Data localization laws can split the central platform your AI depends on
When Saudi Arabia's data localization rules reach Informa in 2023, its live global data platform rests on the opposite principle: centralize everything. Sangeetha Parsan, Vice President of Customer Data Solutions and Services with more than 20 years in data, must split people, process and technology apart without breaking the business, often with only double-digit days to comply. When the same demand lands on your team, where do you start?
Get a four-pillar playbook for localizing data while keeping global operations and AI running
Sangeetha Parsan explains how Informa responded when Saudi Arabia's data localization requirements collided with a centralized global data platform that a live business depended on. She presents a practical four-pillar framework (technology capabilities, data compliance mapping, process change and people readiness) with a phased day 0 to day 180 plan, and discusses how local models and federated learning can keep AI strategies alive when borders fragment data.
“Don't start with what you have centrally. That could end up misleading you because it's built for the entire global organization, not for one region.”
“But people don't move in a project timeframe.”
What you will learn
Start localization work from the capabilities a region needs instead of pulling apart the existing central architecture.
Build a data compliance map with legal and privacy teams to define exactly which data must stay in-country.
Prioritize process changes by impact rather than difficulty, using a day 0, 60, 90 and 180 plan.
Plan change management, training, a managed services model and a longer hypercare period for the people running the new stack.
Protect AI capability with local models and federated learning when borders fragment training data.
Key takeaways from the speaker
Strategic approaches to building compliance-ready technology at global scale
Details
Level
Intermediate
Format
Tech talk
Technical depth
Practitioner
For
Senior individual contributor, People manager, Executive
Length
33:34
Published
August 2026
Questions this talk answers
What is the first step an organization should take when preparing for data localization requirements?
Identify the capabilities needed in the region, the data that powers them and the technology required. Do not start from the central stack, because it was built for the whole global organization, not one region.
How should a team prioritize localization work under a tight deadline?
Prioritize by impact, not difficulty: critical in-scope data isolated with no violations on day zero, secondary processes by day 60, edge cases by day 90 and full maturity with a playbook by day 180.
What happens to AI when data has to be localized?
Training data fragments, inference can miss regional nuances and governance becomes harder across many models. She suggests local small language models alongside global models and federated learning on gradients rather than raw data.
Why can't a company simply stay non-compliant?
Non-compliance means losing trust with local entities, possible penalties or even a shutdown, and losing the AI advantage built over years.
Does every new country require repeating the full process?
No. India's requirements focused mostly on financial transactions and keeping a copy, which is very different from 100 percent localization.
About this talk
Working with data teams before and AI teams now, the common thread has always been: standardize and centralize to enable analytics, insights, data products, and AI. I was deep into this path, building infrastructures meant to centralize, scale, and power enterprise-wide capabilities.
Then data localization hit.
Working for a truly global business with shared systems and processes for cost optimization, I faced a critical question: how do we decouple people, technology, and process all at once without breaking a successful business that depends on all three pillars?
In this session, I'll share the lessons I learned and the framework built to stay compliant while maintaining business continuity. You'll learn how to architect for geographic requirements without sacrificing centralization—and what actually happens to your AI when borders fragment your data.
My framework includes laying the foundational steps and getting compliant efficiently through a layered approach.
The foundational layer starts with data—introspection from data cataloging to understanding data flows. What data do you have? Where does it live? Where does it move?
The next layer uncovers the processes—where does data get transformed, where are the storage points, who interacts with it, and critically, which AI models are consuming it?
The final layer is technology, viewed through the lens of what can be retained and reused. What infrastructure is portable? What creates vendor lock-in?
Then the puzzle starts to take shape. The outcomes of these three layers help you define a cloud-agnostic, vendor-agnostic, compliant data platform that can continue powering local businesses while keeping AI dreams alive for global organizations.
Now, as more countries view their digital infrastructure and citizen data as sovereign assets to protect, this same formula can be quickly replicated without massive disruption or excessive costs. You're not starting from scratch each time—you're applying a proven framework that scales geographically while maintaining the centralization benefits that make modern data and AI possible.
