رونوشت 🔗
Shubhendu Tripathi: Every AI success stories you have heard was edited. This is the unedited version. I'm Shibhindu. I put AI into the systems companies actually run on. So thribe to the integration layer on YouTube, Spotify, or Apple.
Shubhendu: everyone is talking about AI. Almost nobody is telling the truth about it. Ma here's a question almost nobody in your company can answer. When an AI writes something and a human approves it in four seconds, who is actually accountable? Not an org chart. When does it go wrong? My guest today sits in the seat where this question has teeth. He's not an engineer and he will tell you that himself. He works in commercialization, he launches. in patient support programs. The operational side, where AI either changes how the work actually happens or quietly does not. Daniel Banbupwala, welcome to the integration there.
Daanyaal Bandukwala: Thanks, Shu. Looking forward to it.
Shubhendu: One thing before we start, because you deserve to know it, Daniel and I work together commercially. He uses a platform my team builds. I'm telling you that up front, and it is exactly why I'm going to be pushing him harder today than I would push a stranger. So let me start with the rude version. You're not an engineer, Daniel. Why should anyone take AI advice from you?
Daanyaal Bandukwala: Thanks Shub. So not not rude at all. I think it's a very valid question. so I've been at the forefront of commercializations in pharmaceuticals for many years. And I've seen things prior to AI, whether it be new technology or new processes, come into play, and I've seen it fall flat when the implementation doesn't truly work well. So my job as a consultant, as a opinion leader in this space really has to do with implementation, right? And actually looking at things from the lens of how does it work specifically for the pharmaceutical healthcare biotech space. AI as a whole, great concept, lots of interesting avenues. But in healthcare pharmaceuticals biotech, we live in a very regulated environment. So appropriate use, governance of these tools is equally important. and sure, there's some great ways to make your day to day much more efficient, which I appreciate, I use, but using them in a governed manner, and in a manner that truly appreciates how the organization will move forward is key.
Shubhendu: And and actually there is a number that backs you up, Daniel. Boston Consulting Group looked at what separates the companies that scale AI from the ones that do not. About I I think I was reading it this morning, about 70% of the difference is people and processes. So that you mentioned about governance. So that that that's included there. And about 20% is technology. Only about 10% is the algorithm itself. We're saying that Boston Consulting Group sells AI transformation consulting, so they have a stake. Although this particular finding cuts against the thing consultancy usually want to sell people. So if let's say 90% of this is not the technology, I want to start with the part nobody can define. You told me governance was the big one. let's do it properly then. so that's our first segment. we are entering it right now, Daniel. So my question was governance is a word everybody used to mean nothing. What does it mean when it is real? When it is real governance, when you talk about patient support programs or in pharma or healthcare or anything in general?
Daanyaal Bandukwala: Yeah, for sure. So in in the space, you know, we're we're dealing with many different stakeholders, whether it be patients and patient data, physicians and physician perceptions, and there's a lot of moving parts, to be quite honest, right? So there's a few ways to really keep governance top of mind here. and I usually do a little bit of a a test in my head. in terms of You know, when you're putting something into an AI agent or creating an AI process in itself, if this is something that would turn up on the news on the Globe and Mail tomorrow, how would I feel about this? How would I look? And y what would my and my company's reputation look like in that front? Right? so as I was mentioning before, Shub, it's very easy nowadays to put certain things into basic AI agents like Whether it's just ChatGPT or Claude or whatever the case, and get some really good content and good information. Get a deck done really quickly or whatever the case. But the information you're putting in there, would an external party be okay with that? Would they appreciate that you're putting that information into that specific agent? that is governance at its simplest, right? I think beyond that. What I'm seeing in the pharmaceutical space is many companies, now that they know their their staff, their teams, are adopting AI in some cases or looking to adopt AI. They're feeling the pressure of adding this layer of governance in. Right? There's certain tools that they're making available to them to use internally in a much more secure manner, which is great. That's one thing. But those are not always the newest and the greatest tools, right? So Well, a lot of corporate organizations use Copilot as an example, and everyone uses it to some extent. But there might be s certain versions of other AI agents which aren't necessarily available internally. So what happens? Employees use Copilot at work, they have their phone on them, and they enter in certain information that might be considered proprietary, confidential into their personal device. So how How do you deal with situations like that, right? that's a whole other large question that many organizations are starting to deal with now.
Shubhendu: That's that's awesome. But let let me double well get back to that topic in a bit. But one question I was thinking is now let's say assume I run a mid sized company. I could be any company in in pharma space, healthcare space, or in general. What do I do differently on a Monday? like in your opinion when it comes to AI.
Daanyaal Bandukwala:
Load full transcript
Specifically from a governance perspective or generally? Specifically from
Shubhendu: Yes, governance.
Daanyaal Bandukwala: a governance perspective? Yeah. So I think one big thing that I always say is what does your onboarding plan look like from a AI perspective? Staff comes and goes, say Monday morning you have a new team that's starting within your organization, new sales team, let's call it, right? what does your onboarding for AI specific tools look like? for for that t team member and that that group of people. in many cases up until recently, and some of you might not believe this, but up until recently it's just been a very small mention in some large SOP that says, hey, this this you know, as an organization, we we are trying to experiment or use AI or we use AI, these are some key points to keep in mind, right? I would go a little bit further, you know, make it really black and white. What cases are specific cases that should not go in AI, should should stay within a different ecosystem. what are, you know, compliance, those types of functions looking at specifically to make it very blatant and obvious, right? the reality is, and this is a concept in adult learning, you have to hear it, see it multiple times to really remember it. And nothing
Shubhendu: Yeah.
Daanyaal Bandukwala: is different from a AI governance perspective.
Shubhendu: Yeah, and and and it's it's very rightly said, and I think that's such a valuable insight. What we forget is because both you Daniel and myself and my my team, we work with a lot of customers about this. I came across an article recently, actually it's a research report by Grant Horton, and they asked like nine hundred and fifty business leaders across various sectors. A very simple question. Could you pass an independent AI governance audit within 90 days? Like, do you know how many people we're not confident?
Daanyaal Bandukwala: Yeah, I I'm not surprised. It it's
Shubhendu: 78%. We're
Daanyaal Bandukwala: Yeah.
Shubhendu: not confident that they they will pass that. And that's that to me is a big, big red flag to what what where we are as like as an industry in general, whether whichever industry we talk about, it this is like across 10 different sectors. Among the ones which were still in pilot stages, it was seven percent. So imagine so many companies have gone beyond the pilot stage, yet more than 70% of companies are not confident. for in
Daanyaal Bandukwala: That's pretty good.
Shubhendu: 0.5 leaders say they could not pass the audit. Tip that that's the number it comes down to. What do you think? in your mind, how does that show? of where we are in this whole AI frenzy. I know there is a whole MBA concept about the laggers, the early adopters, but wha what do you think where we are at, especially when we talk about pharma healthcare industry?
Daanyaal Bandukwala: Yeah, so I mean, you know, w w with that result, I'm not necessarily surprised. a and I say that because to your point, Shu organizations and it it's not only small and medium organizations, you know, you sometimes usually see that across industry. Small and medium organizations might be a little bit slower to adopt in some cases or might be much more agile in other cases, right? I what I'm seeing in the pharma healthcare space, there's actually a a wide spectrum of adoption and
Shubhendu: Mm-hmm.
Daanyaal Bandukwala: whatnot. And, you know, we we ran a workshop a little while back, a couple of months back, with manufacturers from various sizes, you know, really big pharma to small, medium sized pharma. And it was surprising to see how they've adopted. some of them have really taken AI completely internally. And are not using a vendor-based approach. they're really building everything internally. They're hiring people specifically on the AI function side. And then there's others who are very baby steps. I've I've used Copilot to edit some of my emails, right? And and that's where they're at. So I I'm I think you know, over the next year or so, there's gonna be a a much huger shift in it's really directly related to the fact that we have so many new tools literally coming out on a day-to-day basis. companies will will have to adopt. They'll be in a spot where there is no other choice. And we're starting to see this with agency partners as well. pharmaceutical healthcare companies rely a lot on creative agencies and things like that. Some creative agencies to be honest have adjusted their model of working and have you know really put AI enabled workflows into place which has led to efficiencies in some cases from a cost perspective but also from just a content generation perspective right they can produce content, review content much more efficiently. Other agencies have stuck with a much more traditional approach. And I I think the hammer is coming down a little bit on how that will evolve and change.
