OpenNashPodcasts

AI in Financial Services

Becoming Truly Data Driven - with Tanvi Singh of UBS

1 Aug 2022 · 26m · Tanvi Singh, Daniel Faggella · 22 quoted facts

▶ Play episodeOriginal source ↗

In brief

Becoming data-driven takes more than spending on big data and AI. Tanvi Singh argues that banks need a shared data platform built on a data mesh, in which each business area owns its data as a product. Her case is that central data teams become bottlenecks, and that vendor lock-in and technology-first projects waste money.

The central tension is between a long-term cultural and architectural shift and the near-term results sponsors demand. Singh's answer is to start from specific use cases, such as spotting unauthorized transactions, while building standards that scale. She also stresses planning for the full model lifecycle, including shutting models down.

Key insights

  1. Most firms invest in AI without a data culture

    Singh cites a NewVantage Partners survey of about 500 organizations: under 25% said they had a data culture, yet 99% were investing in big data and AI. She reads that gap as a sign that the way these investments are being executed is not working.

  2. Data should be treated as a product

    Singh says each business domain, such as clients, should own its data along with its metadata, compute and APIs, much like microservices in software. Ownership by the domain, rather than a central team, is what she calls the core principle of data mesh.

  3. Start with use cases, not infrastructure

    Singh warns that firms which built Hadoop-style infrastructure and pulled in data first left sponsors unhappy, because nothing useful came out. She advises solving one specific question, building standards that scale, and only then moving to use case two and three.

  4. AI projects have four lifecycle phases

    Singh describes a sandbox with near-real data, a production stage with security and access control, MLOps (keeping pipelines from breaking), and a final retirement stage. Few firms plan for retirement, so some run expensive cloud pipelines with no useful output.

  5. Avoid vendor lock-in, especially under GDPR

    Singh says buzzwords and vendors change every few years, so black-box solutions make it hard to swap or add providers. She also says European firms should check GDPR, the EU data protection law, before tying themselves to non-European vendors.

Questions

What is a data mesh and why do banks want one?

Singh describes it as a distributed setup in which each business domain owns and publishes its own data as a product, instead of everything going through one central team. She says this avoids the bottlenecks and repeated work that central teams create.

Why do companies spend heavily on AI but still lack a data culture?

Singh cites a NewVantage Partners survey in which 99% of firms invested in big data and AI but under 25% reported a data culture. She says the problem lies in how the work is executed and in how data-literate the organization actually is.

How should a bank start without wasting money on technology?

Singh says to pick a specific problem, such as finding unauthorized transactions, and test it in a sandbox using near-real data. Governance and standards should be set up so they can be reused for the next use cases.

What happens to AI models after they go live?

Singh says they need MLOps, meaning operational upkeep so pipelines don't break, and eventually a planned shutdown. She notes that few organizations plan for shutdown, so costly pipelines keep running with no value.

Should a small company build a data mesh too?

Singh says no, not necessarily. For small and medium-sized firms she says a lakehouse from a single provider, which combines a data warehouse and data lake, can deliver much of the value.

Facts

Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.

Tanvi Singh spent nearly eight years at Credit Suisse, most recently as Managing Director of Digital Transformation and Product Labs.

“At the time of this recording, she was wrapping up nearly eight years serving with Credit Suisse, most recently as Managing Director of Digital Transformation and Product Labs.”

0:00:05 · Daniel Faggella · Guest background

Tanvi Singh is now group head of the ADA Foundations Hub at UBS.

“She is now the group head of the ADA Foundations Hub at UBS.”

0:00:05 · Daniel Faggella · Guest background

Tanvi Singh previously worked on analytics at General Electric and Monsanto.

“She previously worked with giants like General Electric and Monsanto, also on the analytics side of things.”

0:00:37 · Daniel Faggella · Guest background

Tanvi Singh earned a master's degree in software systems at the University of Zurich.

“And before that, received her master's in software systems at the University of Zurich.”