Start with capabilities, not your current stack (73 s)
People don't move in a project timeframe (86 s)
Why your global AI must know regional channels (56 s)
Speaker: Amazing. Thank you, everyone. I hope you're all having a great Friday so far. And since you have confirmed that my audio and video are okay, I'm going to plow through. If you have any questions or if you want me to go slower, faster, please let me know in the chat. I will be monitoring the chat as well.
So first with an intro, I'm Sangeetha Parsan. I'm currently the head of product management. This is a new role and a promo that came recently. So you may see that my title was slightly different when I registered for this event. Things change fast, right? Anyway, so I'm really enjoying my new role, which was actually part of what I was doing before as well. So what is my role?
Speaker: My role sits at the intersection of data and application of data in MarTech and the products that use that data. So primarily customer data platforms, data warehouses, AI, anything, any product or any platform that is connected to data and AI. And that's where my career sits. Speaker 1: So I started as an engineer years ago, then a solutions architect architect. And now as a data leader, I want what I want to share with you guys today is a dilemma that I faced not very long ago. And how that made me think about how it can evolve into the framework, because it was not a problem that was one and done. It is a repetitive problem. And it's going to be for a while for all of us.
Speaker 1: So that's about me and other proud roles I have as I'm the mom of two kids, a wife and that daughter. Speaker 1: Okay, amazing. So moving on to the next slide, I tried to do the slideshow, but it was messing up with me looking at the chat. So I don't want to do that once more. So question for a room here, right? I don't know how much you can put your hand up, but if you can, please do so. How much of your job in the past has been about centralizing things? So you can, please do so.
Speaker 1: So you can have a governed central data strategy for your organization. Speaker 1: I'm going to say, make an assumption here that it's all of us, right? Um, this has been the shift the past five, 10, 15 years since the big data revolution started. Speaker 1: Centralize your data so you can glean more insights. You can understand more about your customer. You can understand what products that can power, right?
Speaker 1: That's the data strategy, right? Speaker 1: That's the data strategy. Centralize it so you can secure it better. You can apply AI on top of it. Speaker 1: And that's the direction we all have been going. Speaker 1: Now, how many of you are coming from global organizations like Informa, which is where I'm from, where you have been going?
Speaker 1: Operations in more than one country. I'm assuming at least, you know, half of the people who put up for your hand for data strategy. Speaker 1: Exactly. No, I see the small explosion there, right? We all operate in a very, very, very global world. Speaker 1: Now, it doesn't mean that we should have operations in the country. Think about it this way. It could also be the systems that are supporting our operations could be hosted in another country.
Speaker 1: That's very normal and common, right? Now, how many of you have a plan when there is a data sovereignty data localization, which is nothing but a certain government of a country saying that some aspects of the data or all aspects of my data relating to the citizens of my country or transactions happening in my country has to be saved within my borders.
Speaker 1: How many of you have seen this either coming towards you or you're already in the thick and in the middle of it. Speaker 1: This is exactly what we are going to address in this session. What do we do when such a request or requirement comes to us? Do we prepare ahead of time? How do we do it when it's already come into our plate and we are not prepared? You know, that's where I was.
Speaker 1: I'm going to quickly pause here and look at the chat. If any questions coming up, we can go into it. If not, I'm going to move into the next one. Speaker 1: Okay. Make it a little smaller so you guys can see better. So going back to the intro, right? Our North Star for data strategies used to be standardized, centralized, right? We touched upon this a little bit in the introduction.
