Listen to Audio
Listen Anywhere
June 2025
This episode, Takeda’s Senior Vice President, Head of Global Quality Compliance & Systems, Magaly Aham, joins the podcast to discuss how AI can enable sustainable quality transformation.
1
00:00:00,000 --> 00:00:04,000
Welcome to the ISPE podcast, Shaping the Future of Pharma,2
00:00:04,000 --> 00:00:09,000
where ISPE supports you on your journey, fueling innovation, sharing insights,3
00:00:09,000 --> 00:00:14,000
thought leadership, and empowering a global community to reimagine what's possible.4
00:00:15,000 --> 00:00:23,000Hello and welcome to the ISPE podcast, Shaping the Future of Pharma. I'm Bob Chew, your host,
5
00:00:23,000 --> 00:00:27,000
and today we have another episode where we'll be sharing the latest insights and thought6
00:00:27,000 --> 00:00:34,000
leadership on manufacturing, technology, supply chains, and regulatory trends impacting the7
00:00:34,000 --> 00:00:40,000
pharmaceutical industry. You will hear directly from the innovators, experts, and professionals8
00:00:40,000 --> 00:00:46,000
driving progress and shaping the future. Thank you again for joining us, and now let's dive9
00:00:46,000 --> 00:00:54,000
into this episode. Our topic today is AI as an enabler for a sustainable quality transformation.10
00:00:55,000 --> 00:01:01,000
To share more about this topic, I would like to welcome Magaly Aham, Senior Vice President and11
00:01:01,000 --> 00:01:08,000
Head of Global Quality Compliance and Systems at Takeda, who recently keynoted at the 202512
00:01:08,000 --> 00:01:15,000
ISPE Biotechnology Conference in Boston. Magaly, welcome to the podcast. We're glad to have you13
00:01:15,000 --> 00:01:24,000
with us. Thank you so much, Bob, for having me. So, Takeda, like most pharmaceutical companies,14
00:01:24,000 --> 00:01:30,000
has initiatives to explore applications of AI and related technologies. Can you tell us15
00:01:30,000 --> 00:01:38,000
about what your company is working on regarding AI as an enabler for sustainable quality transformation?16
00:01:38,000 --> 00:01:46,000
Sure. We have some good stories going on right now. For example, we have a deviation AI assistant,17
00:01:47,000 --> 00:01:55,000
which facilitates drafting of investigation reports based on input entered by the investigator.18
00:01:56,000 --> 00:02:03,000
It has a series of questions per investigation element, such as problem statement, initial19
00:02:04,000 --> 00:02:11,000
impact assessment, scope, et cetera. The investigator then responds to each question,20
00:02:11,000 --> 00:02:17,000
and then at the end, the AI assistant generates a structured narrative,21
00:02:18,000 --> 00:02:28,000
saving, of course, time from repetitive manual drafting, increasing readability, and quality22
00:02:28,000 --> 00:02:35,000
of the final report. It also has multilingual capabilities, which is actually pretty cool.23
00:02:36,000 --> 00:02:45,000
And right now, we have it in four languages, including English, German, French, and Japanese.24
00:02:46,000 --> 00:02:52,000
We're also working on SOP transformation. It's another one of the tools that we have25
00:02:52,000 --> 00:03:02,000
ongoing right now. And what we have at this moment covers three functions, search, compare,26
00:03:02,000 --> 00:03:09,000
and summarize. And the search function, for example, helps you input a specific process27
00:03:09,000 --> 00:03:17,000
you want to know about, for example, process validation. Then it gives you key details on28
00:03:17,000 --> 00:03:26,000
the topic, and also gives you a link to the most relevant SOP from our document management system.29
00:03:28,000 --> 00:03:37,000
The compare function allows you to input any SOP, and will pull from the document management systems30
00:03:37,000 --> 00:03:45,000
SOPs that are similar, and how similar they are by means of a percentage. So not only pulls31
00:03:45,000 --> 00:03:53,000
a list of SOPs, it actually tells you how similar they are or not, and it explains that.32
00:03:53,000 --> 00:04:02,000
It explains the similarity and the key differences. So it's very helpful. It does the search for you,33
