Join the Insider! Subscribe today to receive our weekly insights
David (00:01)
Welcome, you are listening to the IT/OT Insider podcast. I’m your host David and we bring you the latest insights in shaping the world of industry 4.0 and smart manufacturing. Today I welcome Toni Manzano. Toni has decades of experience in managing software projects in the highly regulated pharma industry. I’m gonna definitely talk about that as well today.
He also teaches courses on artificial intelligence and big data at several universities in Spain. And something I find fascinating as well is that Toni is also associated with the United Nations Industrial Development Organization as an expert in industry 4.0 in Pharma. That is a specialized agency of the United Nations promoting accelerating sustainable industrial developments. And…
Toni is the co -founder and chief scientific officer at Aizon. Aizon delivers GxP -compliant software as a service solutions to the life sciences manufacturers. That’s all super interesting stuff to talk about. Toni, very, very happy to have you here today on the podcast.
Toni Manzano (01:07)
It’s an honor for me. Thank you for the invitation, David.
David (01:11)
The honor is all mine. Maybe as an opening question, Toni, as I said, you have so much experience in the pharma industry. Can you walk me a bit and our listeners to your career? why did you start working in the life sizes industry? What experiences shaped your views, et cetera?
Toni Manzano (01:33)
Sure, sure. After teaching some astrophysics at the university, I early realized that this is not for me and for my family. I was involved since the beginning in the pharmaceutical industry, improving critical operations, using maths, using advanced statistics, 20 years ago, 25 years ago, wow, the time is going on, and using software, of course, because it’s the only way that we could have all these insights.
And we started with the first LIMS in Spain. We created the LIMS that was acquired by Siemens. And afterwards, we created an MES, Manufacturing Execution System, that also was acquired by the giant in the US, Aspen Technology. And that means a lot because this is designing how the pharma works. 20 years ago, LIMS and MES was something wow. And today is also wow, after 24 years ago. So I started in this way.
David (02:33)
That’s amazing. It’s fun to say because indeed today we are still implementing MES systems and LIMS systems and all these things. And I also worked with them through my career. And it’s fun to say that you still, that even today, these systems typically are like, wow, this is next level.
Toni Manzano (02:57)
True, true. And this is what we are facing today. Today we are delivering Aizon Execute as a platform where we are recording the electronic batch record system using AI. You know, even today when you go to the market, the market is expecting for this kind of software in order to improve their process and their recipes. 24 years later, the request is the same.
This is very hard to say, very hard to understand, no?
David (03:29)
Yeah, so basically, what you’re saying is today we put the, sorry, buzzword AI on the stuff we are doing, but actually, we’re still working with data, we’re still working with mathematical models.
Toni Manzano (03:45)
Yes, so you know I have my own definition of AI. Every time that I go to my classes for the first time, the first day in my classes, all the students are always asking for the same thing. Toni, which is your definition of AI? Because you have thousands of definitions. And the definition that I always say is AI is like a cocktail. A cocktail with three main ingredients. First ingredient, power computing. Second ingredient, algorithms. Third ingredient, maths. I’m so sorry, but maths is needed. Well.
With these three ingredients, you can do a very nice cocktail. But if you want to work in pharma, the fourth ingredient or the secret sauce of this cocktail is the quality data. Quality data. Without quality data, you cannot bring this kind of AI. So it’s true that we are talking about buzzwords because at the end we talk about our computing algorithms, maths plus data. And we know that since the beginning of the times, a chemometric method.
that we use in pharma, a chemometric method are that: algorithms, maths, computing and data. The difference is that today, this kind of algorithms plus maths plus power computing are more oriented to emulate human knowledge. It’s not just to create a model, something that is replicating a part of the real world. Today, we are talking about AI in these terms, how we can emulate, how we can robotize human knowledge.
David (05:12)
So yeah, very nice. This is the first time I’m thinking about AI as a cocktail. Now I do like my AI and I also do like my cocktails. So this is gonna work for me. Are there in that long area and you started, I would say coding your own MES platforms until today, where you’re offering software as a service solutions to the same industry. Are they like?
pivotal moments where you say like these are times or moments or technologies which really shaped the industry or the way the industry embraces digitalization.