Speaker 1: And the reason behind this is when you centralize, you can apply better standardization in terms of master data management systems, which is the record of truth or metadata systems that tell you what your data is about, right? And that in turn powers analytics where you are looking at what happened. And then insights is why did it happen? Right? And the other thing is, you know, you know, you are looking at what happened and then insights is why did it happen? Right? And the other thing is, you know, you're looking at what happened and then insights is why did it happen? Right? And the Speaker 1: And the other angle, if you think about it like a star, the other angle is data needs to be centralized in order to power data monetization and products. And similarly, with AI as well, to have that holistic view of everything that's going on. And for the models to be trained properly, we need NF data, data going backwards in history, data that is getting generated real time. We need all of it, especially with the agentic shift.
Speaker 1: And the other thing. Speaker 1: Yeah. Speaker 1: You know, the standardization. Speaker 1: Which I keep repeating. Speaker 1: The other one is also. Speaker 1: The cost of operations. Speaker 1: Right. Speaker 1: When you have multiple disjointed. Speaker 1: Stats. Speaker 1: It's very hard on. Speaker 1: Not only on the data quality.
Speaker 1: But also on the people. Speaker 1: Who are maintaining that. Speaker 1: And supporting these data platforms. Speaker 1: Hence. Speaker 1: We went into a platform consolidation. Speaker 1: Now, if you think about the AI teams. Speaker 1: In the past five, six years. Speaker 1: What we have done is. Speaker 1: We need enough data to power the models.
Speaker 1: They're very hungry. Speaker 1: Right. Speaker 1: And we. Speaker 1: Where is that data coming from? Speaker 1: Not from the siloed systems. Speaker 1: It has to be centralized and standardized. Speaker 1: For the models to be trained properly. Speaker 1: And to work properly. Speaker 1: Right. Speaker 1: This is where you turn the data into intelligence.
Speaker 1: Now, if you look at the common theme here. Speaker 1: You standardize the data. Speaker 1: Centralize the data. Speaker 1: Have a standard technology stack. Speaker 1: And that covers this data. Speaker 1: And then you have a standard technology stack. Speaker 1: That also understand their data. Speaker 1: Centralized data that also covers your AI.
Speaker 1: Am I right here guys? Speaker 1: Is this resonating with you? Speaker 1: Is this the direction you all have been going as well? Speaker 1: Okay. Speaker 1: Now, the next one. Speaker 1: So, what happened? Speaker 1: You know, why, why did I, what is this topic all about? Speaker 1: Right. Speaker 1: So we were in the same path as everybody else building a global data platform, building it to scale.
Speaker 1: And then comes personalization, sorry, localization. Speaker 1: For example, Saudi Arabia. Speaker 1: K is the kingdom of Saudi Arabia is the one that came to me in 2023. Right? Speaker 1: And I'm not going into localization itself because what I want to stay is on the framework and how we can still power our operations in when you're localizing the data.
Speaker 1: So for anybody who wants to learn about localization itself, please feel free to connect with me. Speaker 1: I'm more than happy to share any knowledge I have there. Speaker 1: Now, when you have a constraint like this, right, you have a live business also depending on it. Speaker 1: Remember, we have this data platform centralization and we have businesses operating on it.
Yeah. Speaker 1: And you have an engineering team which is very focused and familiar with the current technology stack, right? Speaker 1: Now, legal tells you, you usually have legal and privacy tells you, you usually have double digit days sometimes, not even triple digit days, right? Speaker 1: In order to get to the compliance that a country is applying.
Speaker 1: Yes, there is a lot of information that comes in, you know, it's not like an overnight situation. Speaker 1: And these laws are put into place. There is a grace period. There is a warning period. Speaker 1: But as we all know in technology, sometimes when it comes to us, it's coming in at the very last minute, right?
Speaker 1: And the final part where I struggle this is having the right vendors with the right skill set. So this was the situation that we were in, right? Speaker 1: Now, once you start working on it, this is the framework. We'll go into the framework, right? Speaker 1: Where were the problems when we started working on it, right?
Speaker 1: You think you know the data because you have centralized it, but mapping it into what needs to be localized took us a lot more time, right? That sort of exposes the gaps you have in your metadata systems, which give you an understanding of the data and then translating between a legal and privacy requirement to an engineering requirement.
