Guest Speaker:
Transcripts
Welcome to Tech Lyceum. I'm your host Neerja, and this is the podcast where we explore the technology shifts that are shaping businesses. And today we are turning our attention to life sciences, an industry where the ability to make the right decision at the right time can have a very real impact. Whether it's a clinical trial, manufacturing quality, regulatory compliance, or commercial performance, every decision counts.
Yet many organizations still rely on broad analytics platforms that give them plenty of data, but not always the industry context they need to act on it with confidence. So today we are exploring the shift from horizontal platforms to industry-first analytics, and why this is becoming a priority that CXOs cannot afford to ignore.
So to help us unpack this, we have Aman and Ravi with us today. Aman Lamba is the head of data and AI for banking, financial services, and high tech in Birlasoft, helping clients realize value from their data and AI investments. He has 30 years of experience in the industry and has published papers on data products and the information economy. The perfect person to have on this particular episode, Aman. Thank you so much for joining us.
Yet many organizations still rely on broad analytics platforms that give them plenty of data, but not always the industry context they need to act on it with confidence. So today we are exploring the shift from horizontal platforms to industry-first analytics, and why this is becoming a priority that CXOs cannot afford to ignore.
So to help us unpack this, we have Aman and Ravi with us today. Aman Lamba is the head of data and AI for banking, financial services, and high tech in Birlasoft, helping clients realize value from their data and AI investments. He has 30 years of experience in the industry and has published papers on data products and the information economy. The perfect person to have on this particular episode, Aman. Thank you so much for joining us.
Listen On
Speaker – Aman - 01:32
Thank you. Happy to be here.
Thank you. Happy to be here.
Speaker – Neerja - 01:34
And we also have Ravi Teja, who is an engineering leader at Birlasoft with 19 plus years of experience in life sciences and medtech. He has led large-scale data, cloud, and AI programs, delivering strong business outcomes and consistent execution across enterprise engagements.
His work focuses on helping clients modernize technology, improve decision making, and create measurable value. Incredible! Hi, Ravi. Thank you so much for being here.
And we also have Ravi Teja, who is an engineering leader at Birlasoft with 19 plus years of experience in life sciences and medtech. He has led large-scale data, cloud, and AI programs, delivering strong business outcomes and consistent execution across enterprise engagements.
His work focuses on helping clients modernize technology, improve decision making, and create measurable value. Incredible! Hi, Ravi. Thank you so much for being here.
Speaker – Ravi - 02:06
Thanks, Neerja, for inviting me.
Thanks, Neerja, for inviting me.
Speaker – Neerja - 02:08
Aman, Ravi, let's start with our first question, and I'll direct this one to you, Aman, to start with.
Q: Why are horizontal analytics platforms falling short of life sciences decision making needs?
Aman, Ravi, let's start with our first question, and I'll direct this one to you, Aman, to start with.
Q: Why are horizontal analytics platforms falling short of life sciences decision making needs?
Speaker – Aman - 02:22
Sure, thank you. What's happened in technology over the last few years is the rise of unified data platforms or horizontal platforms. These are engineered to be industry neutral, right? And that's exactly the trade-off. They're excellent at unified storage, compute, pipelines, analytics, but they arrive with no concrete understanding or opinion about how life sciences or any specific industry actually runs. And we see four gaps show up consistently. They have no native model of the business. They understand tables, files, columns, but they do not understand a study, a site, a subject, a batch, an adverse event, and so on, and so every team rebuilds that logic, that ontology, as it were, itself, and it could be different, and the difference only surfaces when there's a leadership review. Related to that is what we call semantic drift. The different functions in the organization may have different definitions of something like, say, enrolled subject, and they get three different answers: screened, randomized, first close. All three are defensible. All three are computed on the same platform, but none of them reconcile.
So the platform is not wrong. It's just that nobody ever agreed on the definition. And linked to this is actually a compliance challenge, right? Because your GXP validation, Part 11 controls, audit trails, all of this get bolted on and try to trace this information through in its own way, and that's a challenge. So at a very high level, these platforms are good at organizing data, but without life sciences context, agreed definitions, and built-in compliance teams spend their time reconciling these numbers instead of actually using them and making decisions.