Shubhendu: Yeah, it's it's it's we weird times we we live in, we work in right now. I'm like there are so many reasons to be excited about, but yet there are so many reasons to sometimes again, this is not my word. A lot of my previous guests have also said this. There are so many reasons to be careful or also discouraged about because again, based on the other report I was talking to you, another survey came up I think it this was like last week or something. I was just I have a I have my agent who scouts these news for me to create like some blogs around that. I think if I'm not wrong, but I'm 100% sure it was some 90% sure it was like 300 C-suite executives at companies over half an 500 million, half a billion
Daanyaal Bandukwala: Mm-hmm.
Shubhendu: of revenue, were interviewed, and 88% of them said governance or risks or ethic reviews had killed or delayed a financially promising AI project. So I see if every day there are companies who are like, we do not have a use case.
Daanyaal Bandukwala: Yeah.
Shubhendu: Now we here we're talking about half a billion in revenue
companies who have a use case who have been seeing the financial ROI, but because of the governance or risk or ethics, they have delayed the project. in the same survey, I think there was a mention about I I now this number I forget, but it was over 70% admitted that they cut the quality assurances tips.
Daanyaal Bandukwala: Mm.
Shubhendu: in their AI project to hit their AI goal or deadlines. Now, this escapes my mind. Like, why is this becoming a race? Yes, what you talk about, governance
Daanyaal Bandukwala: Yeah, yeah.
Shubhendu: is important. And so that's an one extreme. There is also an extreme about okay, let's try what are the problems we can solve. This is a new technology. There are plenty of things we can do. And here we are, we have a deadline. Because we committed to like 10 use cases, we're gonna cut the QA. like out of the project and so that we can meet the deadlines and what happens then the project will eventually fail their their moments will hallucinate or something like that will happen or there will be like a governance issue and then they everybody will blame ai so sale people so governance is killing their project and they're cutting corners to ship it it
Daanyaal Bandukwala: Yeah it's
Shubhendu: it's it's it's both funny and not funny at the same time what do you think is a real problem like is governance killing AI projects or initiatives or there on the other hand, there are people who are trying to cut corners. Are they the problems? Because they are feeding back to the loop that AI doesn't work.
Daanyaal Bandukwala: So it's a great question and I don't think it's governance. I think governance is something that we're used to, right? It working in such a highly regulated industry. Governance is just a concept that's embedded in pharmaceutical and healthcare professionals, right? I think what's tricky is what you mentioned before, Shoop, specifically with these AI goals, right? And I always kind of say that we're not using AI for just the sake of using AI. It's not just for a KPI score code. Right. Reality is, you know, over the last two years, you've also probably been to a number of conferences, I have as well. In many cases, we just saw, you know, concepts and lots of conceptual content on how AI can be used in pharma and healthcare. and you know, I I think a lot of leaders took that. As and it mixed with pressure from global counterparts to say, hey, we gotta implement. We're gonna be too late if we don't implement, right? Someone's gonna beat us to the to the race, hypothetically. And it has landed on someone's scorecard that, hey, by the end of this year, you need to implement X amount of AI initiatives. And the reality is that's not sustainable, that's not the right way of looking at it. The true embedded way of looking at it is What issues do we have currently outside of AI? Like what what challenges are we facing within our organization? And then layering on top, what can we build out, whether it be a workflow, agent workflow, whatever the case, to deal with those issues to assist with it could be something as simple as efficiency, cost, whatever the case. But it it shouldn't be a top-down approach of Hey, we just need to implement AI. So let's see what we can do, right? that's a very backwards way of doing doing it.
Shubhendu: I think having spent 15 years in in ERP, I think it's just a habit of of the masses sometimes that blame the governance. Nothing works, blame the governance. It's it's been the safe answer. And sometimes it's true, and again, to a large extent it is true. But in this case, what's happening, especially in the case of AI adoption, nobody's putting in a survey that, did you assign an owner? Ever and do you did not put in your survey that, hey, did did you gauge the interest of people? Are they like up to speed in what you're doing? No. They'll just simply blame the governance way. No, so well, great answer, that that makes a lot of sense.
Daanyaal Bandukwala: The other point should quickly I'll just add to that from a governance standpoint is when you're looking at these new initiatives that you're rolling out, ideally you're you're building them for the future, right? It's not again, it's not going back to okay, we've hit the governance checkpoint, we check off our our scorecard and we're good. It needs to be built out in a way where for years to come, we can continue to update this system, feed this system, whatever it is, and have it live on. So you can truly see the impact of of it, right? It it shouldn't just be, okay, we hit the scorecard, the the first six months went well and then everything goes bust because it the governance is has not been built out for that.
Shubhendu: Yeah. And again, just to back our claims, what we're talking here, and that you know I don't remember all the reports, so here's another research that explains what's going on. SP Global SP Global surveyed over a thousand IT and business leaders. Only 21% had mature governance for autonomous AI systems. And the analyst described the issue right now. So they said what ex what actually is happening is compliance and legal has been asking that who has the approval authority.
Daanyaal Bandukwala: Mm.
Shubhendu: And where the audit trailer is, and they are not getting the answers. And people blame the governance and the project gets shelved because they did not have the answers. But yeah, this is from the report. If someone wants to go read in
Daanyaal Bandukwala: Right.
Shubhendu: details, Voice of the Enterprise, it was published in October 2025. Very
Daanyaal Bandukwala: Yeah. Interesting.
Shubhendu: interesting. So, compliance is not blocking you, not blocking anyone of us. Compliance is asking who is accountable, and nobody wrote it down. And again. I think it is not 100% accurate based in the pharma industry because they have they're the ones who have really good compliance and and everything is in general when we talk on a like a 10,000-foot
view,
Daanyaal Bandukwala: Right.
Shubhendu: like people do not have those. Like who is the role owner, who is the account owner, who is this process owner? And I think that's where people should start looking at. Which takes me to the thing I actually wanted to ask you about. walk me through what a human loop looks like in practice. Not another diagram, on a Tuesday afternoon.
Daanyaal Bandukwala: Yeah. So let's take for example a a pharmaceutical trade manager who's responsible for distribution management, relationship management with various specialty pharmacies, wholesalers, three PLs. so sitting down, on on first thing they're looking at in the morning is their dashboards, right? They wanna understand what's come out. From an order perspective, the last day, what needs to be fulfilled and movement and flow of where things are going, which layers on to trend analysis, right? is there a certain product that is at risk of out of stock? is there a certain product which is trending much lower than it did over the last quarter or over the last year? Is there something I need to discuss with my commercial marketing sales counterparts and understand
Shubhendu: Yeah.
Daanyaal Bandukwala: what's going on there? Are there sudden spikes from a specific customer? And if there are, we can't just, you know, report up to the executive team and say, hey, great news, we have more sales. The first question is gonna be why? Where is this coming from? Right? What is it? Was it unanticipated demand? What's what's going on there, right? So That is kind of all from a a general flow perspective, you'll you'll continuously do those activities. But then my favorite was when I was in pharma and in this type of role, I used to get a tap on the shoulder. We were in office at this time, from one of my VPs. He used to say, Hey, can you by like five PM have a complete overview of what's been going on over the last three months with these six accounts? And all my reports were generally pretty fragmented. I had a Excel sheet from everyone. And while we used to attempt to put together all of her reports on a monthly basis, a a lot of the heavy work fell on me. So it would always be a hustle bustle to get everything together and not just get everything together in a a plain Excel sheet. I had to keep in mind the audience, right? So this is a VP, he's gonna present this to his leadership team. consisting of the CEO and various other VPs that sit on that leadership team. So it needed to be in language format that was easily translatable, easily understandable, and something that, you know, he could present in maybe fifteen minutes and the message would get across. Right. So as I mentioned, you can't just slap slap a huge Excel sheet with a bunch of data on that. You
Shubhendu: Yeah.
Daanyaal Bandukwala: needed executive level insights. And At a time that was the bean of my existence, that was a a lot of work I used to find, but now there's definitely better ways to do this.
Shubhendu: Yes, that's that's a perfect example of how AGTI workflows. And again, if people think properly, they can I always say this to everybody who's looking to implement AI is break the problem down. Just like you you just mentioned it, it's very like clear to me that okay, step one, step two, collecting data from my fragmented access sheet, step three, this. And again, final state is creating a presentation for your supervisor or your director so senior leaders in your company. And once you have done that, that's like 20 to 30 percent of the work because that's where people get it wrong. a lot of people think it's a magic bullet. I'm gonna give AI
Daanyaal Bandukwala: Yeah.