That's where product comes, right? If you think about it, there is someone that needs to do that translation as to, well, data needs to be localized to what data? Is it attendee data? Is it exhibitor data? Where does the data sit? And what needs to be exactly localized? Is the data all time data or is it recent data or is it data in transit?
There are so many questions that come up, right? Speaker 1: Now, we talked a lot. I mean, I must have used 1,000 times data and a platform in my context so far, but we forget about the process, right? What processes and flows are using this data outside and what processes are bringing this data into an ecosystem? That needs to be thought through, not just the central one.
And finally, people readiness. Yes, you can do this in double digit dates. You know, anything is possible in the engineering world, right? You make it happen. But sometimes we forget about the people that need to support it ongoing, the people that need to use it on the new stack and what training and change management that is needed there.
Isra, I will come to your question. If you're okay giving me two minutes, I'm actually going to cover that. Thank you, Isra. That's a very insightful question. Okay. So, where do we start, right? Like everybody else, I also started with, yeah, let's pull up. Well, there is already a central data platform. Let's put up the architecture diagram and see how we can, you know, pull things apart and deploy it in different places.
Then we start to make technology decisions. Oh, we should deploy it on this stack versus that stack. We should do this versus that, right? But is that what we want? No. Just remember things, right? It takes about, you know, two years or three years to set up a central technology stack. Things might be changing in that time, right?
So now start, rather than just taking what you have, the blueprint you have, and then pulling that apart to make a new ecosystem or target state. Think about what are the capabilities that are still needed? Yes, when we were building it, we would have built it for X, Y, and Z, right? Are they still needed in the region?
What must be working locally, right? Coming to that example of a CRM, when we got to the bottom of it, we knew that, well, it's not really about exhibitor data. Informa is, we are a conference and trade show. We are in the event business. So we have the attendee side and we have the exhibitor side. So you get some more clarity.
Okay, well, it's not affecting my CRM, right? It's only affecting where an attendee is registering for an event. So you find out what capabilities are there first. Don't start with your text tag. This is a lesson I learned in a hard way, right? Then you map the technology that is serving these capabilities. Understand what are the business operations in the region and then map it to the technologies.
Then you can find out what of these technologies are available in the country. It's a registration vendor who can capture, store the data already there in the KSA region. Then you map that to that vendor, right? Think about the capability first and then see who is operating in that space. For example, in my case, marketing automation was a capability.
And what we found out that Oracle was able to provide a local solution for us where the data will be stored within the region. So then it was a one to one map, right? We already know this vendor. We know this capability. Then it's easier for me to work with Oracle and say, okay, let's start working on this local solution, whatever that is.
Yeah. Now you have a foundation in terms of your target state architecture. You're not looking at a blueprint that was created either two years ago or three years ago. You have your capabilities today and then you go into it. Right now I'm going to pause and answer Isra's question here. What is the first step organization should take when preparing for data localization requirements?
That's perfect and that's exactly what I've covered here, Isra. Think about what capabilities you need in that region. What data needs to power those capabilities? What technology stack do you need? And then you go from there. Don't start with what you have centrally. That could end up misleading you because it's built for the entire global organization, not for one region.
Yeah. Okay. I hope that answered your question, Mr. Isra. So then what is the practical model that I'm suggesting? Right. It is not a theoretical model. I want to be very clear of that because I know frameworks can be very hypothetical. This isn't. The order of operations that actually worked for me when the localization hits suddenly is what I'm presenting.
Presenting there. Yeah. Start with the technology part. What are the capabilities you need for your business to run? What are the current technologies that are powering these capabilities? Right. What are the technologies that can available within the country that satisfy the localization needs? Then you come with your target architecture. So, again, you know what you need to fulfill.
The what part is what you're fulfilling with the technology. Now you come to the data, the second pillar. Right. The data. First, start with understanding the compliance policies. And this is the devil in the details. This is where most of your focus should go as data professionals. Again, this is not a one person effort. It is really a team of effort with different skill sets coming into picture for this initiative.