0:00:37 · Daniel Faggella · Education

NewVantage Partners published a report surveying about 500 mostly medium-sized and large organizations about data culture.

“So very recently, New Vantage partners pulled out a report for some 500 different companies, organizations, most of them medium-sized, big-sized, have participated.”

0:03:05 · Tanvi Singh · Data culture survey

In the NewVantage Partners survey, fewer than 25% of firms said they had a data culture, while 99% were investing in big data and AI, according to Singh.

“And the surprising part of the story was that about less than 25 % would say that they have data culture, while 99 % of them are investing in big data and AI.”

0:03:05 · Tanvi Singh · Data culture survey

Singh says firms ran separate data warehouse and data lake platforms, and that data warehouses have existed for decades.

“in our world that used to be data warehouse and the data lakes. So the data warehouse infrastructure was never a very interesting part of the conversation because it's been there for decades.”

0:04:45 · Tanvi Singh · Data warehouses

Singh defines a data platform as one that treats data as a product rather than as a service or an asset kept by individual teams.

“A platform per se is then the right definition of creating data as a product and not data as a service or data as an asset which people tend to keep to themselves”

0:05:41 · Tanvi Singh · Data products

Singh gives the example of a client data product that bundles the data, its metadata, compute, processes and APIs under one domain owner.

“So if you are responsible for the client division in an organization, then you're talking about a client as a data product where the data itself, the metadata supporting it, the compute of it,”

0:06:10 · Tanvi Singh · Data products

Singh says small and medium-sized organizations can adopt a lakehouse, which combines warehouse and data lake functions, instead of a distributed data mesh.

“For a small and medium-sized organization, it's pretty much okay to move into a lake house solution, which is a very technical way of solving the problem”

0:08:44 · Tanvi Singh · Lakehouse

Singh names Databricks and Snowflake as vendors that have tried to offer both warehouse and data lake capabilities.

“A lot of cloud vendors out there, Databricks I can name, Snowflake I can name, have been trying to address being a warehouse provider, trying to be a data lake provider”

0:08:44 · Tanvi Singh · Lakehouse vendors

Singh says European organizations face GDPR challenges when choosing vendors that are not based in Europe.

“Most of you who are in Europe would also have a lot of GDPR set of challenges to deal with.”

0:11:28 · Tanvi Singh · GDPR

Singh says some organizations invested mainly in Hadoop infrastructure and data ingestion, leaving sponsors unhappy because they saw no output.

“They think of Hadoop as the newer technology. So it put most of their investment into setting up the infrastructure, pulling in the data, which makes the sponsors really, really unhappy”

0:11:54 · Tanvi Singh · Hadoop

Singh predicts SQL-based tools for semi-structured data, comparable to those for structured data today.

“What I feel in the future would be coming in would be a very SQL-based set of solutions for semi-structured data as we see it for structured data today.”

0:10:37 · Tanvi Singh · Future technology

Singh says a financial crime use case such as finding unauthorized transactions should be tested on pre-production data rather than synthetic data.

“You're looking into the transactional data. The most important thing to do before you put in any of the regression model or think about using any of the deep learning model is to just have the pre-production data”

0:14:50 · Tanvi Singh · Financial crime

Singh describes MLOps as operationalizing machine learning so pipelines don't break once a model is in production.

“The third area of most of us in the AI world do not play a lot of focus on is the MLOps, which is being able to operationalize your ML.”

0:16:34 · Tanvi Singh · MLOps

Singh says very few organizations plan for the retirement phase of a model's lifecycle.

“Even more important to talk about is the phase four, which I've seen very few organizations even having on their roadmap to say is what happens when a model has gone through its lifecycle?”

0:16:59 · Tanvi Singh · Model lifecycle

Singh says she has seen organizations run costly cloud pipelines that produce no valuable output.

“So I've seen organizations running expensive pipelines on the cloud, which doesn't even have a valuable output.”