Sure, thank you. What's happened in technology over the last few years is the rise of unified data platforms or horizontal platforms. These are engineered to be industry neutral, right? And that's exactly the trade-off. They're excellent at unified storage, compute, pipelines, analytics, but they arrive with no concrete understanding or opinion about how life sciences or any specific industry actually runs. And we see four gaps show up consistently. They have no native model of the business. They understand tables, files, columns, but they do not understand a study, a site, a subject, a batch, an adverse event, and so on, and so every team rebuilds that logic, that ontology, as it were, itself, and it could be different, and the difference only surfaces when there's a leadership review. Related to that is what we call semantic drift. The different functions in the organization may have different definitions of something like, say, enrolled subject, and they get three different answers: screened, randomized, first close. All three are defensible. All three are computed on the same platform, but none of them reconcile.
So the platform is not wrong. It's just that nobody ever agreed on the definition. And linked to this is actually a compliance challenge, right? Because your GXP validation, Part 11 controls, audit trails, all of this get bolted on and try to trace this information through in its own way, and that's a challenge. So at a very high level, these platforms are good at organizing data, but without life sciences context, agreed definitions, and built-in compliance teams spend their time reconciling these numbers instead of actually using them and making decisions.
Speaker – Neerja - 04:08
So they're great for organizing data, like you said, but they often miss the life sciences context needed for faster and more confident decisions, right? And that brings me to this question, Ravi, I'll direct this one to you.
Q: What does an industry-first analytics model look like?
So they're great for organizing data, like you said, but they often miss the life sciences context needed for faster and more confident decisions, right? And that brings me to this question, Ravi, I'll direct this one to you.
Q: What does an industry-first analytics model look like?
Speaker – Ravi - 04:28
Sure, Neerja. So that's a great question. We've all heard about domain-driven design models. So this is a little extension of those domain-driven data models. So what industry-first analytical model actually means is they deal with a specific study site subject and an adverse event, so that these data models are precompiled. For example, we all have been dealing with audit trails, right? And then we have lineage, which is quite important today in the DNA of AI-driven models or LLM models. So an industry-first analytics model looks like a particular model which is driven for a site or a batch, but the lineage and the audit trails are pre-built, not added later. So I'll take some few examples in short. We all might have heard about something called CDISC, which is Clinical Data Interchange Standards Consortium, right? And then we have IDMP identification of medicinal products, or we have a UDI for medtech domains, which is unique device identification. All these are standards put by a certain region. For example, IDMP is widely used across the EMEA region, which is identification of medicinal products. So, if we take paracetamol as an example, even the minute subject of where it was stored, where it was transported, the truck used to bring it from point A to point B has to be recorded.
Earlier, we were treating them as separate database objects, putting them into databases and trying to query them. But now, with the availability, we are creating these precompiled data models, and then feeding data to this industry-first analytical models with audit trails and lineage. So, this is giving the life sciences industry a unique opportunity, where even a small golden batch record, right? If we have to find it, you could get it quickly, or if a batch has to be discarded, you don't need to discard the entire year's batch of that particular product. This is quite revolutionary, in my opinion, Neerja, for the industry.
Sure, Neerja. So that's a great question. We've all heard about domain-driven design models. So this is a little extension of those domain-driven data models. So what industry-first analytical model actually means is they deal with a specific study site subject and an adverse event, so that these data models are precompiled. For example, we all have been dealing with audit trails, right? And then we have lineage, which is quite important today in the DNA of AI-driven models or LLM models. So an industry-first analytics model looks like a particular model which is driven for a site or a batch, but the lineage and the audit trails are pre-built, not added later. So I'll take some few examples in short. We all might have heard about something called CDISC, which is Clinical Data Interchange Standards Consortium, right? And then we have IDMP identification of medicinal products, or we have a UDI for medtech domains, which is unique device identification. All these are standards put by a certain region. For example, IDMP is widely used across the EMEA region, which is identification of medicinal products. So, if we take paracetamol as an example, even the minute subject of where it was stored, where it was transported, the truck used to bring it from point A to point B has to be recorded.