If you are a data professional, then your focus will be on pillar number two. Right. Pillar number one is usually an enterprise architecture solution architecture sort of effort or business architecture even. Now you come into the data pillar, then you're talking about what data is really stored in the region today and now. And create a data compliance map.
This is where all the standardization and centralization you did previously will come into play. You will have metadata about every table, every column in a table, every field. Right. That can cover it now. If you don't have it, it's not a big deal. There are so many AI products today. That's another session for another day. But there are so many products today that can give you this level of insight in a matter of minutes.
Now that's the revolution we have seen in the past three to four years. Right. So create that compliance map first. Now you have two pillars done. What technology I need and what data I have and what data should be compliant. Now that will actually it's actually a loop. Right. It's not a single one, but if you figure out new data points that needs to be localized, then you feed that back into the technology.
So it's more of a recursive item, not extremely sequential. Now also think about the process. What processes are bringing the data in? How do they need to change? For example, think about ID unification. Right. So we had done a central ID unification in our CDP. Now that needs to be split outside of that unification process doesn't work anymore.
What are these processes? And both processes coming in as well as the processes that are going up. Right. Somebody is taking this data and doing something with it. In my case, sending it out to sending a campaign out. Think about the end to end processes and also think about a plan. What is the compliance we need to achieve on day zero and then start to slowly build up on it for your 60, 90 and 180.
Right. You cannot really. If you have two years to do this, that's different. If you have 60 days to do this, it's different. So then you prioritize properly. So these are two levels of processes. Processes around the data and technology and processes in order to achieve a localization. So two sets of processes. Be clear how we are bringing that together.
And the final pillar, like I said, this is also the most important pillar. Yes, your business could be operating in a certain way in their current region. Just by applying, just by achieving compliance alone, you're only solving part of the puzzle. There is new skills needed because you may be introducing a new data process. You may be introducing a new technology there.
Have a change management plan, have a training plan, because this vision of localization can come in as interrupting data, interrupting operations for a business. Right. It can be seen that way. But the way you need to align with the business leaders, start top down, start with the business leaders and make sure that everybody aligned is on the vision that this is a necessary evil for the business to operate in that region.
This is not an option. And once they buy into this vision, then change management becomes much easier. OK, so now I have like another 15 minutes, but I'm going to wrap this up in the next five to six minutes. So there is more time for questions. So then where do we start with the technology? I've been covering this as we spoke, capabilities, what technologies are available in the region.
This is where your infrastructure people will come into picture. Right. What cloud providers can operate locally? What are the costs for them if they don't already have that infrastructure in the region? You need to think about what those costs are, then contrast and compare. Right. And then think about what SaaS tools have data residency options. So this is where you go deeper and deeper into all the way into hosting.
Right. For example, you know, many of us are familiar with AWS and the regions, the hosting regions. Now you go down to that level of saying previously you would have thought about, OK, you know what I'm using, for example, Amazon Airflow or something like that. You might not have thought about in depth about the technology behind it and where it is storing the data.
And now you have also have to think about what data is in transit versus what data is at rest. Now, once you meet once you identify those technologies and then you map it into what the target has available and what is feasible, then you create a target state architecture. So this, like I mentioned, is a collaborative effort between the business where you identify what's needed and enterprise architecture and infrastructure who will put that picture for you together.
And then the next one is a data. Like I said, the first stakeholder for you in this should be your legal and privacy team. They are the ones well versed with what requirement needs to be met. What are the time frames and what exactly does it break down to? Right. When we say customer data. Absolutely, Israel, you're you're very spot on.
It is. But it's about bringing all these entities together is where you will see success. Right. Now, if you have a data catalog, we talked about that. Right. Meta data, et cetera. It's great. If not, first create that and create a map. Right. Create a map of this is the existing data I have. And this is the localization data that needs to be localized.