0:17:26 · Tanvi Singh · Cloud costs

Singh says the sandbox phase mainly involves subject matter experts and machine learning engineers, while later phases involve data engineers and operational teams.

“For Sandbox, you usually have a lot of very excited set of SMEs together with the machine learning engineers trying to experiment everything another way, right?”

0:19:00 · Tanvi Singh · Team roles

Singh argues that centralized data teams become bottlenecks and that decentralizing ownership is key to a data-driven culture.

“I think the decentralization of power, the centralized teams do not work. They just do not work because they kind of become a bottleneck.”

0:22:07 · Tanvi Singh · Data mesh

Singh predicts that over the next three to five years the data vendor landscape will simplify into more standard products.

“a lot of the data landscape or the technology landscape, we see that a lot of different vendors is going to simplify.”

0:24:35 · Tanvi Singh · Vendor landscape

Emerj offers an AI and financial services executive brief at emerj.com/fin1.

“download our AI and financial services executive brief. You can go to emerj.com slash fin1.”

0:25:37 · Daniel Faggella · Emerj resources

Chapters

  1. 0:03:05Why firms need a data platform

    Singh opens with survey evidence that heavy AI spending isn't producing a data culture, and defines a platform built around data products and a data mesh.

    “less than 25 % would say that they have data culture, while 99 % of them are investing in big data and AI.”
  2. 0:11:28Lakehouses, vendor lock-in and problem-first design

    She contrasts smaller firms' single-vendor options with larger organizations' needs, and warns against lock-in and building technology before a problem.

    “never get yourself locked into one vendor. never get yourself locked into one cloud platform.”
  3. 0:14:19Balancing quick wins with the big picture

    The hosts discuss how to deliver near-term value without piling up isolated point solutions, and Singh begins describing the sandbox approach.

    “So what's really important, as you said, is to derive value out of the use cases”
  4. 0:17:26The four-phase AI lifecycle

    Singh walks through sandbox, production, MLOps and retirement, and notes that different teams and sponsors are involved at each stage.

    “how do we shut down certain things? So I've seen organizations running expensive pipelines on the cloud”
  5. 0:22:07Decentralizing to change data culture

    Singh argues central data teams become bottlenecks and that domain ownership, with a reading recommendation and a vendor-consolidation forecast, will move culture over the next three to five years.

    “I think the decentralization of power, the centralized teams do not work.”

Transcript

0:00:05
Daniel Faggella

This is Daniel Faggella and you're listening to the AI and Financial Services podcast. In this episode, we talk about becoming data-driven in the world of financial services, and we have a guest with a robust experience in doing exactly that. Tanvi Singh joins us this week on the show. At the time of this recording, she was wrapping up nearly eight years serving with Credit Suisse, most recently as Managing Director of Digital Transformation and Product Labs. She is now the group head of the ADA Foundations Hub at UBS.

0:00:37

She has a robust experience working with advanced analytics and data and seeing what it looks like inside of very large organizations to level up their analytics capabilities. She previously worked with giants like General Electric and Monsanto, also on the analytics side of things. And before that, received her master's in software systems at the University of Zurich. There are two important takeaways from this episode that I think will be useful for everybody tuned in. The first is a view of the world of data and the future of financial services, where that data can actually be leveraged for AI value, can drive value for customers, and can drive operational efficiencies.

0:01:13

What does it take to get there, and what does that picture look like? How connected is it? How is it governed? And how are smart organizations going about it strategically? Tanvi certainly has some perspective there, and again, has up close and personal experience working on these kinds of projects. So she paints her vision of how a data mesh can come to life and how it can provide the right kind of access and bypass some of the political silos that we often see in the enterprise.

0:01:37

The second half of the episode is all about how do we build that robust data mesh? How do we genuinely become data-driven in a way that also sees some near-term value? Because as most of you tuned in are well aware, companies are not interested in investing in abstract capabilities. They have to be able to see some turnaround, some evidence of results or success along the way. And Tanvi paints a picture of how to bear both in mind, both near-term value and long-term value in the long lumbering path from where banks are today to a place of real digital fluency.