Earlier, we were treating them as separate database objects, putting them into databases and trying to query them. But now, with the availability, we are creating these precompiled data models, and then feeding data to this industry-first analytical models with audit trails and lineage. So, this is giving the life sciences industry a unique opportunity, where even a small golden batch record, right? If we have to find it, you could get it quickly, or if a batch has to be discarded, you don't need to discard the entire year's batch of that particular product. This is quite revolutionary, in my opinion, Neerja, for the industry.
Speaker – Neerja - 06:32
Yeah, that's interesting. In short, an industry-first model is built around life sciences realities from the start, and these are aligned to real business decisions, Ravi, I'll ask you a follow-up question.
Q: How can organizations move from fragmented dashboards to domain-owned data products?
Yeah, that's interesting. In short, an industry-first model is built around life sciences realities from the start, and these are aligned to real business decisions, Ravi, I'll ask you a follow-up question.
Q: How can organizations move from fragmented dashboards to domain-owned data products?
Speaker -Ravi - 06:52
Sure, Neerja. So this is a good continuation of the earlier segment. Imagine if we were storing all this data across separate database objects, separate databases, because each supplier might have their own system of storing data with a different data model. People have created these dashboards to monitor data, like how it is moving on or how it is incoming into the system, right? But in while doing it over the last two decades, what we have created is every enterprise has, in fact, some enterprises have 1000s of dashboards. Some enterprises have ended with hundreds of dashboards. The point of this dashboard, which was supposed to give an information to the leadership, right, has been destroyed. Now it has just been lying there, and people don't know who owns what dashboard because over a period of time, people move out of the organizations they come in.
So now, with the whole concept of data products, right? You could easily control the quality of the data product, the quality of the data contract that you have between whenever there is an exchange of data from a certain point A to point B. And in this whole, with the whole revolution of AI and generative AI. What we could do is we could simply eliminate these dashboards and move to a point of natural query-based responses, which is the whole purpose of building these dashboards.
So, in short, summarizing right, the whole industry is moving away from these fragmented dashboards where you have to collect data from these multiple dashboards to domain-owned data products, which in short means giving the information to the right stakeholders so that a decision can be made, and this is being enabled with the new industry data and event data products that we are building for the life sciences industry.
Sure, Neerja. So this is a good continuation of the earlier segment. Imagine if we were storing all this data across separate database objects, separate databases, because each supplier might have their own system of storing data with a different data model. People have created these dashboards to monitor data, like how it is moving on or how it is incoming into the system, right? But in while doing it over the last two decades, what we have created is every enterprise has, in fact, some enterprises have 1000s of dashboards. Some enterprises have ended with hundreds of dashboards. The point of this dashboard, which was supposed to give an information to the leadership, right, has been destroyed. Now it has just been lying there, and people don't know who owns what dashboard because over a period of time, people move out of the organizations they come in.
So now, with the whole concept of data products, right? You could easily control the quality of the data product, the quality of the data contract that you have between whenever there is an exchange of data from a certain point A to point B. And in this whole, with the whole revolution of AI and generative AI. What we could do is we could simply eliminate these dashboards and move to a point of natural query-based responses, which is the whole purpose of building these dashboards.
So, in short, summarizing right, the whole industry is moving away from these fragmented dashboards where you have to collect data from these multiple dashboards to domain-owned data products, which in short means giving the information to the right stakeholders so that a decision can be made, and this is being enabled with the new industry data and event data products that we are building for the life sciences industry.
Speaker – Neerja - 08:35
So the shift is actually from scattered reports to accountable data products, where each domain owns all aspects of the data it uses. Right, Aman, I will ask you this question:
Q: What operating model and governance changes are required at this point?