Toni Manzano (05:57)
of the most critical steps when you are trying to digitalize industries and even more the pharmaceutical industry that probably is the most conservative industry in the world is that you need to work very closely with humans. So humans we have the knowledge about how to do this kind of digitalization well and this is mainly the first challenge or the main challenge that we are facing today.
In an industry like the pharmaceutical industry that was presenting 1 .6 trillion of benefits, trillion in 2023. If that works, why to change it? So if they are conservative, it’s because that works very well with the status and the ground that they have today. They don’t need to evolve because they are earning a lot of money. So…
The increment was $100 billion in the last of 2022. So it’s amazing. So why they need to evolve. This is the main challenge that you are facing. You work with people in order to improve their processes, but why they need to improve the processes if they are earning a lot of money. So the challenge here is we have one main challenge. One main reason is the patient. We…
We need to understand the pharmaceutical industry with two main targets. One is, well, you have to make happy your stakeholders. Of course, business, we are talking about business. But we have a specific specificity. This industry is serving patients and we cannot do things in any way. So we have to do it in the right way. So these two drivers are configuring the both things. Why the industry is earning a lot of money?
and why the industry doesn’t want to improve or to change what is already working.
David (07:52)
Have you seen over these two decades that’s, I would say, a difference in adopting new technologies? Has it become faster to adopt new technologies? Is there an impact of, for example, now with the cloud and AI becoming so, I would say,
available everywhere. Also, of course, for R &D because we typically in this podcast, we typically talk about, I would say, the actual manufacturing process. But especially also in the pharma, I think from an R &D perspective, there has also been a huge change in adopted technologies.
Toni Manzano (08:35)
Sure, sure. Actually, two decades ago, the main challenge was how we can extract, how we can get data coming from the equipment. That was the main concern, no? So we were trying to put some PLCs, Siemens S7 and so on, in order to connect these kind of things. We put a very big box in order to transform the serial port signals into logic. Well, the thing is that that was the main challenge.
decades ago, how to transform analog to logic information. Today we have a benefit. The providers of these equipment, they are already providing that by default. So the pharmaceutical industry doesn’t need to ask for this digitalization. They are already providing by default. So this is not a challenge anymore. So the challenge today is that the society, we talked about that before.
The full society is pushing in this direction in order to have more information with more easy access in a fast way. So these three requirements, that is the three V’s of big data, volume, velocity, and variety. So we are talking about a lot of different information in a very high volume, but I need that in a very fast way. These three V’s are also pushed by society. So the problem today or the main challenge is that once we have
everything digitalized, how we manage that in the three V’s of big data, and how we can transform this data into information and the information into knowledge. This is the main challenge.
David (10:15)
you still teach courses on these topics as well. I assume that’s because you have a passion for teaching. I like that as well myself, but I also sometimes see that there is a big gap between theory, yeah, academical theory, so to say, and applying stuff in, yeah.
in practice. Do you see that the same? How do you try to bridge that theory -practice gap?
Toni Manzano (10:49)
That’s true, that’s true. So, you know, AI was born in 56, in the past millennium. It was born like a scientific discipline as part of the computing science. And it was completely theoretical. People were doing the algorithms in the notebook by hand with the pensions and so on. After more or less practically 80 years, the situation was the same. So…
my students are coming to the masters and to the this kind of AI degrees and they have to understand the full algorithm but it’s the same way do you remember when you were studying statistics at least once in our life we need to know how the covariance formula looks like how the Pearson coefficient looks like at least once in our life it’s the same when you know how the random forest algorithm works at least once you can rely on that so
The theory is needed like always, but today we have another obligation as teachers. We need to show how to do that in the right way and we need that the students make their own reflections. So you cannot teach this kind of good practices. You need to teach how to arrive to the good decisions in order to avoid bias, gender bias, sex bias, religion bias, zone bias. You need to teach this kind of…
good practices in order that each one applies in their own field because the fields are completely different. So this is how the bridge is fulfilling. So when I go to my classes, of course the theory is there, it’s needed at least once in their life, but afterwards the best part of the classes is when the students, they think in their field in the right way to proceed with the theory.
David (12:45)
Yeah, nice, cool, cool. It’s a lovely way to see it like this. And I also fully agree with the fact that I think you really need to have some theoretical knowledge. You can’t just at random start selecting algorithms and filling in parameters and go like, you know what, let’s see. Let’s sit back and relax and see what happens.