00:04:02,000 --> 00:04:11,000
and it can also help you eliminate redundant SOPs, or when you are trying to work on34
00:04:12,000 --> 00:04:19,000
optimization or standardization, it's very useful in helping you identify all of those35
00:04:19,000 --> 00:04:28,000
related SOPs that may be in scope of your work. The last function is summarize, and this one36
00:04:28,000 --> 00:04:37,000
allows you to input any SOP, and it will provide you a summary of key information from that SOP.37
00:04:38,000 --> 00:04:44,000
So this is very useful when you want to understand certain SOPs, but you don't necessarily need to38
00:04:44,000 --> 00:04:52,000
train on them, so it's an easy way to get overall content. As part of this SOP transformation,39
00:04:52,000 --> 00:05:03,000
we're also exploring other functionalities, including the SOP drafting, and even using this40
00:05:03,000 --> 00:05:12,000
tool to assess if we can evaluate training development content. So more to come on that one.41
00:05:12,000 --> 00:05:19,000
Okay, so a few things come to mind. Change management, for example. So you have a change42
00:05:19,000 --> 00:05:28,000
proposal, and maybe you know that it's going to impact this SOP, or maybe that's part of the43
00:05:28,000 --> 00:05:35,000
change is to you want to alter this SOP. Do you think this system could then kind of44
00:05:35,000 --> 00:05:42,000
go out and say, well, okay, if you're going to change this, it's impacting this and this and this?45
00:05:44,000 --> 00:05:51,000
Well, right now, that's currently embedded in the document management system that we have, because46
00:05:52,000 --> 00:06:00,000
we have linkages, right, to related SOPs. So when you are, to your point, perhaps47
00:06:01,000 --> 00:06:08,000
working on a change control and you pull, I don't know, the process validation SOP,48
00:06:09,000 --> 00:06:16,000
that process validation SOP in its metadata, we have linkages to other SOPs. So when you49
00:06:16,000 --> 00:06:25,000
include that one as an impacted SOP, you will have the information on other related SOPs50
00:06:25,000 --> 00:06:33,000
for change management purposes. Okay. Onboarding of new people. And, you know,51
00:06:33,000 --> 00:06:41,000
you're given this list of SOPs that you need to read and understand, right? Is there any way that52
00:06:42,000 --> 00:06:51,000
some of these tools could help streamline that, maybe reduce the time required for onboarding53
00:06:51,000 --> 00:07:01,000
new people? Absolutely. It's one of the areas that we're exploring, how we can link the SOP54
00:07:01,000 --> 00:07:09,000
transformation with training and really not training, but learning, you know, having a more55
00:07:09,000 --> 00:07:16,000
focused activities in the actual learning piece rather than just reading an SOP.56
00:07:17,000 --> 00:07:22,000
We're not there yet, but absolutely it's one of the areas that we're exploring.57
00:07:22,000 --> 00:07:30,000
Okay. Well, great. So I'm sure you've heard of other companies and I'm sure you've been58
00:07:30,000 --> 00:07:39,000
approached by vendors with different AI apps. What stands out about your approach? How is it59
00:07:39,000 --> 00:07:43,000
unique or different from what others may be working on and why?60
00:07:47,000 --> 00:07:54,000
I don't know what may stand out as I don't have that insight as to how other companies are working61
00:07:54,000 --> 00:08:01,000
on their digital strategy. What I do know is that it's a focus everywhere because there's value,62
00:08:01,000 --> 00:08:10,000
right? I can talk about Takeda and Takeda has been very intentional. One of our objectives,63
00:08:10,000 --> 00:08:20,000
obviously, is to become a digital biopharmaceutical company. So we have been64
00:08:20,000 --> 00:08:27,000
very intentional in creating an engine that will drive this digital transformation that we're65
00:08:27,000 --> 00:08:38,000
striving to achieve, including a lot of obviously training. We have many different66
00:08:39,000 --> 00:08:49,000
courses and certifications that allows many of our employees to gain understanding up to the level67