So the shift is actually from scattered reports to accountable data products, where each domain owns all aspects of the data it uses. Right, Aman, I will ask you this question:
Q: What operating model and governance changes are required at this point?
Speaker – Aman- 08:54
The change is requires a federated operating model, and what that means is the change is less about technology and more about accountability. It means that we are moving from funding projects to actually running governed data products that last beyond the life cycle of an individual project, and are run with a thin center that sets the rules, sets the standards, enables the platform and the domains, as Ravi said, that build and own the products, and the split has to be explicit, right?
The central authority, as it were, owns the non-negotiables, the shared vocabulary for the entities, the core attributes, the data contract and quality standards, and the lineage and policy that is enforced in the platform and audit trails and so on, and a certification gate or checklist that a data product must pass before anyone can say this is trusted and it is official. That is the role of the center. The domains own the products. Every data product needs a named owner who sits in the business: clinical, quality, regulatory, commercial. And that owner is accountable for the definition, the quality SLA on freshness and completeness, the roadmap adoption, and you know the certification of the KPIs. You can look at this as almost like to take an example: a product manager for Tide or Pepsi who owns the product up to the point that it's on the shelf and the customers' experience of that product. It's exactly the same for data products. Now, in this model, governance moves upstream, right? The quality assurance, regulatory, privacy, security are all part of the governance. They are addressed or factored in at design. And in the modern AI-led framework, every individual has a responsibility across that life cycle, in a validated environment like in life sciences, that change is really the difference between determining where issues reside and how we determine you know accuracy that we'll have at the end of a release cycle versus at the beginning. All the funding and incentives will follow this. Products get the funding from the core, but they have to justify, and the product owner monitors the adoption, accuracy, and decision outcomes.
So it's more than just moving data from left to right; it's certifying it and owning it all the way to the consumption.
The change is requires a federated operating model, and what that means is the change is less about technology and more about accountability. It means that we are moving from funding projects to actually running governed data products that last beyond the life cycle of an individual project, and are run with a thin center that sets the rules, sets the standards, enables the platform and the domains, as Ravi said, that build and own the products, and the split has to be explicit, right?
The central authority, as it were, owns the non-negotiables, the shared vocabulary for the entities, the core attributes, the data contract and quality standards, and the lineage and policy that is enforced in the platform and audit trails and so on, and a certification gate or checklist that a data product must pass before anyone can say this is trusted and it is official. That is the role of the center. The domains own the products. Every data product needs a named owner who sits in the business: clinical, quality, regulatory, commercial. And that owner is accountable for the definition, the quality SLA on freshness and completeness, the roadmap adoption, and you know the certification of the KPIs. You can look at this as almost like to take an example: a product manager for Tide or Pepsi who owns the product up to the point that it's on the shelf and the customers' experience of that product. It's exactly the same for data products. Now, in this model, governance moves upstream, right? The quality assurance, regulatory, privacy, security are all part of the governance. They are addressed or factored in at design. And in the modern AI-led framework, every individual has a responsibility across that life cycle, in a validated environment like in life sciences, that change is really the difference between determining where issues reside and how we determine you know accuracy that we'll have at the end of a release cycle versus at the beginning. All the funding and incentives will follow this. Products get the funding from the core, but they have to justify, and the product owner monitors the adoption, accuracy, and decision outcomes.
So it's more than just moving data from left to right; it's certifying it and owning it all the way to the consumption.
Speaker – Neerja - 11:14
So we're looking at a balanced model, right, where enterprise standards are met, but there's also domain ownership for accountability of the outcomes. Let me ask you another question, Aman.
Q: How does an industry-first approach strengthen AI readiness, compliance, and trust?
So we're looking at a balanced model, right, where enterprise standards are met, but there's also domain ownership for accountability of the outcomes. Let me ask you another question, Aman.
Q: How does an industry-first approach strengthen AI readiness, compliance, and trust?