Shubhendu: and it will do it for me. And I'll tell it, just prepare a presentation. But it needs to know the context. And I
Daanyaal Bandukwala: Exactly.
Shubhendu: think that's what that's what you guys have been trying to build when you have been working on the trade fusion concept that break the problem down. What is it that again the context also involves who is who is the user is it the train manager
Daanyaal Bandukwala: Yep.
Shubhendu: or is it like a medical advisor or of you know a a salesperson
Daanyaal Bandukwala: Exactly.
Shubhendu: or a marketing manager etc
Daanyaal Bandukwala: Exactly.
Shubhendu: so that that's really exciting what you you guys have been building and yeah looking looking forward to know knowing more about that now
Daanyaal Bandukwala: Definitely, yeah. More to come.
Shubhendu: in more to come and I wanted to share study again and Please do give me your comments if it's too much of information that should be researching
Daanyaal Bandukwala: Okay.
Shubhendu: for both to Daniel and my gu my listeners as well. But I want to keep like things honest here, the actual facts, right? So this study changed the way I think about all of this. Researchers took 27 radiologists, from trainees to very experienced specialists, and had them read mammograms with an AI assistant. Sometimes the AI was wrong. When it was right, accuracy was like, I think 80% across the board. When the AI was wrong, accuracy for the least experienced readers fell to just under 20%. For the most experienced, it fell to about 45%. Now these are trained specialists whose entire job is to look carefully. If experts drop to 20% when the machine is wrong, What is your human in the loop protecting you from in those perspectives?
Daanyaal Bandukwala: Yeah, it's it's really interesting. So I I I think there's been a number of studies like this shoot that have come out I think over the last say two years. and there's more that are underway, that's for sure. I think, you know, people get a little bit edgy about this specifically, right? And AI being used in medicine and whatever the case. No, that's me. I
Shubhendu: I don't blame them, honestly, but yeah. Yeah.
Daanyaal Bandukwala: don't either. I've had a few discussions with physicians, some that are very much for it, and others that think
it is not the way to go, and it'll lead to more problems down the line. The ones that are for it, ultimately their thinking is similar to our thinking in our business processes, right? If AI can do the first draft or something for me, and then I can verify it. then I'm able to see more patients in my clinic, my efficiency gets better, capacity for the healthcare system gets better, and we all win. I think the trouble that they have, especially the physicians that aren't for it, are they don't believe everyone, all physicians will actually adopt it in that manner. they believe that th there will be a select group and it just takes a select group for things to go wrong, who will use it in a mass use way where they feel that their agents or whatever the case are well trained and they don't really need to look over the results. And the relationship between a physician and a patient is truly something that can't be completely replaced. Yes, there might
Shubhendu: Yes.
Daanyaal Bandukwala: be pieces like this which make things more efficient, but there's a high level of trust between a patient and their physician.
Shubhendu: Yeah, well where and that's great. Where I'm trying to get at Daya is if I am the ps expert who is leveraging AI
Daanyaal Bandukwala: Mm-hmm.
Shubhendu: and I I am like you're my customer or you're my patient or something, I think I should still be held accountable. Not
Daanyaal Bandukwala: One hundred
Shubhendu: the machine. It cannot be the machine. It I at least as of where we are today, I know they talk about the AGI, it's coming, it's here. Think Jensen from Nvidia said that Asia is already here. Anyways, but if someone is using AI, I think they should be held accountable. I think that's a good solution for for where things are. And and this is something I thought I invented something, so I researched and it's not something new. it's aviation psychology. named this in the 90s and like 90s, solved it in 2000. Skipka
Daanyaal Bandukwala: Okay.
Shubhendu: atmosphere. they they are they were the ones who read the study and the people who were genuinely accountable for the decision made fewer automation errors. Because people who were
Daanyaal Bandukwala: Yes.
Shubhendu: not accountable just deferred to the machine.
Daanyaal Bandukwala: Yeah.
Shubhendu: Even when the evidence in front of them contradicted it. And that is twenty like twenty-six-year-old. It's it's foundational, not stale. and I think that is the point. This was answered before most AI governance policies were written.
Daanyaal Bandukwala: It's true.
Shubhendu: Yeah, accountability changes the behavior. And I think nothing else in that literature does. Why is every AI policy about processing said? What do you think? So I think that that brings up to my next question. if you agree to what I said right now that accountability still stays with the person who's running the AI, why is or I'm trying to brainstorm with you here. Why are all the AI policies focused on the process? I'm not the person who's using it. Whether it is in concept of trade manager, whether it's can in concept of teleaviation psychology, or we build a process around initial triaging patients, for example. we we always focus on the process and to an extent it's correct. But the policy should should also focus on on who is actually gonna be working on that or using that. Well
Daanyaal Bandukwala: Yeah.
Shubhendu: it does.
Daanyaal Bandukwala: Yeah, definitely some thoughts. So a hundred percent agree with you on the accountability piece. I think that's that's key. And that was I think well relayed in in the physician example that I gave too. where where those who were for it and you know using AI and AI agents cross the board, ultimately the the final decision still came to them, right? So yes, if there was a patient issue or something was missed. They couldn't necessarily blame it on their AI agent. They would have to take accountability for that, right? So on your piece around policies and how they're being built out, I think it's a very valid point. From what I've seen recently, there has been a little bit more of a shift to how even in in small wording and how some of the wording is built out on some of these AI policies, and the you know, some real language around the use of responsible AI and things like that. in things that are simple, employee handbooks, you know, I've seen it in the recruitment space recently as well. when job postings go up, they they put a statement on there that really says that, hey, by the way, this company uses AI for screening and things like that, but we still on the back end have someone looking at applications to ensure that we're not missing candidates and things like that. So I think it is shifting a little bit, but there's more work to be done in that space, that's for
Shubhendu: Yeah, quickly changing topics. this is not directly related, but maybe we'll give some ideas to you. When when the a lot of system social media tools started labeling the the blogs or labeling the posts as AI generated, there was a survey which was done. I I I don't remember the specifics about this, but this is like f from the quarter first of this year. It did not change how much persuasive or like the effect it the blog or that content had on people. So even after telling them that this is AI generated, people still were persuaded by the product. So that's in interesting. So what I thought was it it since it's not changing in your mind now if you talk about the job market which you mentioned, if you talk about hey, we have a product in Pharma or PSP solution that uses AI. I think it's it's gonna be similar in in some maybe the percentage will be different. Now in the disclosure policy for AI.
Daanyaal Bandukwala: Mm-hmm.
Shubhendu: Why do we have to why do we even have to mention that? And this is something which I'm I'm curious about. I I do not have enough experience to understand, but why does it ha need to be on the
disclosure policies that we have gener used this or we're using AI in some parts of the process.
Daanyaal Bandukwala: So I can speak to some of my experience in the medical education space, and some of the work we've done in the past. So I think it's truly just transparency. sure, you know, I'm not a lawyer, just like I'm not an engineer, I'm not a lawyer. And there's probably
Shubhendu: Yeah.
Daanyaal Bandukwala: some legalities and compliance pieces to it as well. But it's truly in in the face of transparency. So through Inc. one of our companies, we produce a lot of medical education content, podcasts, things like that, that are AI generated. And we've had clients come to us and say, Hey, we wanna send this out to a group of physicians, let's say, to capture a new study that's come out. And our rule always is, that's fine. you know, Shub, you know how good some of this content usually sounds nowadays. It doesn't seem like AI. and I
Shubhendu: Yeah.
Daanyaal Bandukwala: think that's part of the reason of the disclosure. Maybe a year and a half, two years ago, when something was AI, you could tell it was AI. Right? and now we're getting to a point where maybe this whole p podcast is AI. Yeah, yeah, fair enough. It's not,
Shubhendu: It is not yeah.
Daanyaal Bandukwala: it's not. But what I'm getting at is it's become so good that You know, you don't want to be in a position where you're deceiving someone, and I think a lot of companies have just taken the upfront approach of let's be transparent and whatnot. And that's some of the same concepts we've had. and you also have to be careful with when, you know, if you're using someone's likeness or something like that, or their voice or their face or whatever the case, it's very clearly cut that, hey, this isn't them actually. This is you know, their their AI agent or whatever the case. I think that's common in our industry, but I'll give you another example. I was scrolling on Instagram the other day, as I do sometimes, and I came across on my Explorer page some influencer promoting some product, right? And I started looking at the page and you know, because I I'm on a lot of AI stuff, I I can still kind of figure it out when something is AI. But this was such a you know close depiction of a real influencer, someone who is, you know, doing their day in the life, promoting a certain product, whatever the case. And you know, they had millions of followers on Instagram. And it doesn't say anywhere that they are an AI based influencer. So to me, you know The promotional abilities, they have sponsorship deals with some of these companies, I'm sure.