00:08:49,000 --> 00:08:56,000
that they want to get, right? There's some training that allows some of our employees to68
00:08:57,000 --> 00:09:06,000
even develop their own AI companion tools. We also have innovation centers,69
00:09:06,000 --> 00:09:14,000
internal innovation centers, which I think is an advantage for us because we have these70
00:09:14,000 --> 00:09:23,000
innovation centers across Takeda, which what they do is basically we go to these centers and71
00:09:23,000 --> 00:09:31,000
present our ideas for solutions or things that we envision. And we have the technical72
00:09:32,000 --> 00:09:39,000
powerhouse there to allow us to develop those solutions and evaluate whether they're scalable73
00:09:39,000 --> 00:09:48,000
across the organization or not. So I think that's a very useful thing for us to have and allows us74
00:09:48,000 --> 00:09:58,000
to drive innovation faster. The other thing that comes to mind with this question is we also have75
00:09:58,000 --> 00:10:07,000
an AI labs, okay, where people can test their AI ideas for feasibility in a safe environment76
00:10:08,000 --> 00:10:15,000
and without impacting any of our systems. So this allows people to feel not only more comfortable77
00:10:15,000 --> 00:10:25,000
with AI and the possibilities, but also excited on what it can do in facilitating our daily work.78
00:10:26,000 --> 00:10:30,000
So let me go back to this innovation center where you take your ideas.79
00:10:31,000 --> 00:10:40,000
There's always a tension or a conflict as companies try to innovate between letting people80
00:10:40,000 --> 00:10:49,000
sort of play and experiment and try different things versus, okay, here's the vision and we're81
00:10:49,000 --> 00:10:55,000
creating this and then this and then this and then this and anything else doesn't fit in.82
00:10:56,000 --> 00:11:02,000
How do you balance allowing for creativity and experimentation versus83
00:11:03,000 --> 00:11:10,000
descending into chaos and the resulting need to systematize everything?84
00:11:11,000 --> 00:11:19,000
No, that's a very good question because especially if you provide for the environment, right, and when85
00:11:19,000 --> 00:11:25,000
we talk about the AI labs for people to experiment, I think that also helps people86
00:11:25,000 --> 00:11:35,000
feel more comfortable and it helps us demystify AI and gen AI in a way. So it actually helps us87
00:11:35,000 --> 00:11:41,000
augment capabilities and help people feel more comfortable. But to your point, yes, we do have88
00:11:41,000 --> 00:11:50,000
governance, right? So we might have solutions that could be developed. However, not every solution89
00:11:51,000 --> 00:11:59,000
is approved for wider implementation. First of all, we need to understand the problem that we are90
00:12:00,000 --> 00:12:09,000
solving. We are also evaluating for scalability, right? Is this a solution that is scalable91
00:12:09,000 --> 00:12:18,000
across the entire Takeda organization? And what's the level of effort? Also, what's the92
00:12:18,000 --> 00:12:28,000
value creation? Because everything takes time and time represents also cost. So what is the value93
00:12:28,000 --> 00:12:40,000
creation? How is these tools or digital tools that are being created contribute to our vision94
00:12:40,000 --> 00:12:49,000
of value creation of being better, faster, and more efficient? So we have governance in place95
00:12:49,000 --> 00:12:57,000
to determine that. I assume then from what you said that there are companion teams in,96
00:12:57,000 --> 00:13:04,000
for example, MS&T or other groups who are also working with AI, machine learning,97
00:13:04,000 --> 00:13:16,000
or digital twin technologies. Is that true? Absolutely, yes. We have an enterprise DD&T98
00:13:16,000 --> 00:13:28,000
organization that establishes the framework for the enterprise and the strategy that will99
00:13:28,000 --> 00:13:34,000
enable Takeda to identify and address the unmet needs across the different business units that we100
00:13:34,000 --> 00:13:43,000