Speaker – Aman - 11:32
Sure. Let me take this, and Ravi, I'll pass it on to you. All these three AI readiness, compliance, and trust are usually treated as separate programs, but they all draw on the same foundation, and that is governed, well-modelled, lineage-rich data that exposes or surfaces through these data products. And so, your AI is always only as good as the meaning you feed it. It's no longer is AI is a data ready for AI, but can we feed our data into models and agents that are going to work autonomously, and these domain-owned products are the grounding layer. So, how do we ensure that trust through the lifecycle? And you know, we have built examples where an assistant has a supported RAG model that is actually grounded not just on a general document store, but actually pulls from the certified data product knowledge with version control and media edge carried through. Ravi, you like to add anything?
Sure. Let me take this, and Ravi, I'll pass it on to you. All these three AI readiness, compliance, and trust are usually treated as separate programs, but they all draw on the same foundation, and that is governed, well-modelled, lineage-rich data that exposes or surfaces through these data products. And so, your AI is always only as good as the meaning you feed it. It's no longer is AI is a data ready for AI, but can we feed our data into models and agents that are going to work autonomously, and these domain-owned products are the grounding layer. So, how do we ensure that trust through the lifecycle? And you know, we have built examples where an assistant has a supported RAG model that is actually grounded not just on a general document store, but actually pulls from the certified data product knowledge with version control and media edge carried through. Ravi, you like to add anything?
Speaker – Ravi - 12:25
Yes, Aman. I think you covered it, but I would just like to add one point. Like, if you Neerja and Aman, you guys know, right? If you ask for any LLM application or any generative AI leader who is building applications on data, right? They would first say that the main problem has been governed and trusted data because the challenge in nowadays industry Neerja that I am facing with our customers is every time LLM model or an application build using RAG doesn't give the same response and this is very hard to justify when the actual business users who are non-technical are using it, so creating that you know a governed data product which gives consistent responses has been the key. And I think with the best practices that Aman stated, and also maintaining ownership, right? Taking the example of Pepsi, who take the ownership till it is there on a customer's hand, right? That is the need of the hour, even for life sciences industry.
Yes, Aman. I think you covered it, but I would just like to add one point. Like, if you Neerja and Aman, you guys know, right? If you ask for any LLM application or any generative AI leader who is building applications on data, right? They would first say that the main problem has been governed and trusted data because the challenge in nowadays industry Neerja that I am facing with our customers is every time LLM model or an application build using RAG doesn't give the same response and this is very hard to justify when the actual business users who are non-technical are using it, so creating that you know a governed data product which gives consistent responses has been the key. And I think with the best practices that Aman stated, and also maintaining ownership, right? Taking the example of Pepsi, who take the ownership till it is there on a customer's hand, right? That is the need of the hour, even for life sciences industry.
Speaker – Neerja - 13:23
So I suppose the takeaway here is that trusted AI starts with trusted data. And I have one more question. I'd love both your thoughts on this.
Q: How should CXOs measure business value?
So I suppose the takeaway here is that trusted AI starts with trusted data. And I have one more question. I'd love both your thoughts on this.
Q: How should CXOs measure business value?
Speaker – Ravi - 13:37
As we discussed, Neerja, in the last 10 - 15 minutes, we've seen that people are moving away from dashboards, which is the need of the hour, which were created over the last two decades. You know, and every organization has a bit of tech debt, right? No matter what, how large the enterprise is. So now CXOs are actually thinking in a different way. How quickly the decisions can be made, and what is the impact of a decision based on the data or the governed data product-based application that we are building? Rather than measuring the number of dashboards, rather than asking their teams to you know collate the information from these dashboards, people are actually doing the work. So earlier, a CXO had to you know rely on his tech advisor to collect this information. Now he has the application in its hands to figure it out. So I think it is obvious. So CXOs are already measuring business value based on the quality of the data rather than the quantity of the number of dashboards. And I think we will see this shift getting into starting. I mean, from LSS to all the industries, pretty much. Aman, do you want to add?