Shubhendu: Yeah.
Daanyaal Bandukwala: and, you know, they're they're influencing. They didn't influence me to buy anything, unfortunately, but they're influencing other people, I'm sure, on specific products. And it's hitting the algorithm on Instagram, right? So it's very targeted targeted to specific people. so I think the way we market, the way we promote,
is going to be a total, total game changer with this.
Shubhendu: Yeah, it's so I mean what essentially I understand from what you're trying to say is as a consumer, now the consumer could be like as patients or someone who's receiving any kind of services or products. We should be made aware of what we are being served as as content and then we can be the best judge.
Daanyaal Bandukwala: Yes.
Shubhendu: Right. But now we if you talk about specifically sensitive industries so again going back to healthcare again if let's say and I have I'm gonna try to be very plain and simple and not ranting about any of my past experiences I'm gonna be talking about the healthcare in Canada so if I call virtual medical service for an appointment etc
Daanyaal Bandukwala: Yep.
Shubhendu: how How do we define that? What is the escalation to a human? Like, how do we define that okay, this this request should be or could be solved? Because I I also understand from their point of view, from by them, I mean the other party, that there are some redundant requests or questions, etc. How do we kind of deal with that?
Daanyaal Bandukwala: Yeah, and we were talking a little bit about this before. and again, not getting into too many details about current state of of healthcare in Canada and whatnot. That's not that's not for this podcast. but I think, you know, it it's very known, well known that wherever we can find efficiencies in the healthcare system to get more patients seen by the right specialists and the right people is important and sh should be done where possible. Now, the tricky kind of balance between that is exactly what you're getting to shoot, which is at what point is an AI agent potentially not enough to answer a specific patient's questions. you know, I think during COVID before all of this AI truly bubbled up, we saw a lot of a spike in virtual medicine, right? virtual calls, things like that. Yeah.
Shubhendu: Mm. Yeah, I remember those days, yeah.
Daanyaal Bandukwala: So I think, you know, that has continued on in some cases in some ways. and what has been layered on top is okay, before you even speak to a physician or a nurse practitioner or a nurse or whatever the cave pharmacist, let's do a set of screening questions. To ensure that they have the right information and can cue you up accordingly. Now, in theory, that's a great idea, right? Because that means ultimately when they actually get you on the phone, they have your chart pulled up, they have all the right information, they know what the prognosis is, at least at a high level, and can give you a treatment plan and next steps within minutes in a very ideal situation. The reality though, is ultimately Some of those questions patients actually need to talk through with their provider to really understand what's happening. They're not always all black and white. And if you have an AI agent on the other side of that, the way it captures it and documents it might end up being a waste of time, quite
honestly, and being double work. So going back to some of the earlier points we made, when something like that is deployed. actually having you know a a very comprehensive training and real world use case perspective for those models is like at the utmost importance
Shubhendu: Yeah.
Daanyaal Bandukwala: while layering in the governance piece, right? We need to you know, and shoop you have lots of experience with this, from like a data control perspective and information perspective. what the patient is actually comfortable with sharing with that AI agent and where is that information actually going. Right? That's where I think patients are still a little bit iffy about it, because they're not comfortable with sharing as much over what they seem like is a automated system, versus what they would share with their physician in person, let's say.
Shubhendu: Yeah, and and on that note, it's it's a very, very sensitive point. And not even just the customers, Daniel. I go to events, I I talk and I see other vendors talk about AI, I I see other vendors talk about the data resiliency and and all that requirement about the data, how is is it being used? Sometimes it it it it bothers me so much that even those vendors who are promoting the products
Daanyaal Bandukwala: Yeah.
Shubhendu: Do not understand the distinction. And that is very, very, very concerning to me because again, not just from the perspective of data privacy, but also I'll give you an example. we often talk about, how is your data being used? Where is it stored? That's called data resiliency.
Daanyaal Bandukwala: Right.
Shubhendu: And where is the model stored or where is the model running?
Daanyaal Bandukwala: Frame.
Shubhendu: So that is the model resiliency. There's also a concept of sovereignty right now. So whether the data is you might have the data stored on, for example, an Azure or AWS data center in
Daanyaal Bandukwala: Right.
Shubhendu: in Toronto. That's still not sovereign because why? Because the company which governs that is a US company.
Daanyaal Bandukwala: They're not based here, yeah, that's right.
Shubhendu: Yes. And when I see vendors in the in various industries say that they are sovereign and unlike Dude, you don't even have the basic distinction on like what is data sovereignty, what is data resiliency. So that I think to all the listeners, when you are consuming some any any using AI, any AI model, know what is what, know where is your how is your data being used. I know these tools have incognito mode, etc.
Daanyaal Bandukwala: Yeah.
Shubhendu: Yet if the government wants, and again, this is just one use case, should a entity. I'll just say entity instead of government. If an entity wants access to your data, there might be ways they will be able to get your data without you even understanding that, hey, how did they these
Daanyaal Bandukwala: Yeah, for sure.
Shubhendu: end up happening? And and I think one of the things which people should understand very carefully, like back in the days when we were sharing our emails everywhere and then we were on these newsletters or these mailing list people, random person trying to sell us something. And it used to bother all of us. I think AI is now starting to get into that territory.
Daanyaal Bandukwala: That's for sure. That's for sure. And I I think Shub, it's such a good point with regards to the definitions of these different pieces, right? And the fact that, you know, sometimes the people you think should know these in and out don't necessarily know it. We've come across some of the same scenarios as well. And what we tell our clients in our consulting practice is when you're looking to partner with a vendor. These are some of the key questions you need to ask them. Whether you're doing a full-blown RFP or just a capabilities presentation, first of all, they should be coming prepared to discuss those questions as it is. But dig
Shubhendu: Yeah.
Daanyaal Bandukwala: a little bit deeper so you truly know these pieces. Because once you're into that, it's not that it's impossible to get out of, but it's challenging. Every transition is challenging.
Shubhendu: And and and this time it's not just your email or phone number being tracked, right? Now it's time is how you talk.
Daanyaal Bandukwala: Yeah.
Shubhendu: What's your background? What's your accent like? Like I I see a lot of people use like the ch the voiced mode, which
Daanyaal Bandukwala: Yes.
Shubhendu: which is cool. I use it too, but does do you guys know that if you if I someone is able to get a hold of a little bit of more like that piece of information, they can create like a voice note as if it was coming from you and do anything basically
Daanyaal Bandukwala: If you literally saying anything and it literally I don't know, correct me if I'm wrong, but thirty to forty five seconds max is all you need of that to to do that,
Shubhendu: Yeah. Yes.
Daanyaal Bandukwala: right?
Shubhendu: Yeah, yeah. So again, I know we I went a little sidetracked, but this is something very important and it is relevant. And it is relevant in in whether you talk about AI from a governance perspective or AI any any other format. But okay,
Daanyaal Bandukwala: Cross the board.
Shubhendu: let's let's get back to our our our spine of the conversation. There is also a case which is worth knowing and talking about in Australia. a Royal Commission found that from 2016 in a government debt recovery scheme, human review was removed until the debt notices were being issued without anyone looking at them. Now about 400,000 like audits, something like that, people were wrongly issued unlawful debts.
Daanyaal Bandukwala: Wow.
Shubhendu: 400,000. That's like the yeah a population of a town
Daanyaal Bandukwala: Let's eight. Yeah.
Shubhendu: in in Canada sometimes, or more than that, four four
Daanyaal Bandukwala: Yeah.
Shubhendu: towns. And I want to be fair about this: that this was rules-based. This was rules-based data
matching, not AI as we know it today, but the
Daanyaal Bandukwala: Right.
Shubhendu: failure is exactly the one we are describing in general that we have talked about right now. The review step was still on the process map. It had just stopped happening. Now, Daniela, how would you know from your point of view if that was happening in any of your organizations we work with right now. How would you how what's your way to kind of find that out if you have to?