And this is where your legal and privacy team will come in very handy. The governance team will come in handy. Right. That way you create a perimeter for it. In my case, like I mentioned, exhibitor data did not have to be central. I mean, localized on day zero or day one, which was great. Then I was able to focus on the most critical speaker or attendee data that had to be localized.
Then the third pillar we covered is the process. Right. Identify how every data related process is going to change. And the key factor here is prioritize the impact, not by the difficulty level. We all tend to do that. Let's handle what is most simplest to do. And then we will go to the most complex one. I ended up going in that direction, but we course corrected very soon.
Right. No, that's not going to help us. Go with the impact. Which one has the highest impact? And then you order your priority. For example. Right. Day zero. What is the must have? Critical in scope data, mood or isolated. No violation. Go live. Day 60. You are target the secondary processes. Day 90. Look at all the edge cases.
Right. And day one, one 80. You have full maturity and a playbook ready for the next country. Yeah. Then the people don't forget about them, please. They are the linchpin in this. You can deploy, redeploy technology. You can migrate data. Everything can happen in a project timeframe. Right. But people don't move in a project timeframe. That's why, like I mentioned, start top down with the end users.
And also at the same time, keep thinking about what capabilities you need. One example for me is we were almost halfway into this migration. Right. You are so tied up in getting the technology ready, data migrated, and making sure you achieve compliance for the business that you tend to forget what happens after that. We were very cognizant about it, thanks to the great team that we have.
We started working on a managed services support model midway into this initiative. It really took us that reminder of the time to put a structure in place. You will also have a hypercare period. Think about that, right? Because you're not going to transfer it over immediately into your operations. You'll definitely have a hypercare period in between, whether it's four weeks or six weeks.
Lessons learned from me. Have a longer hypercare period. And because this is the first time you're doing it, you're learning. The business is learning. Technology is learning. So take your time. Don't rush hypercare. Right. And I think it also helps with the change management for people, knowing that they don't have to put a ticket and wait for the next 40, you know, 40 days.
They have someone to talk to, whether it's a chat or whether it's a Slack or whether it's an email. They have one person they can reach out to or a team they can reach out to. That goes a long way in change management. Okay. So finally, we come to the most interesting bit. Well, we are AI strategy was always built on centralizing data.
Right. But localization pretty much cuts through that. It goes into fragmenting that data. Then how do we grow AI if our central ecosystem is getting fragmented into 10 different regions? So what really breaks? One is, we talked about it, fragmented training data, having enough data for training. And also that if data is pretty much looking at one region, you may miss the nuances that are happening in the other regions.
Right. That's the second one, the inference quality. And then the governance audit. If you have like 10 different models running around, it becomes very hard. Like Isra mentioned, governance is very hard. Now, I'm not the expert in this. I'm still figuring this out because the AI is relatively new for all of us. Right. So how do you fix this?
One is by design. You may end up having local models, you know, small language models for your local data, while you have your large language for the global one. Right. And you may have to train on local data. But you don't really train on the data itself, but on the gradients of that data. So that way, your global models are aware of the nuances in the region.
One simple example. Right. So I'm in the MarTech world. So if you think about how we reach our people. Email is a very common method of communication in North America. But SMS and WhatsApp and WeChat are very, very, very common in the other regions. Right. Like Asia. So this is a small nuance. But it can become a big one when you think about a model who doesn't understand this nuance or doesn't know about the nuance.
And you're asking it to design you the best campaign strategy. If it doesn't know about it, you're missing the massive channel where you can reach your audience. Right. And hence, you need that federated learning. It doesn't know the SMS number of the person you reached. But what it needs to know that SMS is a very viable channel in the next region or Y region.
Now, going into the data strategy itself. Again, it goes into that federated learning aspect as well. Right. It's only the key personally identifiable information and financial transactions and things like that that needs to be normally localized. Not everything. So architect for what you can move. Right. So the strategy is keep your global model. But feed it as much as it is allowed and you can about your local data.