have through share investments. They steward the DD&T span across Takeda and drive enterprise101
00:13:43,000 --> 00:13:51,000
innovation. As part of that enterprise DD&T organization, we have business partners who sit102
00:13:51,000 --> 00:14:01,000
on the DD&T leadership team, and they represent many T&T organizations across Takeda. And T&T103
00:14:01,000 --> 00:14:08,000
is an acronym for Takeda Executive Team. And what I mean by that to your question is these104
00:14:09,000 --> 00:14:19,000
organizations are like global manufacturing and supply and quality, R&D, Japan, PDT or plasma105
00:14:19,000 --> 00:14:27,000
derived therapies, among others. So we have the enterprise DD&T or data, digital, and technology106
00:14:27,000 --> 00:14:35,000
organization. And then we have DD&T heads in all of these sister organizations or business units107
00:14:35,000 --> 00:14:45,000
across the organization. The structure makes us more agile to drive value for our users and108
00:14:45,000 --> 00:14:53,000
customers. And based on the aspiration I mentioned earlier of becoming a digital biopharmaceutical109
00:14:53,000 --> 00:15:04,000
company, yes, we are working with AI, AI, virtual reality, augmented reality in all parts of the110
00:15:04,000 --> 00:15:12,000
organization. And yes, we are also piloting and using digital twins, not only in manufacturing,111
00:15:12,000 --> 00:15:20,000
but even in R&D. So that's just some of the examples that come to mind.112
00:15:20,000 --> 00:15:28,000
So at the Facilities of the Future conference last January, ISPE's conference in San Francisco,113
00:15:28,000 --> 00:15:36,000
I heard Sanofi make a presentation about a digital twin aseptic filling machine.114
00:15:37,000 --> 00:15:44,000
But it took them four years, and the first three was pretty much just getting the site organization115
00:15:45,000 --> 00:15:53,000
to buy in to doing this at all. So what issues or stumbling blocks have you encountered,116
00:15:53,000 --> 00:16:03,000
and how did you solve them? Well, from the manufacturing perspective, if we talk about117
00:16:03,000 --> 00:16:11,000
digital twins, we're piloting that, like I said, in R&D, in manufacturing.118
00:16:12,000 --> 00:16:21,000
From the quality perspective, one of the stumbling areas that I can share119
00:16:22,000 --> 00:16:34,000
has to do with the data quality, right? And perhaps the inconsistency that we have found120
00:16:34,000 --> 00:16:43,000
in data structure, and particularly when you are dealing with different data sources.121
00:16:44,000 --> 00:16:50,000
So having data that is not standardized or from different systems has been a challenge,122
00:16:50,000 --> 00:16:58,000
at least for the digital tools that we are working on in quality. To that extent,123
00:16:59,000 --> 00:17:08,000
Takeda has put in place a data governance organization that is working on different124
00:17:08,000 --> 00:17:17,000
aspects of data, because obviously that's the raw material, right, for all of these initiatives.125
00:17:17,000 --> 00:17:23,000
So master data, data quality to ensure that the requirements for new systems are in place,126
00:17:24,000 --> 00:17:30,000
and for tools that are being developed that need to leverage whatever systems we have right now,127
00:17:30,000 --> 00:17:38,000
that we have better guidance in terms of what we will need in order to make those initiatives or128
00:17:38,000 --> 00:17:47,000
tools successful in achieving what we want to achieve. And in some cases, that might be129
00:17:47,000 --> 00:17:54,000
data cleanup and standardization. So yeah, I understand what you mean about taking130
00:17:56,000 --> 00:18:02,000
some of these years. We have had some situations where we have tried some tools131
00:18:04,000 --> 00:18:12,000
sometimes, and we have to recall that AI, gen AI, we have kind of learned as we go.132
00:18:12,000 --> 00:18:23,000
And so in many cases, we have designed some solutions with the output and not the input133
00:18:23,000 --> 00:18:32,000
in mind. And that has represented some challenges. So the apps that you have developed and deployed,134
00:18:33,000 --> 00:18:36,000