0:02:09

So some interesting perspective from somebody close up to these problems and hopefully an episode that's helpful for all of you tuned in. Without further ado, let's fly into this episode. This is Tanvi Singh with us here in the AI and Financial Services podcast.

0:02:29
Daniel Faggella

So Tanvi, glad to be able to have you on the program. Thanks so much for being here.

0:02:33
Tanvi Singh

Hi, Dan. Glad to be here too. Thanks for inviting me.

0:02:36
Daniel Faggella

You bet. And you're introduced by another excellent guest and those are the best kind of guests in the world. So I'm really happy we're getting to RIF today. And we're speaking on a topic that at least in financial services, I don't think we've ever unpacked just unto itself. And this is the idea of an AI platform in this industry and space. I know you sort of define an AI platform in a specific way. We're not talking about a particular vendor.

0:02:59

Can we start with maybe your definition for this platform of AI and why it's so important? I think I'll go the

0:03:05
Tanvi Singh

other way around. Why is it important? Why should we even talk about a platform for AI, platform for data, just in specific. So very recently, New Vantage partners pulled out a report for some 500 different companies, organizations, most of them medium-sized, big-sized, have participated. And there, the conversation was always about data. How much of the data culture do these organizations really see? And the surprising part of the story was that about less than 25 % would say that they have data culture, while 99 % of them are investing in big data and AI.

0:03:44

And that kind of equation, for anybody who understands maths, doesn't go really well. So something within the paradigm of what we are doing with the investments, with the focus, with the culture is not working through. And I think it boils down to how the execution of the different analytics or the AI function goes. But more importantly, how data centric, how data literate as the organization, and how much of a data-informed decision-making does happen in an organization irrespective of whether it's financial services, pharmaceuticals, manufacturing, plus plus.

0:04:18

So that brings in the concept of platform to answer your question of why platforms are important, but what exactly is a platform? Platform is a place where one can just go in and dig in the data they can trust, that they can play around with it for experimental reasons. when we talk about machine learning or AI, or to be able to use the same data to run metrics, to run KPI, to run reports and dashboards for various different regulatory reasons.

0:04:45

But if you do have to pick in and pull in different sort of infrastructures to do the reporting part of the world and then have a different sort of infrastructure to do the analytical or more proactive set of insights kind of work by using machine learning or AI, then you're creating two set of parallel platforms. in our world that used to be data warehouse and the data lakes. So the data warehouse infrastructure was never a very interesting part of the conversation because it's been there for decades.

0:05:15

And then we talked about data lakes, and that became a big conversation over the decade. Then all of us were trying to pull in any data that we had over to the infrastructure. And a big part of the investment was just going and creating this massive set of pipelines from the operational sources to the analytical sources trying to bring value into the data with very little sort of output that is meaningful that could say that the organization has a rich data culture.

0:05:41

A platform per se is then the right definition of creating data as a product and not data as a service or data as an asset which people tend to keep to themselves but more data as a product, just as the way we are very used to the microservices in the operational side of the world to use data as a product in a similar way into the data insights, into the analytical pieces of the world. To be able to create a very domain-specific ownership of the data set.

0:06:10

So if you are responsible for the client division in an organization, then you're talking about a client as a data product where the data itself, the metadata supporting it, the compute of it, and All the associated sort of processes and APIs are very much part of a data product where you are trying to get your customers excited about your data product as much as you're excited about exposing any of the product as part of your business output.

0:06:36

And more important, when we talk about the platform, it's also the self-serve part of the platform. How well can your platform service the various different capabilities of the different actors or different personas playing, which I said about deriving APIs, pulling out reports, dashboards, or data scientists pulling in machine learning set of algorithms out of it. And finally, some really fine-tuned set of AI that you're exposing to your customer and that are adding value to the overall business growth.