As we discussed, Neerja, in the last 10 - 15 minutes, we've seen that people are moving away from dashboards, which is the need of the hour, which were created over the last two decades. You know, and every organization has a bit of tech debt, right? No matter what, how large the enterprise is. So now CXOs are actually thinking in a different way. How quickly the decisions can be made, and what is the impact of a decision based on the data or the governed data product-based application that we are building? Rather than measuring the number of dashboards, rather than asking their teams to you know collate the information from these dashboards, people are actually doing the work. So earlier, a CXO had to you know rely on his tech advisor to collect this information. Now he has the application in its hands to figure it out. So I think it is obvious. So CXOs are already measuring business value based on the quality of the data rather than the quantity of the number of dashboards. And I think we will see this shift getting into starting. I mean, from LSS to all the industries, pretty much. Aman, do you want to add?
Speaker – Aman - 14:44
Thank you. Two practical points, right, on how to measure value. So first, most important is to capture the baseline before we build anything, because that is what we need, after go live or after the new products. Nobody will agree on what the old number. And the benefit case will become diluted. The second is to keep our scorecard very tight. Keep only 8 to 10 metrics with a named owner and a target date against each, right? An objective and a key result linked to business value, and to review that quarterly for the with the owner alongside the targets and the P&L, rather than a separate data forum, we've often found that data discussions don't make their way or are not surfaced on the business side, and so it's important that the measurable case is reported and reviewed with or through the business rather than in a separate data ecosystem. And you know the critical thing is measuring the before and the after number.
Thank you. Two practical points, right, on how to measure value. So first, most important is to capture the baseline before we build anything, because that is what we need, after go live or after the new products. Nobody will agree on what the old number. And the benefit case will become diluted. The second is to keep our scorecard very tight. Keep only 8 to 10 metrics with a named owner and a target date against each, right? An objective and a key result linked to business value, and to review that quarterly for the with the owner alongside the targets and the P&L, rather than a separate data forum, we've often found that data discussions don't make their way or are not surfaced on the business side, and so it's important that the measurable case is reported and reviewed with or through the business rather than in a separate data ecosystem. And you know the critical thing is measuring the before and the after number.
Speaker – Neerja - 15:41
So the takeaway really is: CXOs should measure value by the quality and speed of decisions, business outcomes improved, risks reduced, and effort saved, and not just by the number of dashboards created. Would that summarize this conversation accurately, Aman and Ravi?
So the takeaway really is: CXOs should measure value by the quality and speed of decisions, business outcomes improved, risks reduced, and effort saved, and not just by the number of dashboards created. Would that summarize this conversation accurately, Aman and Ravi?
Speaker- Aman 15:57
Absolutely. Yes. Yes, that would be right.
Absolutely. Yes. Yes, that would be right.
Speaker – Neerja - 16:00
Thank you so much. Today's discussion highlighted why life sciences organizations need to move beyond horizontal platforms and adopt industry-first analytics models that bring together domain context, ownership, governance, AI readiness, and measurable business value. Thanks to Aman and Ravi for joining us and sharing their insights. Thank you so much once again. It was a pleasure speaking with you both.
Thank you so much. Today's discussion highlighted why life sciences organizations need to move beyond horizontal platforms and adopt industry-first analytics models that bring together domain context, ownership, governance, AI readiness, and measurable business value. Thanks to Aman and Ravi for joining us and sharing their insights. Thank you so much once again. It was a pleasure speaking with you both.
Speaker – Aman -16:25
Thank you.
Thank you.
Speaker – Ravi - 16:26
Thanks, Neerja.
Thanks, Neerja.
Speaker – Neerja - 16:27
And to all of you listening in, we hope this conversation helps CXOs rethink how data and analytics can drive faster, trusted, and more impactful decisions. For more queries, you can write to us at
contactus@birlasoft.com. We'll catch you on the next one.
And to all of you listening in, we hope this conversation helps CXOs rethink how data and analytics can drive faster, trusted, and more impactful decisions. For more queries, you can write to us at
contactus@birlasoft.com. We'll catch you on the next one.
You were listening to Tech Lyceum, a podcast from Birlasoft.
Other related Tech Lyceum episodes