can you quantify the benefits that you've achieved?135
00:18:36,000 --> 00:18:45,000
Well, value creation is part of what we consider in every project. But I can tell you that136
00:18:46,000 --> 00:18:53,000
so far, overall, in Takeda, it will be millions of dollars. We also have been working with137
00:18:55,000 --> 00:19:00,000
automated visual inspection system using AI, and that has helped us reduce138
00:19:01,000 --> 00:19:15,000
rejection rates from around 30% for parental biologics to less than two, and has allowed us139
00:19:15,000 --> 00:19:22,000
to save at least 16 hours per batch in the places where we have incorporated this, just to name a140
00:19:22,000 --> 00:19:30,000
few examples. Well, that's impressive. And it's great that you've been able to implement that141
00:19:30,000 --> 00:19:38,000
kind of technology. Thank you. Going forward, what innovations do you see possible with AI142
00:19:38,000 --> 00:19:44,000
and related technologies? And what organizational regulatory and resource challenges might exist?143
00:19:45,000 --> 00:19:50,000
And how broadly and in what ways do you see these technologies being used in five years?144
00:19:51,000 --> 00:19:58,000
Oh, wow. I see this applied everywhere, everywhere that is possible, right? I think145
00:19:59,000 --> 00:20:10,000
as an industry, as I mentioned earlier, I think we all see the value of applying these technologies146
00:20:11,000 --> 00:20:23,000
internally. For once, it allows us to be more predictive. And I think it changes the mindset of147
00:20:24,000 --> 00:20:33,000
going from firefighting to more risk awareness and being more predictive on everything that you do.148
00:20:33,000 --> 00:20:42,000
There's definitely a need to augment understanding, because you don't really validate AI,149
00:20:42,000 --> 00:20:50,000
right? It learns. So, it's something that is still kind of an area that we all need,150
00:20:50,000 --> 00:21:00,000
particularly regulators, get into understanding these models a little bit more, because it does151
00:21:00,000 --> 00:21:09,000
require a shift in mindset. Of course, we will have to work with regulatory guardrails to monitor152
00:21:09,000 --> 00:21:18,000
its application. But I see it's used everywhere. R&D, for instance, conducting in silico studies,153
00:21:20,000 --> 00:21:27,000
and perhaps even digital runs to demonstrate process optimization outcomes. And this could154
00:21:27,000 --> 00:21:33,000
perhaps minimize the amount of PPQ batches that you might need to run in the future for155
00:21:34,000 --> 00:21:39,000
submissions and new product launches. So, I think the possibilities are endless.156
00:21:40,000 --> 00:21:48,000
Well, thank you for sharing these initiatives and the success stories. Just to recap,157
00:21:49,000 --> 00:21:57,000
Takeda has a center of excellence where it sort of brings together these different initiatives.158
00:21:58,000 --> 00:22:07,000
Your quality department has created this AI for structuring deviation, adjudication, and159
00:22:07,000 --> 00:22:13,000
development in the reports. You've got something that looks at SOPs, and you're exploring ways160
00:22:13,000 --> 00:22:21,000
this can help with training. That's all really great. I think it's wonderful that these161
00:22:21,000 --> 00:22:26,000
technologies are really getting some airtime. So, this brings us to the end of another episode of162
00:22:26,000 --> 00:22:34,000
the ISPE podcast, Shaping the Future of Pharma. A big thank you to our guest, Magaly Aham, for163
00:22:34,000 --> 00:22:40,000
sharing more about how global quality Takeda is using AI to enable sustainable quality transformation.164
00:22:41,000 --> 00:22:46,000
Please be sure to subscribe so you don't miss future conversations with the innovators,165
00:22:46,000 --> 00:22:52,000
experts, and changemakers driving our industry forward. On behalf of all of us at ISPE,166
00:22:52,000 --> 00:22:57,000
thank you for listening, and we'll see you next time as we continue to explore the ideas,167
00:22:57,000 --> 00:23:04,000
trends, and people shaping the future of pharma.