Toni Manzano (13:07)
No, no, no. In this way the pharma doesn’t work. In pharmaceutical industry we need to apply science. At the end we’re talking about scientific things for patients, so it’s not heuristically. You have to know what you’re doing.
David (13:24)
And something I also really like to talk about is, I would say, the regulations in this industry. FDA, GxP, whatever. Can you, I’d say for our audience and especially for the ones who are less familiar with what is a regulation, why do we have regulations? Could you like in a very shortly explain a bit more about…
Toni Manzano (13:45)
Mm -hmm.
David (13:51)
what that is and what the impact of that is and why we do these things.
Toni Manzano (13:55)
Sure, sure. Actually, every single pill that you take when you get a cold, the pill that you are taking, or when you get an vaccine, the vaccine that you are getting has been never tested. Never. So could you think on that? The pill that you are ingesting when you have a cold or a powder and so on has never been tested because it’s in the envelope or in the syringe or this liquid is never tested. So…
The regulation is required in order to ensure that this medicine that you are getting that has been never tested will work as expected. So there are three main commandments that the regulators or the regulations are looking for when they are approving a drug. The drug must be safety. They have to have the quality expected and the efficiency. Efficiency is very important. So you know when this pill that you are getting,
this pill is not having the efficiency expected, your ill will be delayed or maybe it’s not healed. So these three commandments, safety, quality and efficiency, are the three main commandments that the regulators need to be sure in every single medicine. So this is why they are there. They need to be sure that every patient takes this medicine when they need, in the right moment, in the right proportion, in the right situation.
David (15:24)
Cool, and maybe to just, I would say, throw a couple of acronyms to you. So if I say GxP, for example.
Toni Manzano (15:32)
GxP means good whatever practices. If we talk about manufacturing, we talk about good manufacturing practices. If we talk about laboratory, in quality control and so on, we talk about good GLP, good laboratory practices. If we talk about warehouse management system, GWHP. And we are preparing the GMLP, or the GAIP. So this is common good machine learning practices and good AI practices.
David (16:01)
something that’s being worked on now or…
Toni Manzano (16:04)
Yes, this is working now, yes, you’re right.
David (16:07)
That’s interesting, that’s interesting. And that’s maybe also a good bit because how do regulations or regulatory bodies, I would say, speed up or speed down innovation? What is your opinion on that?
Toni Manzano (16:22)
Well, let me share one thing with you. Do you know that FDA, they have all their servers in Amazon Cloud?
Second thing, the FDA was the first one publishing a good practice based on how to use AI as a medical device. For example, in 2023, we had by the first time, the first guidance from the FDA about how to use AI in drug manufacturing. You can Google that and you will find. The EMA, the EMA published their reflection paper on AI for medicines life cycle.
Also past year and even more they are not just publishing the guidance and so that’s all first of all They request the feedback coming from the pharmaceutical community and with this feedback they improve their guidance so it’s an open discussion and this is Let me say I’m feel very proud about the regulation that we have nowadays because they are taking this this this responsibility they want to provide a
all the things necessary in place in order to be sure that the patients are held and managed in the right way, in their own way. So they are first.
David (17:44)
That’s an interesting insight. This probably to make, I would say, a step towards validation of ways we work. So I think in the, well, in many industries, but especially in the pharma industry, when we implement a certain algorithm, when we implement a system.
when we implement the control loop, et cetera. And you already mentioned that we wanna make sure that if we produce a pill or a vaccine or whatever we’re making, that everybody who is getting that pill or that vaccine from that batch gets it in a controlled manner. And that means that the way we, I would say, build software and build hardware in these manufacturing plants needs to be validated.
And I also just want to be a bit more on your experience on validation, also the why, the how, how that also changed over the last years, I assume, because a while ago, validation was really, you had an engineering diagram and you just, you saw like, okay, if this happens, then that happens, okay, validated, but now also with software that becomes really more, yeah, it becomes more complex, I assume.
Toni Manzano (19:02)
more complex. Yeah, you’re completely right. You know, in one of the biggest drug manufacturing sites in Spain, I cannot say the name, 25 % of the staff are validation people, quality people. 25 % of the staff. And this is increasing the price of the drug. And this is why the price of the drug can be increased. Well, the thing is that…
David (19:19)
Wow.
