a moron,” and it said, “You already are.” So I’m waiting for it to say “more off than on,” but that’s okay. I thought that would have been a better response, but anyway, there’s been one. Actual citation, that was published on AI, and it’s really not. It’s kind of half baked when you read it. So it actually says we can’t read it in the back. This client was in a company, and it came back, and they said AI agents were used to generate specs, procedures, master production, control records, without quality unit review. That’s not AI. That’s just stupidity. So you write it down. You don’t have quality. Look at it. That is AI’s problem. Let me ask you this. Why didn’t the FDA come in and say? The AI drift associated with this, or the hallucinations that it was generating, was X Y Z A B C D E F G because it actually understood what it was doing. Let me assure you, the FDA couldn’t even begin to address that. I used to work for the FDA. It’s okay. But anyway, regardless of that, the reality is that they can’t go to that level yet because they don’t understand what it means. Anybody know what AI drift is? And no, you do. Anybody know what it is? It’s a changing performance over time. So it changes performance over time, right? So if you, as your model learns, it degrades. It actually does, right? It actually falters, and people say, “Well, it doesn’t give me the same answer.” Of course not, right? I don’t know what the metrics are anymore. But they say that the internet doubles what every what six and a half seconds or whatever the heck it does. The data that’s out there, whatever it does, it’s a lot. So it’s got access to all this stuff. How the heck is the answer going to be the same as it was six minutes ago? It’s impossible. So unless you have a defined LLM somewhere, right, and even within that, then you set artificial boundaries. Sorry, I’m using the word to define itself. Of how it’s going to interpret that data. What’s interesting about this? There’s nothing wrong with the first part of this statement. There’s actually nothing wrong with this whole thing. Just look at the data. Anytime the AI you use it, if you don’t read it before you publish it, you should get fired. I turn to all my salespeople on AI, and my recruits, and we sit down and we try to train them all the time about do this, and this, and it is amazing to me how many things I get back that I go, did you actually read this thing before you sent it to me? I did. I said, really? And when you look at it, what’s the very first thing when you use? Claude, that will be the initial thing. As soon as you see it in the output, what does Claude always do in its output?

[31:57] Speaker 1: You’re absolutely right.

[32:00] Speaker 2: It puts in what? That’s a long dash. M dash explanations. M dash all over the place. As soon as you get one of those, you know exactly where it came from. They either use sonnet or use fable, right? The lower level ones don’t do that, right? So every engine has a tickler. ChatGPT has one. Copilot has one. You just need to know which one. And as soon as you see it, you’re like, okay, you generated it this way without modifying. So you know what you do to trick it. I’m going to give you guys a trick. If you generate Claude, take the output and say run it through Copilot. Right after that, say clean it up. Guess what? It fixes it. Doesn’t put the new stuff in, and it actually looks like you wrote it. Then send it to you. Before you do that, so I’m going to just jump to a couple things. This is not theoretical. This is real. These are real numbers. Today, there are about eighty percent of large farms, top one hundred, that use some form of AI in their validation process. The two largest pharma’s actually built their own. The largest one spent 3.8 billion dollars just on their LLN alone. 3.8 billion. And they will only tell you that they’re about 60% confident of its results. They, the same company has also laid off sixty percent of their validation team. That’s real. That is a real number, written down and it’s fact is published. If you optimize, there’s two point four million that is on screen there. That actually comes out of a client by site that has about 27 sites across the world. They’re saving about 2.4 million. They spent on their initial input into validation when they were going into AI. They spent about three million dollars. That return is in 18 months by site. Think they had a pretty good return on investment. Ask me how much of that was reduction in staff. Zero. The big fallacy about AI, and it is a fallacy, management will tell you we’re going to cut heads. I wish I could change the rhetoric, but I can’t. The reality is, though, that just think about your own self. About how you might use a tool. Anybody know the greatest tool that displaced the largest number of people in human history? Anybody know what the tool was? Before electronics, the tractor. The tractor changed the entire landscape of how the world worked. It displaced hundreds of thousands, and forgive the language I’m going to use. Slaves, hundreds of thousands of people. Could you even conceptualize working on a farm without a complement? Of course not. People farm today for what used to take hundreds of people with automation. Why is that a bad thing? You adapt to a changing landscape. Instead of validating for months on end and getting nowhere with slow results, sit back and say, I can use the tool. I’m going to validate it in days. What is wrong with that? Or more importantly, I can ingest the change I got sent from the software vendor, and it’ll dynamically update my trace matrix and automatically go through and test it for me, producing the results I’m now going to review. That it’s complete. Why is that a bad thing? You mean to tell me you have nothing else to do in your jobs? If it actually did that and it was reasonably accurate, and you actually reviewed it, you couldn’t show the value to your company and to your boss and everybody else that it is that says. Hey, I’ve expedited this. Give me five more. I’m the one who’s going to maintain this process. I’m the one who’s going to do this. This is the way I’m going to think. What is wrong with that? So I’m going to just go. There’s a whole lot in here, and I apologize, but I am going to jump to one really thing, and then I’m going to wrap up here. AI. I’ll leave you with three pieces of information. Most important element of AI. What is it? How do you define it within your organization? Because it is being used like the panacea of God knows what. Mike, since only a couple of you heard me talking here, I had a laugh about a year ago. This is old. Sorry that it’s not new. My wife brought home a new toaster. It was three hundred and sixty dollars. Bro, I’m gonna stop. My wife told me she spent three hundred and sixty dollars on a toaster. Almost fell on the floor. It had this really great little symbol on it. It says, ‘I’m AI.’ And I said, doesn’t it toast bread? I said, no, this is great. What are you talking about? She said, you can connect your phone to it. I said, I hate to say this. I sit right here, and the toaster’s right here. I said, I’m gonna connect my phone. Why? She said, well, it’ll tell you when it’s done. Doesn’t it pop up when it’s finished? She said, oh no! It’s even better now. You can monitor how well it’s toasting. Okay. Is there still the dial on the toaster? Oh no, no, no! It’s computerized. It’s got numbers and all kinds of stuff, and it auto adjusts. What do you mean it auto adjusts? Based on the thickness of the bread and this and that and everything. I swear to God, who told you this thing? She said, he was a really nice guy in the store. Huh. Thank God the toaster broke. Because we don’t have it anymore. I won’t tell her I broke it, but we broke it. Here’s the thing: not everything has to be in that kind of paradigm. You have to understand where does it fit within your organization. So when you’re talking about validation, you’re talking about anything associated with this. The first question I would ask is this: do you look at deviation? Do you look at root cause? Do I look at this? Do I cap? Do I look at capital? Do I go through with this? Look at the area that is the least consistent in your organization. That has the most number of variables. The most number of consultants. You all use consultants. You use consultants. So here’s what you do: they all write different. Everybody gives you a different answer. Take all their data, run it through an engine, and determine what’s consistent. And get rid of your consultants because you don’t need them. Yes, I just undersold my sales team, just going freaking nuts right now. But you don’t need it because you can use this. And you can simplify. And then what do you do? You have somebody from QA actually review the data, that knows what they’re talking about, or the engineer, because maybe it’s on the shop floor, and actually review the information. So you pick up AI, you look at the data, and you understand where that’s going to go. That’s how you harness the power. You harness it through the use of a tool to accelerate the knowledge base of the human. That’s what it is. So if you take anything away, think of it as. Another element to make you even better. That’s what it’s about. I want to thank you all for your time. I appreciate it. I hope you all have a great show.

Related

• AI Validation and Acceptance Criteria MOC
• Home
• Validation is data, not an event - the continuous-validation thesis

---

← Start | Home | Adoption panel →