Daanyaal Bandukwala: So first of all, I mean, I'm not completely surprised that it is happening. I think complacency, right? the system's working. we've seen the first few samples, it seems to be issuing it right. Let's take a backseat, right? What we encourage a lot of our organizations to do, whether it comes to this type of work or it's on, you know, their creative partners or patient support program partners. You know, most of the contracts that they put together have a formal audit clause in them. But
Shubhendu: Mm-hmm.
Daanyaal Bandukwala: doing a formal audit, if I'm being very honest, is a lot of work. It should be done periodically at certain points. But at a minimum, it doesn't hurt to on occasion, maybe every quarter, every six months, do a spot check. And a spot check isn't as invasive or crazy as an audit, but you'd be surprised with what you find when you do these spot checks. whether it's, you know, on a specific vendor that you're using. And, you know, that that way it gives the vendor an opportunity to have a very transparent dialogue if they've missed something and to pursue corrective action. But at the same time If the vendor is not as open to the spot check, it tells you right away there should be lots of red flags going off on, hey, what's actually going on here? Is there something we should be looking closer at? So across the board, whether it's a AI initiative or any contract that one of our clients is getting into, we encourage you know, ensuring that a spot check is in place.
Shubhendu: Nice. Yeah, so but like if AI writes something for you and you take a couple of seconds to approve it, like the question is like who is accountable and I mean figure out a good way to say it. But how do we say this? Like who
Daanyaal Bandukwala: Yeah.
Shubhendu: gets fired when the agent k gets wrong is what people should be noting down and people should be thinking about.
Daanyaal Bandukwala: That's that's what I was kind of referring to prior as well, right? When
Shubhendu: Yeah.
Daanyaal Bandukwala: I mentioned the case of what if this ends up on the front page of the Globe and Mail or on on a news story or something like that, right? That's kind of the mindset that needs to be ingrained into Teams, right? Would we put our own name behind this? Would we be willing to truly risk our career, our job over this? Or Should we kind of shift the accountability to like we were saying before? Yes, let's use this. We want to be efficient. We want to make use of AI in this specific situation.
But ultimately, I need to be the one who looks over this to ensure that I feel comfortable enough to sign off. And
Shubhendu: Yeah.
Daanyaal Bandukwala: especially in specific situations, some of the examples you gave, it's more and more apparent definitely in the healthcare and pharmaceutical space as well.
Shubhendu: It's I sometimes envy people working in your industry. And here's the reason. you guys have unfair advantage on on the question I'm about to ask. Like in January, I think it was last year or January or February, the it was winter time, the FDA published its first draft guidance of using AI in regulatory decisions. how do you use AI in making decisions? And in in in the document. they mentioned that they have received more than 500 drug and biologic submissions with
Daanyaal Bandukwala: Mm-hmm.
Shubhendu: AI components since I want you to hear this up 2016.
Daanyaal Bandukwala: Wow.
Shubhendu: Like and again, listeners and viewers, no, we're not talking about chat GPT. This is like before AI was cool. So Daniel, your regulator has already had to answer who's accountable
Daanyaal Bandukwala: Yeah.
Shubhendu: and not any other industry. so I
Daanyaal Bandukwala: Yeah, it's true.
Shubhendu: yeah.
Daanyaal Bandukwala: It's true. It's it's interesting, right? And I think one of the most first and this definitely proves it, one of the most first and earliest use cases of AI, not necessarily in its form like how it is today, like you were saying Shub in just very early, early stages was regulatory and drug discovery, right? Drug discovery is another huge one in terms of finding new pathways for treatment. companies that are looking at expanding their pipeline and figuring out what's the next product that we can bring into play, whether it's licensing in from like a business development perspective or truly from a scientific perspective, right? What pathways have not been studied yet and it is something that we could make novel and patent and could have a true use case, right? Could be backed up with data. So that that side I think is one side in pharma from an AI perspective that has been ahead and pushing forward. But I would say it's it's at the global level, right? So when I say it's at the global level, it's the it's the head office that may be using this in some of their workflows and whatnot. In many cases it hasn't been pushed down to the local organizations.
Shubhendu: Yeah, the way I see it it's pharma is being early on this. and every industry is about to get get its own version of a regulator asking for the signed off now. And most of them I don't think they have any answer already. I don't know and they're struggling, they're figuring it out. Now I have to be straight with you, Daniel. I went looking for real data on this total cost of ownership for building versus buying. How often do internal builds get abandoned? How these decisions actually get made, there is nothing credible out there. Every number I found traced back to a
development agencies, of course. then writing a blog with no sample and no methodology, no framework, etc. So there is no good data on this. and I looked. I looked a lot. Now this means in my opinion, this is an experience, not a finding. Now tell me what you have actually seen about this rational about companies making decisions about building versus buying.
Daanyaal Bandukwala: Yeah, so good good point, good question. I think one big thing that is important is we as as corporations, as companies, we don't like to publish our failures, right? and if
Shubhendu: Yeah.
Daanyaal Bandukwala: conferences as we were mentioning before, you always give the examples that really worked out well or whatnot. So there are I'm sure tons and we've heard about these anecdotally as well, of AI pilots and things like that that have happened internally, in some cases that have failed. And the reality is it goes back to some of what we were discussing before where there's the governance piece that comes into the flow, or perceived governance piece, let's call it.
Shubhendu: Yeah.
Daanyaal Bandukwala: and then the other piece is lack of adoption. I've heard about that a lot lately, and you know, I I think What's interesting is once a organization decides to stop a a certain program or a certain way of doing things, when they look at it in a retrospective manner, they quickly learn that, yeah, the issue is lack of adoption. Maybe we should have provided some more training to our sales rats or whatever the case.
Shubhendu: I can't
Daanyaal Bandukwala: And we we should have just expected them to pick it up right off the back. So One thing I've seen a lot of organizations do now is these actual retrospective reviews, which is becoming a little bit more and more common, where their executive sponsor who's behind the team for a specific initiative says when they initiate the project that you know y you have till X date to complete the project, let's say. And if there are flags in between, we need to do a review and if we're at the point where there's enough flags where it doesn't make sense to proceed with this anymore, I need a true retrospective review on what went wrong and
Shubhendu: Yeah.
Daanyaal Bandukwala: how we kind of adjust accordingly. whether it be you know a case w where you've used a external vendor or built this internally, how can we course correct for the next initiative, right? So we don't get into this rapid cycle of pilot and it didn't work out.
Shubhendu: Yeah. It's it's funny 'cause our last guest, Michael McKenna, also said the same thing. And I I really wish what you're saying it makes total sense in terms of r retrospective, but I really wish the vendors would also do this. Well like because
Daanyaal Bandukwala: For sure.
Shubhendu: like you you meant just mentioned as well, the case studies are always that you know what we did successfully. But one thing which Michael said in that episode was if There is an AI vendor which is coming to you and they do not disclose the failure. Hey, here's what we tried and this failed. Don't trust them.
Daanyaal Bandukwala: Yep.
Shubhendu: Don't like that's that's the more important part, but okay, how many times they have failed to get this right? Because that's a lot of learning which goes into this. Now, this brings it us into a more like formal segment about how to choose and who benefits. Now, Daniel, how do I choose the right vendor when we talk about the pharma industry also? You have a real criteria.
Daanyaal Bandukwala: Yeah, so I I think there are a few key building blocks that you need to think about, right? so one of them I mentioned early on. do they truly understand your your space? And I'm not saying that they need to be X pharma or anything like that. Ideally that's great, but do they understand the operations of how your day to day like you were asking me before, loop me through a day in the life, right? I think that's a core criteria and that you should evaluate them on. I think the other piece is start with the actual problem, not just with AI as a whole.
Shubhendu: Yeah.
Daanyaal Bandukwala: we need to use AI. What is the problem internally? So actually aligning on that with all stakeholders internally because you might find when you have that discussion, each stakeholder sees it slightly differently. So don't ask what AI platform we should buy, ask which business problem we're actually solving. So whether it be field forced effectiveness or I don't know, market access analytics, PSP optimization, right? Align on the core bucket. Outcome-based, exactly.
Shubhendu: Outcome based.
Daanyaal Bandukwala: and then evaluate governance before the features. Sometimes I think this is an interesting concept because we as consumers, we're all consumers, right? We get drawn in on the shiny object, right?
Shubhendu: Yeah.