Toni Manzano (19:31)
In the old world, let me say in the 20 years ago, the CDs and the DVDs were fully controlled by the validation team. They can control in which computer the CD has been installed. I can see the version, I can stamp my signature. No one can do that in another way. Today we are talking about a completely different thing. Everything is in cloud. No one knows when…
Microsoft team is changing the version. When Amazon is modifying some services, you know, Amazon is modifying, is updating the services every five minutes. Some of the services are already updated every five minutes. How I can control that, no? In the static world, that validation team, the CSV team, the Computer Software Validation team, thinks that we’re thinking on something static. The validation that we’re talking about 20 years ago and today,
David (20:11)
Yeah.
Toni Manzano (20:27)
is based on static assumptions. Everything is under control in my CD, my DVD, no one can change. The regulators, the regulatory bodies, they are already aware of that and they want this change. They promote the innovation, they promote the modernization of the pharmaceutical industry. You can make a Google FDA modernization industry and you will see a lot of papers requesting to the industry to be modernized. But you cannot do that with the classical approach. Everything is static. So for the reason they created the new concept.
CSA, computer software assessment, no validation assessment. That means if you see the difference, you will see that the main difference is that you need to put inside the validation concept the risk assessment and the human rationale on top of that. And this is one of the main problems in the industry. In the industry, classically, everything needs to be considered as static, nothing changes. And in the real world, everything changes continuously.
So the validation teams were always trying to avoid this kind of continuous variability. But the variability is per se in everything. Teams are changing, my equipment is changing, my provider is changing, my decisions are changing continually. So the classical approach for pharmaceutical industry is that everything needs to be under statistical control, variable by variable. What happens when a variable changes?
the other ones are already immediately affected. But I’m very happy once all of them, they are under control. But no one was taking care of the real essence of the nature of the changes. Because it’s so complex. When you talk about biopharmaceutical industry, you are talking about bugs, bacteria that they are building a bioreactor. This is a life. You cannot control a life of things. The only way is today is putting something…
really smart on top that is able to understand what’s happening. Let me explain this experience. It’s very nice. I was completely surprised when I entered in a big pharmaceutical industry with 1 ,500 liters bioreactor doing insulin inside. And I entered with the director and he was an old man and this guy said,
this smells not very good, this batch is not going well. But all the sensors were saying that my batch is going well. You know, this guy was able to condense all the information, all the sensors, pH, density, temperature, humidity, the sugar, blah, blah, in only one observable. How does it smell? So I think that this is the concept of AI, you know? Today with AI, you can aggregate.
this kind of reality complex and variable reality in something that is understandable by a complexity Mind that can be a human or can be a machine the problem is that if this guy goes in my vacation in vacationing holidays No one knows that that is going wrong
David (23:42)
It’s amazing how experienced people have a so fine -grained understanding of what they are making and it is so difficult to grab that knowledge into a system.
Toni Manzano (24:02)
So why they can do it? Because they have, of course, experience, but experience means a lot of data with a lot of rationale behind that. So if you try to simplify that with data and rationality, and we can emulate that with maths and algorithms and power computing, we have AI. But 24 per 7. This is the difference.
David (24:31)
And it also clearly shows that just applying a black box model, we get some data and we’ll feed that through algorithm A, B or C and we expect something to come out of this, it’s not gonna happen. It’s…
Toni Manzano (24:50)
I can ask you something, please don’t say black box in the regulatory world because this is completely forbidden. No, you know, black box means that you don’t know what happens inside. And the regulatory bodies, they only believe or they only accept models that has been created understanding what’s happening inside. So you need to demonstrate that you have everything on the con…
David (25:04)
Yeah.
Toni Manzano (25:17)
For the reason the black box at the beginning, when AI was a black box, as you say, because we have a neural network with some hidden layers, I don’t know how many, with some input cells and so on. Today, we are very lucky because the community is working and is creating standards, for example, like OpenNN If you look for OpenNN Open Neural Networks, you will see that there are systems that they are able to transform black box into transparent box. So…
David (25:46)
Wow, that’s interesting.
