Ep #1 Data PM Interview Series — Madhumita Mantri, StarTree AI, Paypal, Linkedin
What is Data Product Managment?
Following is my interview with Madhumita Mantri, Product Lead at Star Tree AI, Formerly a Sr. Product Manager of Enterprise Data Platform and Data Products at Paypal, and a Staff Technical PM of Data Platform, Data Products, and AI at LinkedIn. In addition to this she’s, also an avid writer and thought leader in the data product management space, and the Founder of AI for Everyone.
In this interview we talk about the evolving field of Data Product Management, what it is, where it’s going, and what the biggest challenges are in the field.
Intro
Brett: [00:00:00]
I’m really excited today to be talking to Madhumita Mantri. Madhumita is the Product Lead at StarTree AI, and formerly worked as a Data and AI Product Manager at Paypal and Linkedin. She’s, also an avid writer and thought leader in the data product management space.
In this interview, I’m hoping that we can learn more about data product management as a profession. It’s certainly a newer field, and I’m curious what your experience has been like so far. I’m also hoping that we can talk a little bit more generally about the challenges that companies are facing today that make data, product management, particularly relevant and exciting.
So with that being said, Madhumita, would you mind just giving a quick intro to yourself and what you do?
Madhumita: [00:00:56]
Thank you for having me. And I was looking forward to connecting with more people who are passionate about Data Product Management. It’s a new area and speeding up. So great to meet with you and have this forum to chat with you.
A little bit about me- I am a product leader at Star Tree AI, an upcoming startup in the Bay Area that is primarily focused on real time analytics and anomaly detection. The real time analytics platform was something that was incubated at LinkedIn. I was luckily part of that incubation journey and have worked on features that allowed users to do things like see who viewed your profile in near-real time on LinkedIn.
This platform was powered by the same team who built Apache Pinot, a recently open sourced real-time analytics platform. At StarTree, we’re taking that and giving it an improved form, and contributing to the open source community along the way. So I’m fortunate to be part of this team and continue my work on real-time anomaly detection.
My true north, the mission and vision I have, is to fast-track problem solving by detecting anomalies in real-time and generating meaningful insights out of those anomalous events.
What is Data Product Management?
Brett: [00:02:48]
Wow, that sounds really interesting. I could see how that would be used at LinkedIn and other places. You know, we’ve thrown this term Data Product Management around. It sounds like in your current role, you’re working on a data product that is the core product offering for that company. But in other roles, you’ve been working on products and features that were more internally facing. So how do you define Data Product Management?
Madhumita: [00:03:19]
Yeah, that’s an interesting question. In fact, it has interesting journey for me as well. Having worked at LinkedIn, Intuit and PayPal, where the focus was the internal data platform that we were building, which was primarily geared towards serving internal users such as Data Scientists or Data Analysts or operations teams, etc — there was wide spectrum of data producers, consumers and decision makers.
Obviously at LinkedIn, even though it wasn’t a core product that we were selling, it was indirectly used and sold to our customers globally. For example, there’s a product called LinkedIn Talent Insights that was powered by the real-time anomaly detection platform. It was an enterprise product sold to some of our biggest customers. So as a Data Product Manager at LinkedIn, there were some data products that were externally facing.
But here at StarTree, the core platform is a data product, and what we do is focused on getting this data product in the hands of millions of users. I think the impact is really big, and that 0 to 1 journey is what is promising about working for startup. However, it feels like these kind of companies are turning out to be fewer and fewer these days, because analytics platforms and solutions are becoming more centralized, and more cloud focused. And so I think that’s another evolution that is on the horizon.
Madhumita: [00:05:08]
If you look at like large spectrum of industries, they are not into data, their core business is not selling data products. However, they do need data, and they’re racing to get into data-driven business.
And that’s why I feel like data product management is becoming very important.
What is data product management like? I think in simple terms - you’re analyzing data, building metrics and KPIs to inform product feature decisions or business growth decisions. You’re advising the business on how to collect data and store it in the right format.
So I think end of the day, what data is leading to is better decision making. It’s also opening doors, I think, to build and sell data products like Star Tree AI. Which consist of end-to-end data lifecycle management with some enrichment on top. I call it as data-ops or AI-ops, where you’re able to build a pipeline ICD code, ensure sales are met, quality is met and the complete lifecycle management. So I guess data, product management is quite a vast area as you can see. While for companies like Star Tree, selling data products is their bread and butter, in my opinion it exists in every company in some shape or form.
Internal vs External Data Products
Brett: [00:06:53]
Yeah, it’s funny to hear that you’ve kind of been on both sides where like data was a product or data solutions were the product versus kind of being more an internal, you know, having internal stakeholders. That’s kind of why I asked the question. I find that data, product management is such a broad and fuzzy term, and that’s why I wanted to have this conversation. I thought I knew kind of what data product management was. So I was previously working in product management for a data as a service company where data was the product. So I’m like, okay, I got this. And since I’ve shifted to urgently to more of like a, I think maybe a traditional data product management role where data isn’t the core product, but it’s really more of like an internal capability that we’re trying to establish within the organization. It’s just kind of made me laugh about, you know? Maybe how little I knew about what the space is and really how broad it is. So I think that’s an important question.
Madhumita: [00:07:51]
Yeah, I think some Data Product Manage roles are more vertically oriented, where you are serving a customer a product from end-to-end, and others are more broad, where you may be serving a variety of internal customers and use cases across the business.
Brett: [00:07:58]
Exactly.