0:07:09

So, I mean, if you would hear me seeing all these different topics, I'm pretty much talking about the new fundamental way of approaching data and being able to use a very federated set of distributed architecture, which in today's term is called the data mesh. So that's a very exciting way or a very new sort of paradigm to think about data to bring in the maximum amount of value based on the investment that goes into any of the data programs in bigger organizations.

0:07:39
Daniel Faggella

Big time. And you've seen it over the years. We certainly have across all sorts of different industries. AI has gone from this sort of band-aid point solution, plug it in where we might need it, to a genuine understanding, at least among some, certainly not among most, that this is a capability we are to build. And this is to unlock the value of data to really be able to make some bigger picture changes and truly derive value from these things as opposed to focus on these little pocket solutions.

0:08:08

you are definitely talking about the big picture and the strategic changes. You had mentioned paradigm shift. And because it's a paradigm shift, Donvi, you know as well as I, it's tough to move a whole paradigm at once. It's going to have to be moved in pieces. And you've worked at different organizations that are maybe farther along or not as far along or what have you along this journey. When you think about what an enterprise has to go through to move towards this greater data mesh, to move towards this platform where data is awake and they can be nimble with using it in a productive way, Where do you often see first steps or what do you recommend for leaders?

0:08:42

It's such a big topic. I'll let you take it wherever you'd like.

0:08:44
Tanvi Singh

Great question, Dan. Again, very important to emphasize that I'm talking about the overall paradigm and one can think about platform to just take a little bit forward from our last conversation. For a small and medium-sized organization, it's pretty much okay to move into a lake house solution, which is a very technical way of solving the problem of having a similar data infrastructure for a warehouse kind of solution plus a data lake kind of solution. A lot of cloud vendors out there, Databricks I can name, Snowflake I can name, have been trying to address being a warehouse provider, trying to be a data lake provider for all the ML kind of projects and vice versa.

0:09:26

All the three big cloud platforms trying to solve that challenge as well. If you're a smaller organization, it necessarily doesn't have to be a distributed data mesh solution that needs to be implemented because you can get a lot of value of the data by outsourcing a lot of that sort of effort to one single technical provider based on your needs. But when it comes to organizations that have already gone underway, creating a huge set of data in different parts of the organizations, but having a problem to interconnect that sort of information and having to redo the same set of data exercise over and over again to answer very different set of questions using the same data domain.

0:10:09

For those set of organizations, it's important to think of data problem at an organizational level. But before we get into the complexity of that, I think one of the most important topic that I talk about is to not get locked in into a vendor. Because that's the challenge. As you kind of see the collage of different vendors over the last year, especially over a decade, it keeps changing the buzzwords from one to another. So never get involved yourself into a black box solution.

0:10:37

where it's very difficult for you to enhance, expand, get rid of the vendor, have another vendor, or have them do a part of the piece that they do very well and augment it with another set of vendor, a new set of technology coming in. So, for example, what I feel in the future would be coming in would be a very SQL-based set of solutions for semi-structured data as we see it for structured data today.

0:11:01

So a lot of the complex data that would fall into that process would then be able to be answered for all of your predictive models using SQL without having to worry about ML or AI set of engines. So to be able to embrace the set of change and the amount of investment that has gone to all that different space of creating better data products, give yourself that way so you can enhance and embrace all of those set of technologies as it comes by.

0:11:28

So never get yourself locked into one vendor. never get yourself locked into one cloud platform. I think that's equally important. Most of you who are in Europe would also have a lot of GDPR set of challenges to deal with. And it's very important to check those set of standards before having a lock-in with any of the vendors that are not based out of Europe, which most of them are not. So those are the set of very simple checkboxes that are needed.