Daanyaal Bandukwala: So we see that you know, every AI vendor can demonstrate a chatbot or something that looks really cool in theory. this would make my life so much easier. But like we were talking about before, where's the data actually stored? is customer data the data that you as a client are putting in there going to be used to train models for your competitors. candida residency remain in Canada. audit logs, right? Pharma is used to audit in all functions. So this is something that'll be top of mind as well. role based permissions, things like that, right? You have your VP like I was talking about. You have your trade manager as an example. How does that all kind of fall into the the realm of Yeah, the processful. Then another thing that I think about time and time again is integration. So it's one thing when they deploy it and I see it in consulting as well. Part of the reason I launched my consulting practice was when I hired consultants when I was working in pharma, there always used to be lots of strategy work done. And then it was kind of like the consultants disappeared, right? They did the strategy work. But the
execution kind of disappears. And you know, I it's a concept I call of hiding behind the slide deck, right? So beyond the the visuals, the strategy behind it, the actual product, how will integration work? So if we're talking about a Salesforce effectiveness tool, your vendor should be asking, Hey, what CRM do you use? Are you on Viva?
Shubhendu: Yeah.
Daanyaal Bandukwala: Are you on Salesforce? Whatever the case, do we have a path forward to integrating within Current systems, or is this being built in a complete silo as well, right? that's super important. And then the last couple of pieces I'll go through. I don't want to go too too deep into this, but understanding the underlying models. What what is is the vendor asking you what you're currently using? or are they again building in silos? What is the real ROI? you're gonna get it from this. So I've seen many cases because we're in sometimes this pilot mode where people will love it or it looks like an amazing interface or great responsiveness. What truly when you implement it in its full steady state, what are the hours that are saved for X function? What's the cost reduction we can anticipate? Is there a fixed error reduction? Can we expect faster review cycle times, right? those are really, really important to come up with at the base of this relationship with the vendor versus okay, we'll try it for six months and then see what happens.
Shubhendu: Yeah. that's one way to see it. And now now because I am the lazy person, so w to make decision faster.
Daanyaal Bandukwala: Yes.
Shubhendu: Is there something which if the vendor says, you guys, like the listeners, they should just walk out of the room?
Daanyaal Bandukwala: Well that's interesting. so one thing that stands out would be that shiny object piece that I mentioned before, right? If they are solely focused on look at how good our platform looks and look at how you know quickly we can have it set up for you, they don't truly understand your problem statement. what's going on within your business. so that's something that's key there. Couple of other ones, just rapid fire. our AI model is a hundred percent accurate. I don't believe that. it can't be, right? It needs to learn and i
Shubhendu: Yeah.
Daanyaal Bandukwala: you haven't given it the opportunity for your specific use case. going back to governance, don't worry about governance. We'll figure that out later. We work with lots of clients in heavily governed environments. That's not an appropriate answer. Pharma is different than any other space. something I mentioned before, on asking whether the data is used to train models. If if your data is helping them train models for other clients, that could be perceived as a red flag. You have to really Objectively think about if you're comfortable with the information that's going in there. another one I I mentioned before, pharma experience. If they don't have pharma clients yet, I I don't think there's a major red flag, but if they see something around the lines of, well, AI is AI, it can be implemented in any space, that I would be
Shubhendu: Yeah.
Daanyaal Bandukwala: a little bit hesitant towards. There's Always in shoop, you've probably heard this too. There's this concept of AI is gonna replace people, right? as a pharma client, that's not exactly what you wanna hear.
Shubhendu: Yeah.
Daanyaal Bandukwala: mature AI vendors, I think, truly position their platforms as augmenting, you know, and being yeah,
Shubhendu: The workforce again.
Daanyaal Bandukwala: yeah, yeah. Really working with your workforce in nature, not replacing people. and building on that as well, that integration piece, right? Sometimes we say that okay, integration is is phase two. they should be thinking about integration right away, right? Yes, it it might
Shubhendu: From the get go. Yeah.
Daanyaal Bandukwala: happen at a later time point, which is fair, that's possible. But if they're not willing or looking to find a way to integrate within your current ecosystem or making suggestions from that perspective, that gets really, really tricky.
Shubhendu: Yeah, and and again, for those of you who are listening or watching, make a note of this. This is really good because I'm gonna add more f flavor to it. And what happens in usual cases is you take these questions which Daniel is asking and you ask them very clearly. A lot of the times the people you're dealing with, the salesperson, may not have the answer. I would like or like suggest all of you to push harder, to push harder to get those answers because a few of these things are often embedded in the master subscription agreement, in the contractual terms, which you're not aware of what you're signing to. And again, for those of you who are from the big companies, maybe you already have experienced this, but for those who are not, check the terms. It's very important, especially the fact that Daniel mentioned that is your model being trained by by a customer data, etc. That is very important to know because that is Going to be very, very relevant in the coming days, especially now that every jurisdiction, every, every government, every municipality or the state government there or provinces, they are pushing a lot of laws and regulations around it. So the least you would do by this is avoiding heavy fines by by like paying attention to that. All right, Daniel, I want to put one hard thing to you. Now I
Daanyaal Bandukwala: Okay.
Shubhendu: don't want to spring it. Now this is an uncomfortable question. So I'll start with like let me put your case as strongly as I can can first. The argument is that AI is not really a technology decision, that the organizations pulling ahead are the ones who work out who owns what, who escalates what, who signs their name on on the output, and that everybody else is buying tools and hoping. I I find that argument very persuasive. I've seen it play out. And I should say I have a stake in that being true because I sell
into that belief. Right? You have a written playbook. You have written a playbook that lives
Daanyaal Bandukwala: Yeah.
Shubhendu: in that belief too. So here's what I found when I went looking for the evidence. Model op says governed organizations move 12 times more projects to production. Model op sells AI governance software.
Daanyaal Bandukwala: Mm-hmm.
Shubhendu: IDC and NetApp found governance leaders see much higher revenue. Netapp sells data infrastructure and it sponsored the study. And I know we're chuckling, but let's get to the more interesting as well. PWC found most AI value concentrated. It was in the small group of firms correlated with PWC's own index. PwC sells AI consulting. Now
Daanyaal Bandukwala: Interesting. Yeah.
Shubhendu: Let's talk about McKinsey. McKenzie found that CEO oversight of governance correlates with better results. But McKinsey's own methodology notes what they have internally, concedes that a single year of survey data cannot separate cost from correlation. Meanwhile, the most cited independent field study of the Ph is that the core issue is not talent, it's not infrastructure, and it's not regulation. It found that employees are getting real value out of ungoverned personal AI tools while the official governed
Daanyaal Bandukwala: Yeah.
Shubhendu: and the dip like governed official deployments are stalling right now. So can you name one in independent non-vender study that isolates governance as the cause rather than a marker of companies that we're lying to that that that are going to win.
Daanyaal Bandukwala: So it's a great question. offhand, I don't know of any study that claims that. we would have to do a deep dive. but I'm not sure it exists, Troop. I I don't think it's it's out there. Right?
Shubhendu: Hundred percent. So you got read that. Yeah. Where I've landed with this, Daniel, after my research, is nobody has has run a study. Nobody has compared a heavily governed company against a lightly governed one. It has just simply not happened. And no one has measured what has happened if you govern versus if you do not govern. So this is not proven and it is not disproven either. It's a conviction. And I happen to share that conviction. Based on my experience, it's my intuition. I just think we should do be honest that it is what it is. It's conviction from all of us. So, what is the test? How do I tell your version apart from governance theater? Let's let's do something about that. And I'm going to say five things that senior managers believe, and you will just simply reply with true and false.
Daanyaal Bandukwala: Okay.
Shubhendu: so again if you want to speak more like like you you are invited to do so as well. So
Daanyaal Bandukwala: Chip.
Shubhendu: first is governance slow you slows you down.
Daanyaal Bandukwala: False.
Shubhendu: The pilot proves the value.
Daanyaal Bandukwala: is there like a medium, like a in between?
Shubhendu: Stify, yeah.
Daanyaal Bandukwala: Yeah, so I think there's a in between, right? So there's something I call pilot fatigue which exists. I think some companies are constantly in this cycle of let's do more pilots, let's do more pilots, let's do more and then they get to the point where nothing's really stuck
Shubhendu: Mm.
Daanyaal Bandukwala: how they would expect it to. And then they're nowhere further ahead than they were at the beginning of those pilots, right? So it goes back to what I was mentioning before on really defining that business problem, the objectives and whatnot, instead of throwing everything at the wall and saying, Hey, let's just try it, see what happens.
Shubhendu: So it's true with caveats.
Daanyaal Bandukwala: Yes.