Toni Manzano (25:47)
Yeah, and graphically. And they are open source, and these tools are open source and also graphical. So you see how the input is going in the machine, in the all black concept. You see how the neurons and the weights in each layer has been assigned. And you see the output. And everything can be traced, and everything can be exported. So it’s not black box anymore. At the end, we have a transparent box that can be accepted by the regulation.
David (26:17)
Interesting. So here again, and you mentioned a couple of times, but you really, you always talk about this, how can we push innovation? How can we keep on taking steps in the sector? And what I also mentioned in the introduction, Toni, is that you’re involved in the United Nations Industrial Development Organization, UNIDO in short.
How did you, what does that agency do? What is your involvement?
Toni Manzano (26:52)
Sure, sure. You know, in the United Nations, they organize missions. In this case, in UNIDO. UNIDO is a branch of the United Nations dedicated to develop good practices and to promote the industrialization in medium -developed countries and underdeveloped countries. And they have missions. And in these missions, you participate as an expert in order to help to this…
countries because at the end we are talking about countries, Slovenia, Cuba, Africa, South Africa, West and so on. You contribute with other experts in order to promote the industries. And United Nations, they are completely convinced, they believe that AI is a tool, AI and digitalization in general, is a very good tool to democratize this advance, this evolution of the industry.
There are countries like Cuba, for example, that they are completely isolated by the American laws and so on. But they are extremely powerful, believe me. They are so, so good. They don’t have access to Amazon. They don’t have access to Google. But they created their own cloud. And they are amazing creating AI systems that they can, they are advancing a lot in terms of drug discovery. For example, the COVID vaccine that they created by themselves.
was a success vaccine and no one knows that. So we only know Moderna, Pfizer and AstraZeneca. No one is talking about the Cubans vaccine and it was amazing, it works very well. So they are very advanced. So this is an example of how United Nations are supporting these countries in order to help them to be promoted. And United Nations believes that digitalization in general and AI,
are fields that can help a lot. Can help a lot.
David (28:49)
So that’s then the, I would say, that’s the mission is to, I would say, be a knowledge body, transfer knowledge.
Toni Manzano (28:57)
Knowledge and also to find out potential relationship between countries. For example, if you find a country that is very good in something, another country that is very good in other things, or they have the opportunity to work with Europe, for example, why not to work together? These kind of collaborations are working very well.
David (29:18)
Cool. Hey, and also, I also just want to, of course, touch your company, Aizon which you co -founded. I assume that you started with founding Aizon because you identified a certain gap in the industry, something which wasn’t solved yet.
Toni Manzano (29:40)
Yeah, yeah. So you know, 20 years ago, the same three co -founders, we have been working in other iterations previously. We know each other since I was here. So after 24 years, we found that we were living, by the first time in the history, a special moment. It was how the cloud computing was completely
democratized to everyone. So you know, you remember when you had these external hard disks close to your computer in order to make your backups and your pictures with all your holidays and so on? So the price for one gigabyte encrypted, backup, and with replication in multi -region is one cent, one cent per month. That’s nothing. So this is the…
most amazing democratization around the world for data and data access. So we believed on this change 10 years ago and we went to Silicon Valley and we arrived to an agreement first with Google. They didn’t believe in the life science vertical and for this reason we went to Amazon and we created this kind of partnership with them. And that was because it was the first time that we had all the technology in our hands in order to
to comprehend the complexity and the continuous variability in the pharmaceutical industry. Always in the pharma industry, everyone with the recipes, with the old patents and so on, we are looking at the processes in a very limited way. We are humans and we only see in three dimensions. So we only see three parameters at the same time, always within limits. But we knew for all the previous experience 20 years ago that
the reality is more complex and is continuously varying. How to tackle that? Not from a scientific perspective, but based on the data that we have. And for the reason we went to Silicon Valley to found this agreement. And the problem to solve is that, is how to understand complexity in a scientific world.
David (31:55)
What are the typical type of problems you solve for your customers?
Toni Manzano (32:01)
Well, everything that is related with a problem that cannot be solved with the usual tools that they are using. So all the customers that we have, they are amazing. They have fantastic statistical teams with very knowledgeable teams. But again, the problems are complex. And let me explain you a problem. For example, you know, in the pharmaceutical industry, when you work in chemistry,
and your raw material is coming, you analyze the raw material, you analyze the density or the color, the organoleptic quality attributes, and you can say, wow, this is not compliant with my specification. You can say no to the provider. Mr. Supplier, I need another batch that is committing with my specifications. But now imagine that you work with human plasma and you transform human plasma into something. You cannot say ‘no’ to human plasma.