0:11:54

But the most important thing as you get yourself more and more involved into the data platform is to try and solve for a problem and not create a technology and then look for problems later. That's a big mistake a lot of organizations have seen in the past have done. They think of Hadoop as the newer technology. So it put most of their investment into setting up the infrastructure, pulling in the data, which makes the sponsors really, really unhappy because they don't get to see the output.

0:12:21

So create an infrastructure, create a platform that's able to answer a very specific question while you create the standards and the governance in a way that can be scaled up after solving use case one to use case two to use case three. That's the most fundamental way of augmenting a platform one over the other. Only bring in the data over to a platform of your choice that can derive value for you and be responsible for the data quality of the data and not just the usage of the data, which is usually very challenging.

0:12:54
Daniel Faggella

Yeah, and this is bringing us to really one of the most pivotal kind of seesaw points in AI adoption in the enterprise in general. We opened up here today with this paradigm shift idea of really making an entire enterprise more AI nimble, being able to wake up this data, apply it to fulfill our customers' desires in new ways, to operate maybe in new ways, to operate in more efficient ways, what have you. And now you're bringing up another point, clearly very important, which is, hey, let's find areas where we can really add value.

0:13:29

Let's adopt these technologies where we have a business case and we can go to a sponsor and we can show a result and we can show real progress. sometimes those are counter balancing forces in some way in other words if we focus on an individual point solution the easiest way to make a boss happy is find the smallest most surface level data often that we can get to and be able to create some kind of a dashboard or some kind of functionality that can maybe prove something that we can measure financially as soon as possible with no literally no attention paid to strategy or to the building of ai capability but if we tack on 30 or 40 of those

0:14:05

onto an enterprise, obviously, we can get sort of bogged down. Obviously, you're able to keep both of these things in mind. How do we find those areas to add value, but also develop them in a way that keeps this bigger picture of moving towards a

0:14:19
Tanvi Singh

data mesh in mind? Great question, Dan. I always think about a full boxer in that case, right? So what's really important, as you said, is to derive value out of the use cases as the first set of exercise in a paradigm. Anybody in the domain, whether they, and I can speak much more at length for financial services, so let's take an example of somebody sitting in the compliance organization. You were to figure out where is an unauthorized transaction happening, which is a classic case of financial crime.

0:14:50

You're looking into the transactional data. The most important thing to do before you put in any of the regression model or think about using any of the deep learning model is to just have the pre-production data because you really need to have access to much closer set of data and not synthetic data to look for anomalies and transactions in a sandbox environment. So once you're able to get an answer to come out in a sandbox environment, and in this case, the bosses or the SMEs or the compliance officers kind of like, oh, really, that was the transaction I was looking at, but I was doing it in a manual way or using it in a very rule-based way,

0:15:24

so your algorithm somehow is working. That's an innovative case. And if that set of sandbox approach of yours has been successful, and most of the time, the sandbox approach shouldn't be successful. So it's very important that failure is an option when we talk about this first set of iteration towards our journey to a more mature AI methodology for an organization. And if any of the experiments within the sandbox are successful, and the reason why I call it a sandbox and not an experiment or a POC, the proof of concept of the standard way of doing software engineering, because here you really need to have access to as close data to real as it is possible for most of the machine

0:16:04

learning models. And that's why we kind of call it as a paradigm within the sandbox. If any of the sandbox exercises or the algorithm proves not to be successful, it's very important to put it into an environment, which is what we call as the production environment, that the standards have to be in place, the access control has to be in place, the security has to be in place. It's very, very important for any sort of organization to continuously making sure that the sandbox environment is very, very different from a production environment.

0:16:34

And that's where what we talked about before of a platform is really, really relevant because you've got to manage certain sort of standards and governments into that space. Once we move something to production, it's also very important. The third area of most of us in the AI world do not play a lot of focus on is the MLOps, which is being able to operationalize your ML. So your pipelines aren't getting broken once it has stabilized in production.