And also keep your local models alive. Yes, it takes effort. It takes time. So this is where you need your AI governance to come in. Anyway, we are almost towards the end of it. Like I mentioned, the problem in this case, the full playbook is that we had a problem. We had a global platform. But then the mandate was, yeah, well, the citizens data cannot leave it.
We had a live business depending on it. We went through all the four pillars that we discussed. You map the capabilities. You select a local infrastructure. You make sure you map your data properly. Build your infrastructure. Design your processes. You train your team. Right. The outcome of this was that we have a purpose-built stack. Global operations continued.
There was no interruption to it because you're separating data out of the central ecosystem. Local business was fully operational. And then now we sort of have a replicable playbook as much as we can because the requirements, the localization requirements are not the same in every region. Anyway, as I promised almost towards the end, remember I said Saudi Arabia.
Immediately after that, there were two regions that hit us one more time, India and Vietnam. Right. So we are working through that. In some cases, we didn't have to go through this whole process again because for India, the requirements were mostly towards financial transactions and having a copy. And that's where, like I mentioned, privacy comes in. It's different when you are allowed to have a copy versus you have to have 100 percent localization.
Anyway, what is the cost? I mean, why are we going through all this trouble? Right. Why can't we not be compliant? No, that's not an option. You lose trust with the local entities. You may face penalties. You may face a shutdown as well. That's a looming threat. And then you also have the A capability loss that we discussed.
Right. You lose the A.I. advantage that you spent four years or five years doing. Anyway, again, the repeatable formula. We already saw this is about understanding the capabilities, get the data compliance. Think about the horizon horizon process and make sure you are also tying. But there was a reason why I had a thread tying all of this.
In some cases, it's not very sequential like this. It has to feed back into each of the other pillars. That's about it. Thank you, everyone, for your time. Not easy on a Friday, but I'm happy so many of you joined. Thank you. Now I'll go into the questions. Isra has one. Okay. It's not a question. Yeah. Thank you, Isra.
It is a very complex topic. You're absolutely right. And I'm not oversimplifying it here. But all I'm saying is, this is the approach that worked for me. Any other questions? Going to stop sharing now so I can watch one screen, which is the question screen. Okay. If that's the case, then I can give you all your five minutes back so you can get ready and go into the next session.
Thanks, everyone. Really appreciate your participation.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
How do you meet data localization laws without breaking global data and AI?
Sangeetha Parsan shares the capabilities-first, four-pillar playbook Informa used when Saudi Arabia required citizen data to stay in-country.
In short
Data localization laws turn the standard data playbook upside down: years of centralizing data for analytics and AI suddenly meet a government requirement that citizens' data stay inside national borders. Sangeetha Parsan argues that the answer is not to pull the existing architecture apart, but to work through four connected pillars (technology capabilities, data, process and people) with a phased plan that becomes a reusable playbook for the next country.
Key takeaways
Strategic approaches to building compliance-ready technology at global scale
When centralization meets localization
Parsan opens by asking the audience how much of their work has been about centralizing data. Since the big data revolution, she explains, the North Star has been to standardize and centralize: master data management and metadata systems, analytics and insights, data monetization and products, and AI models that are hungry for historical and real-time data. Platform consolidation also lowered the cost of operations and eased the burden on the people supporting fragmented stacks.
Informa, a global conference and trade show business, was on the same path, building a global data platform to scale. Then, in 2023, the Kingdom of Saudi Arabia's localization requirement arrived. Parsan describes the constraint many data leaders will recognise: a live business running on the central platform, an engineering team familiar with the current stack, legal and privacy teams who often give only double-digit days to reach compliance, and the struggle to find vendors with the right skills.
The hardest surprises were about understanding, not building. "You think you know the data because you have centralized it, but mapping it into what needs to be localized took us a lot more time, right?" (9:54) Someone, usually product, has to translate a legal requirement into an engineering one: attendee or exhibitor data, all-time or recent data, data at rest or in transit.