Shubhendu: So next one is a human in the loop means the human is accountable.
Daanyaal Bandukwala: That's a good one. I would say it's another one of those medium ones, right? I would say that because of the the following. So a human in the loop, to what extent, right? is is their AI model kind of running in the background and that means they're in the loop and they're just reporting back the high level findings to the leadership team and checking the box? If that's the case, then yeah, sure, they're accountable. If it means that they're actually spot checking like we talked about and verifying what's coming out of the model, then they need to be a little bit more than just in the loop. They have to be in tune with what the outcomes actually are.
Shubhendu: Hmm. Okay, the next one is related. So the bottleneck is the model.
Daanyaal Bandukwala: Yeah, true, to some extent. I think we were just talking about vendor alignment and choosing the right vendor and and whatnot as well. So on one side, I think the model you choose and and the way you choose to work with it can be the bottleneck. We know this, there's certain models that are better for certain things. That's That's out there, that's proven, I would say, confidently.
Shubhendu: Okay. Yeah.
Daanyaal Bandukwala: at the same time I go back to the same sentiment of, have you truly defined your business problem appropriately?
Shubhendu: Hmm. Now this is the last one. Building means you own it, buying means your trap.
Daanyaal Bandukwala: These are good. Shoot, I I like these. So building means you own it. I mean I think to some extent, optics-wise, that might seem true. capabilities-wise, that may seem somewhat false as well. and what I mean by that is there might be certain things depending on your organization that you can't get beyond a certain level by doing it internally, right? so yeah, you you do own it, but are you also creating a stop gap for yourself in terms of how much more efficient a certain model could be if you took that expertise externally?
Shubhendu: You did really well from my experience. So let me arm two of those questions right now. So Boston Consulting Group and Columbia, Columbia Business School surveyed, I think it was more than a thousand employees, and 76% of the executives say their people are
enthusiastic about AI. 31% of the people actually are, and 77% of executives. Think managers are prepared to guide AI conversations, while 91% of the employees say there's a manager they're not prepared to. Now, alongside that, what's interesting is 90% of executives who are confident they can see what AI tools are being used in their organizations for, 49% of the employees admit using AI tools without approval.
Daanyaal Bandukwala: interesting.
Shubhendu: The the part that should worry a governance person like someone sitting in your your chair right now is 69% of the C suite said they're personally fine with unapproved use. Because they want the speed, because
Daanyaal Bandukwala: Wow, that is shocking. Yeah.
Shubhendu: they want they want to compete, they want to outdo their competitors. Now, so leadership is wrong about how people feel. They're wrong about what people are doing. And their worst offender, like about governance as well. Is your governance policy describing a company that exists? And
Daanyaal Bandukwala: Yeah, that's
Shubhendu: again, this has been this way before, yeah. I I'm I'm sure if you did this survey on cloud adoption, we
Daanyaal Bandukwala: Yeah.
Shubhendu: would probably be somewhere similar. But when you you talk about is your governance policy describing a company that exists, like what what do you think about?
Daanyaal Bandukwala: So I I think it's a interesting thought, right? and I don't wanna say this in a in a negative manner, but I think and we d discussed this a lot before. I think a lot of these studies aren't taking very highly regulated industries into into place, right? and
Shubhendu: Yeah, yeah. Yeah.
Daanyaal Bandukwala: and I think that's part of the challenge. in healthcare pharma governance is is everything, right? Risk mitigation is everything. So I I wonder what those stats would look like if you asked a pharma executive and how they might differ. I I think one thing that truly stands out when we have conversations with clients is actually fear, right? It's it's fear of what could happen. And, you know, w we talk with them a lot about this and building a true risk mitigation plan into place, right? That's when everyone gets a little bit more comfortable. Here are all the risks, here's what's defined, here's how the risks fit within our governance policy and what we're what risk we're actually comfortable with taking. And if there is a way to mitigate that risk in a appropriate manner, then yes, we are more willing to take that risk. versus, you know, just kind of going with at the side of your pants and and saying, Yeah, it We just want speed and we're just gonna move things forward no matter what
Shubhendu: Yeah, and Daniel, great answer. And I'm w the reason I I threw so many numbers, I want something to share of what I feel. Again, this is how I feel. First of all, the emotional aspect of how people are addressing this, you're absolutely spot on on that. Now, what is happening is, in my opinion,
AI is making everybody feel like they are a McKenzie class consultant. Yes,
Daanyaal Bandukwala: Yeah, it's true.
Shubhendu: and Please tell me honestly, have you heard about this phrase that ninety five percent of AI pilots fail?
Daanyaal Bandukwala: Yeah.
Shubhendu: Do you know how many people have heard that? Everybody. Yeah, it's
Daanyaal Bandukwala: It's it's a com common piece. Yeah.
Shubhendu: a common thing. And then everybody on on social media is talking about it. Why I wanted to bring this up because I I actually used it in one of the episodes, first episode. And I and then like a follower, he reached out, shared something. So I dug deeper into it. That number is coming from one MIT affiliated report from August last year.
Daanyaal Bandukwala: Interesting. Yeah.
Shubhendu: It is not peer reviewed. It
Daanyaal Bandukwala: It just got a lot of social media attention, I'm sure.
Shubhendu: exactly, and that is the problem. And it does not mean 95% got cancelled or can like it means so basically what the study says is researchers found that 95% of the pilots had not shown a measurable bottom or like ROI inside their study window.
Daanyaal Bandukwala: Yeah.
Shubhendu: Which was well under a year. A lot of changes take take again when you talk about big companies, whether it's pharma or non-pharma, they're big. So it will take sometimes it takes more, more than a year, more than a few months.
Daanyaal Bandukwala: Yep.
Shubhendu: Now s plobal surveyed just over a thousand leaders and found something related but different. Forty-two percent of companies abandoned most of their AI initiatives in 2025. Now this is up from 17% the year before. That that's a very different thing. But the what I'm trying to say is does the real number matter or like or does everybody just enjoy the big one, right? So it's 95% are failing, so this is a bad time to start RY. This is actually compelling or like making people leaders think and stop that okay, maybe we should wait a little bit, maybe we should and that is also something which is feeding into the resistance or the the challenges that the adoption of AI is is facing. Now I want to end on the part nobody puts in the business case, which is the people. Okay.
Daanyaal Bandukwala: Yes.
Shubhendu: So this last segment is about what does this do to people? You mentioned something very important and and and close to me. And then again, I I support that in like full wholeheartedly the emotional aspect of it. But let's start with whose job gets smaller if you're right. Well you say all all the things you're saying today, Tunnel, what do you think?
Daanyaal Bandukwala: So you know, I I actually think of it in a in a very streamlined manner. I think across our industry, across the board, capacity for many employees has been a challenge, right? whether it's your marketing function or your commercial functions, your trade functions, what your medical functions and you know Being on the other side of the industry, not directly working for a pharmaceutical organization anymore, I hear this from my clients all the time. I have so much to do, you know, and and so on and so forth. So in an ideal scenario, I don't see anyone's job necessarily getting smaller. I see as getting a little bit more efficient. I see as getting to the point where, you know, we're using AI and AI models to make our day to day. a little bit cleaner from that perspective.
Shubhendu: Mm-hmm.
Daanyaal Bandukwala: So I I don't necessarily see it as like a mass extermination of jobs. It's just a different
Shubhendu: Yeah.
Daanyaal Bandukwala: way of working. Now what I will level set and this might be a little bit of a bold statement, is those within organizations that are not willing to adopt these new ways of working with AI, I think yeah, your your position or you as an individual are at risk. Right. So that's why, you know, keeping up to date and ensuring that you're you're keeping these concepts top of mind, your way of thinking is is in line, is super important. that's what I think, you know, employers now, even when they're bringing on new hires, whether it's for a really junior role or something that's at the executive level, that's what they're looking for, which has changed the criteria a little.
Shubhendu: It has. And if you see it this way, that confidence about job security, it rises as you go up, right?
Daanyaal Bandukwala: Yes.
Shubhendu: That's how it has been. Now 23% among the middle managers. 31% are of like when you go up are the upper managers. Now 35% in the C suite. Now Gardner projects that one in five organizations will eliminate more than half of their middle management layer by the end of this year. By leveraging AI. I want to be precise. Gardner projects it. It is a forecast, not a measurement, and I'm citing it secondhand.
Daanyaal Bandukwala: Yeah.