0:16:59

They've got all the accolades, people have all their bonuses, but the model isn't working later because it's just very fragile. It hasn't been built with keeping operations in mind. And that's why MLOps becomes a really, really important topic to think about when we get into the phase three of that process. Even more important to talk about is the phase four, which I've seen very few organizations even having on their roadmap to say is what happens when a model has gone through its lifecycle?

0:17:26

How do we shut down certain things? So I've seen organizations running expensive pipelines on the cloud, which doesn't even have a valuable output. So once we go through the full cycle from sandbox to moving something into production, then maintaining something in production over its cycle and finally taking it off production and kind of putting it into its grave so it's not running anymore. It's an important set of cycle for making sure that we end up creating, you know, just a certain set of bots sitting all over with very, very brittle

0:17:57
Daniel Faggella

sort of architectural setup. Yeah. Wonder how much Amazon makes every day, nevermind every month on zombie enterprise AI applications of that kind that are sitting there without really any dashboard or output that's really benefiting any companies. People really don't talk about that fourth phase. And what I'm picking up from what you've said, there's a few things that that fourth phase is really interesting. Maybe there's more to discuss there. But one very important point that I'm noting personally, and I want to make sure that this is the proper takeaway for the listeners here, is that when we get started with a solution, we should really be thinking about all four of those phases, at least at kind of a planning level,

0:18:37

to really ask the question, is this something that's substantial enough for us to really want to sink value into? Is this something we can really plan to have be a part of our future, as opposed to spinning it up in a dark corner and having it just be a little band-aid? Because 30 band-aids does not make an AI fluent company. Let me know if you would alter or add to what I just mentioned there.

0:18:57

I'm trying to kind of sum up some of the takeaways for our listeners.

0:19:00
Tanvi Singh

Absolutely. The only thing I would add on is it talks about investments when we talk about a full life cycle over the four phases, but it also talks about different actors in the game. For Sandbox, you usually have a lot of very excited set of SMEs together with the machine learning engineers trying to experiment everything another way, right? People who understand the data and are super excited trying to solve every new problem that's thrown at them.

0:19:24

When we move into the second phase, you have people who have written pipelines that are of value, that are optimized to run on a platform infrastructure, whether cloud or on-premise platforms. And they're not usually the same set of people. They're always going to be overlaps. I'm not saying they're completely different set of teams, but a very different mindset of people, in my experience, over many, many years, having worked with data engineers and data scientists. The third one is usually when it moves over to operational teams and there are no formal data scientists being involved, primarily just the data engineers involved.

0:19:59

And the fourth one is also very important because you're talking about archive, you're talking about making sure that certain set of reports or the results that came out of the models are secured in a certain way. It's a very different part of the organization that's involved in that. So when we talk about the four cycles, it's the time period that we talk about, it's the investment. It's also the buy-in from the senior sponsors that are involved.

0:20:22

But also, it's the different parts of the organization that the different roles that have to be touched upon as we go through the cycle. It should be considered very similar to the way we think about software systems. And the moment we start thinking about data, very similar to the very standard process of software engineering and system engineering, I think we would be able to manage all these different set of conversations that I'm saying and see it in a very similar light.

0:20:48
Daniel Faggella

You know, there's certainly some ways that enterprise AI projects are wildly different from traditional IT. And we really do need to shake up the enterprise audience and get them to see those differences. But in the respect that you just articulated, we really should see it as how would we do software right? Let's make sure we have some of those same elements with our AI efforts.

0:21:09
Tanvi Singh

Absolutely. Just the way we think about microservices, we have to think about data products.

0:21:13
Daniel Faggella

Yeah, well, that's a writer downer for our folks who are tuned in. I know we've got a few minutes here, Tanvi. I wanted to touch on something that you opened with that you have experience with and that I really wanted to shake out before we wrap. And that is sort of the cultural change elements of becoming a really data fluent organization altogether. Clearly, it's a journey. Clearly, it's not a week. It's not a year to get us there.