Human plasma is gold because it’s coming from donors. So when you work in medicine, in drug manufacturing, and your raw material is blood, this is gold. In whatever state it arrives. So this company, and I cannot say the name yet, but you can imagine, there are not too many companies dedicated to human plasma. When you are receiving this raw material and you see that the final product that you are doing are decreasing the quality, the yield,
of the processes, you cannot do anything because the raw material is there and you know very well your process, you are very knowledgeable in all the operations that you are doing. How to tackle this kind of problems, no? Well, and this is very complex because you are centrifuging this plasma, you are filtering, you are centrifuging and you are transforming in different phases this subsequent product that you have coming from the original one. But if the original plasma that you are getting is not good, the final yield is not good.
and you have to apply these recipes that has been patented. Well, how we can help on that? AI is able to understand the complexity between the quality attributes of the raw material, the quality attributes that you expect from your final product, and all the critical process parameters in the middle. So we’re talking about hundreds, hundreds of parameters. So it’s not possible from human perspective. So when you put in place AI and you have very good data,
quality data, remember the secret sauce of the cocktail, and you have this data working and transforming this reality, this complex and variable reality in something that is understandable, you can act on the process in order to know which is the perfect speed, the perfect filtration process in order to have operation by operation, phase by phase at the end, the final expected yield. This is an example that I can share with you because it’s amazing.
David (34:53)
It’s absolutely amazing and it also has a real tangible impact. There is only so much plasma available.
Is there also the role for IT OT convergence? So on the one end, the operational systems on the other end, and more the IT world. Do you see the gap closing in the life sciences industry, or do you see the gap staying the same or even becoming bigger?
Toni Manzano (35:22)
No, I think that the gap is being filled in part because we have two main reasons. One is, as I said to you before, the society is pushing, but the second one is the executive levels. They are being replaced by young people. Young people like you, no, believe me, this is a problem when you have a chief executive officer, a top level, that they don’t understand how AI works.
or they are not working as, or they don’t believe on what they are doing with digital because my industry is earning 1 .6 trillion per year so I don’t mind. But when this generation is being replaced by young people like you, they understand and they live every day, every moment with that. So it’s not something strange, it’s something that is needed because I live with that. You know?
When people is coming to me and they say, well, my data will be in the cloud. I don’t believe in the cloud. Why my data is in the cloud when I have all my recipe, all my pH during 24 hours in the cloud? Someone can steal that, no? And I always say, do you know where is your money? Which is the last time that you went to the bank, the physical bank office. I don’t remember. So if the money’s in the cloud.
how you cannot rely on your pH on the cloud. So this mindset is changing because people like you are getting this kind of executive level. So this is the gap that is filling.
David (37:01)
Interesting. And Toni, you seem to me like a guy who does have kind of a glass ball. And so I just want to end this podcast by asking you this I would say this broad question. How do you see, especially in your industry, I would say industrial digitalization, IT OT convergence, whatever name you want to give it, how do you see that changing in the years to come? What are the things we’re going to see?
or the other things which are going to happen.
Toni Manzano (37:33)
In 2003, Tim Menzies is one of the fathers of AI. He was old in 2003. He said, in the future, so now, AI won’t be a hype anymore. AI will be a commodity. So I see life science using AI and using digitalization because the curve of the innovation, so at the beginning of the, you know, the innovation is like a Gaussian.
At the beginning of the curve you have the early adopters, people that is risky and they buy an electric car. But in the last part of the Gaussian, we have the laggards in adopting innovation. So it’s just a matter of time. So in the future, in five years, the pharmaceutical industry will adopt by necessity because everyone is doing that AI. So there is no way. So we have to get the bus or get the bus.
David (38:32)
I like this final outlook. Toni, thank you a lot for joining this conversation. I really enjoyed it. Keep on doing the work you’re doing. It’s amazing.
And to our listeners, thank you for tuning in and you can find more episodes on ITOTinsider.com. Until next time.
Toni Manzano (38:55)
Thank you so much, David.
David (38:57)
Thank you.