Start with capabilities, not the current architecture
Parsan admits her first instinct was to pull up the central architecture diagram and decide where to redeploy each piece. She calls this a lesson learned the hard way. A central stack takes two or three years to build, and needs change in that time. Instead, she asks which capabilities must work locally. At Informa, closer analysis showed that the CRM was not affected; attendee registration was. Only then does she map the technologies serving those capabilities and check which are available in-country. For marketing automation, Oracle could provide a local solution with data stored in the region, which made the mapping straightforward.
Answering an audience question about where to begin, she was direct: "Don't start with what you have centrally. That could end up misleading you because it's built for the entire global organization, not for one region." (15:05)
Data maps, processes and a phased plan
For the data pillar, Parsan names legal and privacy as the first stakeholders, followed by governance. The goal is a data compliance map that sets a perimeter: what data exists, and what must be localized. An existing data catalog helps; if there is none, she notes that AI products can now generate this level of insight quickly. In Informa's case, exhibitor data did not need to be localized on day zero, which let the team focus on the most critical attendee and speaker data. She stresses that the pillars form a loop: newly discovered data points feed back into technology decisions.
Processes come next, both those bringing data in and those sending it out, such as central ID unification in the customer data platform or campaign sends. Her rule for ordering the work is clear: "And the key factor here is prioritize the impact, not by the difficulty level." (23:49) She admits the team first drifted toward the easiest items and course corrected. Her plan runs from day zero (critical in-scope data isolated, no violations, go live) to day 60 (secondary processes), day 90 (edge cases) and day 180 (full maturity and a playbook for the next country).
People do not move on a project timeline
Parsan calls people the linchpin. Technology can be redeployed and data migrated within a project, "But people don't move in a project timeframe." (24:40) She recommends top-down alignment with business leaders, so that localization is understood as a requirement for operating in the region rather than an interruption, followed by training and change management for new skills, processes and tools.
Halfway through the migration, her team started designing a managed services support model, and she urges others to plan for hypercare early. "So take your time. Don't rush hypercare." (26:06) Giving users one person or team to contact, instead of a ticket queue, goes a long way in change management.
Keeping AI alive when borders fragment data
Localization cuts straight through an AI strategy built on central data. Parsan lists three things that break: fragmented training data, inference quality that misses regional nuances, and governance and audit across many models. She is candid that she is still working this out, but suggests local small language models alongside global models, and federated learning on gradients rather than raw data. Her MarTech example: email dominates in North America, while SMS, WhatsApp and WeChat are very common elsewhere, a nuance a global campaign model must know without seeing anyone's phone number. Only key personally identifiable information and financial transactions typically need localizing. "Not everything. So architect for what you can move." (29:26)
The outcome at Informa was a purpose-built stack, uninterrupted global operations, a fully operational local business and a largely replicable playbook. India and Vietnam followed, with India's rules focused mostly on financial transactions and keeping a copy. "It's different when you are allowed to have a copy versus you have to have 100 percent localization." (31:36) Non-compliance is not an option, she concludes: it risks lost trust, penalties, shutdown and the AI advantage built over years.
Watch the talk
"Localizing in the world of global operations" by Sangeetha Parsan (34 minutes): https://www.womentech.net/video/localizing-in-world-global-operations
About the speaker
Sangeetha Parsan is Vice President, Customer Data Solutions and Services at Informa, with more than 20 years of expertise designing and implementing data, technology and data integration solutions for B2B organizations. She heads a team of data experts focused on integrating businesses onto data platforms, particularly in complex corporate mergers and acquisitions.
Questions this talk answers
What is the first step an organization should take when preparing for data localization requirements?
How should a team prioritize localization work under a tight deadline?
What happens to AI when data has to be localized?
Why can't a company simply stay non-compliant?
Does every new country require repeating the full process?
Drafted with AI from the talk transcript. Editor review pending.
Sangeetha Parsan. "Localizing in the world of global operations." WomenTech Network, 2026. https://www.womentech.net/en-au/video/localizing-in-world-global-operations