0:21:37

But you've articulated the general platform, this idea of a data mesh that we've covered in some past episodes. I know you have some experience with. What are the things over the course of the next five years that you think will be the biggest help to move the needle on enterprise culture? What are some of the things leaders or companies will do kind of internally, or maybe forces in the market that will help bigger organizations move in that direction?

0:22:00

Because like you said, the math doesn't work out. They're investing their money in this stuff, but they know that they're not really a data fluent organization. How does that change?

0:22:07
Tanvi Singh

I think the decentralization of power, the centralized teams do not work. They just do not work because they kind of become a bottleneck. There's always a war for talent. And if every part of the organization is trying to prioritize what the central data teams are going to be answering for them, it becomes a constant set of not deriving value at the time that it's needed. I feel a big part of what data can give to you is at what point in time you're trying to answer the questions.

0:22:36

And we do have the technology. We do have the data. We just have to figure out how quickly can we mesh up our way of organizing the teams, structuring the organization to be able to best address those questions. And what we also get to see is increasingly the problems that we are trying to solve, the questions that we are asking is becoming more and more complex when it comes to data. And it's just humanly impossible for a central team to be able to address the set of complexity, timeliness of delivering those insights.

0:23:06

So, very federated set of teams and organizing the culture of the organization to be data-centered. It's very important that we empower the people within the domains who are responsible for not just the AI or building up of very exciting models and really the decorative part of the system, but also the operational part of the system, also the engineering part of the system, also making sure that we have the right set of data quality within the systems.

0:23:32

How are we making the mix and match of the different data sets together to derive value is going to be only attained once we have the data product ownership within the domains and not with the central tubes. I think that's also a huge principle of what data mesh says. And we've all tried to experiment with this new paradigm. So I wouldn't say that I know of an organization that has done it superbly well and we can all copy paste that set of models.

0:23:57

Yes, of course not. It is going to be iterative. It's going to be iterative. But I recommend everybody who feels that they are part of the organization that's big enough, has put in enough sort of investments, but just doesn't feel that you're able to leverage that sort of platform, that investment in a way to make the organization data centric. To read the book from Samark on data mesh, really. It's a really heavy book in terms of technology, the way we think about data, the way we think about technology, but it's a really good one to get a lot of these concepts, a lot of questions that we ask ourselves to be put in a very, very structured form.

0:24:35

Very quickly on what's going to happen in the next three to five years around the data topic is a lot of the data landscape or the technology landscape, we see that a lot of different vendors is going to simplify. And we would be able to see a lot of very sort of standard sort of products that one can work with without having to look for a small set of value chain with different innovative products coming into our way.

0:24:59

So please make sure that it's going to be important that any architecture that one would define for your organizations, it's important it's been source enough for you to be able to integrate some of the newer technologies, some of the newer innovations to make its way to your organization.

0:25:14
Daniel Faggella

So some strong closing notes for those of you tuned in and a book recommendation, which some of you who are listeners, hopefully you are also readers. Tanvi clearly got a lot out of Data Mesh. Maybe you would as well. I know that's all we had for time, Tanvi. It's been a real pleasure being able to chat with you. Thank you so much for being able to join us.

0:25:31
Tanvi Singh

Thank you, Don. It was a pleasure.

0:25:37
Daniel Faggella

So that's all for this episode of the AI and Financial Services Podcast. A big thank you to you, our listener, for tuning in all the way through to the end of this episode. And a thank you to Tanvi for being able to join us. If you're interested in learning more about AI and financial services use cases, make sure to download our AI and financial services executive brief. You can go to emerj.com slash fin1. That's F-I -N like finance.

0:26:01

And then the number one, emerj.com slash F-I -N-1. And download our AI and financial services executive brief, you'll get not only a quick rundown of some of those popular AI use cases in financial services, but also a quick glossary of terms that are important to understand in this space as well. So again, that's emerj.com slash fin1. Thanks so much for tuning in with us. I look forward to catching you next month here on the AI and Financial Services Podcast.