Shubhendu: And there is a strange thing underneath it as well. 46%. I I did a lot of research for this in this episode. I I think people I got a lot of feedback that they want more numbers as well. So, anyways, I I I had to say Forty six percent of managers say AI will not affect their job right now. Or that they're not showing it. Almost 80% of those same managers use AI tools every week. That belief was fifty-four percent in like two years ago. Like and it was still forty-four percent in twenty twenty-five, flat. It's not changing, but the usage has exploded of
Daanyaal Bandukwala: Right.
Shubhendu: AI. Nearly half of them, and again this and I'm just quoting the sources which I got, use AI every week and still think it will not touch their job. What are they not letting themselves see, in your opinion? Have you ever like Again, you w what's your thoughts? And I'm talking generic, like not just.
Daanyaal Bandukwala: Yeah. Yeah for sure. Yeah, not just pharma. Yeah. I think it's you know, a lot of people looking at it from the mindset of they're using AI for convenience, right? For their personal convenience right now.
So, maybe it's a little bit under the radar in some cases. maybe it's not all the AI platforms that have been mandated to use by their company. They're using some other stuff. on the side, which in turn seems to make their job easier. I think there's also this concept and you know I speak to a lot of new grads as well. and there's been this concept that you know in in many cases where once you're in, once you put in your time, you're not replaceable. I don't believe that. I think, you know,
Shubhendu: Yeah.
Daanyaal Bandukwala: for sure there's a a very emotional look to it and a a people look to it and you gotta look at the person individually. I one hundred percent am a big believer of that. But everyone to some extent is somewhat r replaceable in an organization, right? so I wish good luck to those people that think that they're not replaceable completely. I think there's certain functions of their job that may evolve. And it goes back to my earlier point on the adoption piece. It seems like these people are adopting, but they're not aware of what they're actually adopting, right? so having that awareness of what you're adopting will be key. Roles will evolve. So the same traditional roles may not exist, but there'll be other roles out there for those that have adopted and are aware of what's going on. and you'll see New titles and new positions come to be.
Shubhendu: Yeah, so let's finish on something that's if someone is listening can actually or should actually do. one last number from me guys. so Microsoft surveyed 20,000 knowledge workers across 10 countries this year. Only 30% feel rewarded for redesigning how work gets done with AI. And only 26% say the leadership is clearly and consistently aligned on AI strategy. Microsoft sells AI totally. So factor that in for for this. Now 90 days, one thing, how how should like people address this or approach this?
Daanyaal Bandukwala: Yeah, so I I think you know and take out the AI piece for one second, right? So think about someone who is really motivated and they're in a new role or they're starting at a new company, someone that's a real go getter, right? automatically you create a little bit of a action plan for yourself of How am I going to show my value while learning in those first 90 days? how am I going to show my value to the organization? Right? I think that concept is no different from as you're building out AI tools and and workflows, right? So I think for senior management and for executives, they need to understand that this is a transitional time, right? there shouldn't be this wide expectation as such that ultimately, yeah, this person, this individual built out this AI workflow that saved us X amount of hours. No big deal. I mean, if that's gonna have some sort of effect on your bottom line or your general efficiency from a hours perspective, that needs to be measured. And what we were talking about earlier, I think re-emphasizes this point very heavily. Setting those KPIs for what efficiency looks like, what you know cost management looks like what are you trying to achieve from redesigning this specific workflow or whatever the case
Shubhendu: Yeah.
Daanyaal Bandukwala: set those up Don't just do it for the sake of doing it. and same same thing with AI generally, like we talked about. We're not just adopting AI because it's the new fresh thing on the block, right? We're doing it for Yeah, exactly. We're doing it
Shubhendu: Good cool.
Daanyaal Bandukwala: for for specific business reasons, right? We're bringing it into the fold because there is a business challenge or something that we want to address within our organization. And just like we measure other projects and and problems and things like that by using KPIs or or key indicators, we should be doing that with this.
Shubhendu: Now last thing. So in terms of AI Daniel, what's the one thing which you're least sure about as when you talk about the AI technology, the industries, response, etcetera?
Daanyaal Bandukwala: You cut out shoop. I couldn't hear that piece.
Shubhendu: So w wha about AI, what's the thing which you're the least sure about? Whether it's AI adoption, AI technology itself, the industry's response.
Daanyaal Bandukwala: Yeah, so I think adoption as a whole, will will be a challenge, right? We're it's not that we're not seeing adoption, I should correct myself, but I think the challenging part that comes to be is within your organization, are you truly creating a plan for it to be used widely, right? Sometimes when we get into these pilot groups too, The users that we put within that pilot group are your super users, I call them, right? Who are very much on board with it and whatnot and you test it with them. everything's went super well, we're in good shape. let's just roll it out to everyone now. That's not an adoption plan for the masses, right? You've created an adoption plan for a very select group of people. And I'll refer back to a specific example here. From back let's go back to COVID days for one second, in the pharma sales environment. So remote calls, virtual calls, zoom calls with physicians were not really a thing prior to that. Like you would have the occasional phone call, the occasional video video call. You'd use Skype, things like that.
Shubhendu: Yeah.
Daanyaal Bandukwala: but it was not common. For a pharma sales rep, for anyone customer facing a medical science liaison, whatever the case, you would show up in person to the clinic, to the hospital, over dinner, lunch, whatever the case. We went through an environment there where, as I was mentioning before, there were some customer-facing employees that jumped right on this and said, Okay, I I'll try to schedule calls, I'll try to find space, I'll try to meet with my customers this way. Now, what was layered on top of that where you had lack of adoption from reps who had maybe been in their role for many, many, many years. They were probably the best reps in person, but didn't
necessarily adopt wealth in a virtual environment. They really struggled in that environment. Right? they weren't usually able to do their usual side conversations with the staff and all of that. it was a very different environment. And then you factored on top of that to sure, the first maybe few months, physicians were busy, obviously, with everything that that was going on. But they were also open to trying this new model here and there. And then you you went through this fatigue period. So I just used this example in a way to demonstrate how many AI models need to be thought out. That was the biggest lesson for us, because it was a very quick change. We we moved to virtual calls overnight in some cases, and no one was expecting to to do that. Here we have a little bit of runway. Yes, as we've talked about before, there's a little bit of a a race between some companies on how quickly they adopt and whatnot. But we also at the
Shubhendu: Yeah.
Daanyaal Bandukwala: same time have a little bit of a runway to map out, okay, let's look at all the stakeholders involved in this. Let's look at all of their profiles from an adoption perspective and and segment them, right? Like internally
Shubhendu: Yeah.
Daanyaal Bandukwala: and externally. How will that all play out?
Shubhendu: Okay. I could talk to you for hours. It's it's really
Daanyaal Bandukwala: I don't
Shubhendu: fun. one tradition we have in this show, Daniel, is every guest leaves with a question for the next one.
Daanyaal Bandukwala: Okay.
Shubhendu: So my last guest was Michael McKernan, very renowned executives who have worked at Wayfair, Amazon, and his question to you was have you let AI enter in your personal life in in the form to kind of support your hobbies in some way.
Daanyaal Bandukwala: Yep, one hundred percent. So simple really simple example. I like to travel. most of my travel itineries are completely built on AI now. it it kinda knows what I like, the types of restaurants I like, the types of hotels I like, the types of experiences I like. So usually the the first draft that comes out is pretty close to what I would book for myself. and that makes things seamless for me.
Shubhendu: Very nice. And so now, Daniel, what is your question for our next guest?
Daanyaal Bandukwala: Okay, so this is a a little bit of a fun one. what's one thing AI should never replace?
Shubhendu: Okay, that's a good one. I'm sure it'll make the speaker think you know the guest by the way, so I'll keep it a little surprise for you when the episode is launched.
Daanyaal Bandukwala: there you go.
Shubhendu: here's what shifted for me today. I came in thinking that the hard part of AI was deciding what to build. I'm leaving thinking the hard part is no is deciding who answers for it and that almost nobody has written that down. The technology is actually the easy half. Daniel has written about this in more depth than we can actually get in a conversation. The book is called Pharma AI Strategy Lab and it is co-authored by Nicole Serena. It came out in June, and the link will be in the show notes. if anybody's interested, go check it out as well. Daniel, thank you, thank you, thank you for this very educative session. I love the conversation and you let me push and you did not flinch. And that, my friend, means means a lot. So If this made you want to go and find out who signs their name to AI output in your company, guys, subscribe wherever you are watching or listening from. I will see you next week. This has been the integration layer.