Madhumita: [00:07:59]
And then it feels like you are in a part of an ocean and there’s so much to do. Obviously, the only difference I see is in an internal product, probably your number of users are in terms of thousands where you’re taking something external, then you’re putting in the hands of millions and millions of users. At least you’re taking the product out to the door. Now people are coming in, adopting it. How much is the second question? But the market opportunity is so large spot to share because the data is in everything and it’s a basic need and there are a lot of companies that don’t even know what I mean. They’re still in the learning journey that data literacy is a big gap right now, the right skill set like I think. I mean, I’ve also co-founded a nonprofit startup called AI for everyone dot org and the mission of the company is primarily close the skillset gap and trying to help people like underprivileged people who are not able to give either time or money to learn AI or data in everything. So how we can help them to be more skilled at this and then landing job because there’s so many opportunities people are like crowded in other areas with so fewer people and that is also causing a lot of roadblocks in terms of how fast we can grow and catch up with the trends that we are seeing.
Bringing Data Skills to Underserved Communities
Brett: [00:09:28]
Well, that sounds like an incredible effort. Would you mind sharing a little bit about the challenges that you face so far with bringing those skills to underserved communities and where you’ve found success as well?
Madhumita: [00:09:42]
Yeah, that’s a great question. We are still, I think, iterating on that. We haven’t I don’t think that we have hit that sweet, sweet spot yet, although I do see it as a huge need because our first trial was to help non-profits who are not able to use data to make informed decisions and drive their business growth. Then we had two beautiful success stories. They were nonprofits, not startups, but they were quite decent size in India and one was to in the education field, the other one was of like more in the education space but more in the career management side of the world where they were trying to help underprivileged students who can’t afford money and then give them advanced education, like maybe some form of MBA like courses. And so the challenge was there’s so much data and the one is they had limited resources which nonprofits always deal with and how they can get collect all the data and even the simple thing to deliver a dashboard and view the dashboard. They were not able to do that at scale. That was their fundamental problem. And we were able to help with a group of volunteers, data scientists and data engineers. We delivered one first dashboard for them and from there they were able to pick it up. I think the first, like the roadblock of understanding how data product is being built was the journey we took them through.
Madhumita: [00:11:20]
And similarly, the other company was also very same, same bucket. But this is our side gig and we didn’t have enough resources and we’re not yet sure we can turn it to 100% time. So we decided to pivot and our goal was to close the skillset gap, primarily because this was the learning which kind of led us to think that, okay, we have limited resources, we had only three volunteers. So then our journey was to open up how we can help people to learn things quicker. And we did a couple of iterations there and a couple of other interviews. The problem we identified is people are somehow intimidated with the word data and AI, and if it’s need, it needs a lot of skill set and to get into the space. So I think people are just hesitant. So yeah, we were again not very successful in that. Now we are pivoting and trying to help people who are not necessarily technical background. You can be non tech background and how we can remove the initial friction to really get into data II space. And we are planning to give that level of coaching and expertise given we all had that expertise. Yeah, that’s the recent experimentation that’s being launched. So let’s hope for the best and see what happens next.
What challenges are companies facing today that make data product management so important?
Brett: [00:12:49]
Well, best of luck with that. You know, it sounds like this is something, you know, data product management, the data space in general is something you’re really passionate about, even where you have this side gig. What do you think it is about either the challenges that companies are facing today or just kind of the state of the world today? That makes this a really exciting and special place to be working in.
Madhumita: [00:13:13]
Yeah, I can answer into folds. One is the challenges. Obviously, those turns out to be opportunities and also the upcoming trends that talks about the future. And definitely future looks also very promising and combining the two for sure data, product management is a very promising area and I guess an interesting area that everyone should start considering in my opinion. And obviously there’s no easy path to this. You have to build certain or close certain skills, skillset, gap, but it is not hard if you are passionate about data and there’s so much things to do, like, I mean, so many things to do in terms of not being technical. Also, you can bring in so much impact. So I can quickly highlight the data challenges that I have seen based on my experience working at these companies and doing my own startup. First of all, like if you talk about bigger companies versus smaller companies, I will segment it into two buckets. The smaller companies, of course, they don’t have huge needs of doing data or running the data at scale, so the scale problem goes away. So that’s the only difference I found between the bigger company and smaller companies, the scale of problem we are dealing with. But most of these problems are common between the two except the scale. But now at a bigger company, another thing I’ve seen is the data silos. So there’s so many data team from getting formed and this data silos and over the course people are getting introduced to this new data world in a slower pace. So what has happened organically? These bigger companies have grown and they have several data teams and they talk different language, they follow different architecture principles and they had pinned up their own storage and pipelines and processing.
Madhumita: [00:15:12]
So bringing it all together and yeah, so that had contributed a lot of challenges and which we are using like a typical raw data mess you might have heard a lot about, it’s been talked a lot into it and I think LinkedIn. So as a PayPal, PayPal, the bigger challenges were the so many teams grew data teams that they had a big challenge of like so many data hops, they had about 800 petabytes of data. It’s not that big of a scale, but they had so many data hops. And because of which there’s a lot of challenges around one, not only data management becomes huge problem, but besides that cost of operations, even business cannot innovate and iterate quickly. So that’s the bottleneck that I have seen people who have done that in a proper way. So agility, I think, is one biggest thing. Another thing I’ve seen people who are working in the data space, as I earlier said, because of limited knowledge and learning or skill set. So there’s only handful of resources who know this and the need is very big, right, as I’m talking about and so, so many things to do. So like so, so little resource so and your existing team cannot scale and serve everyone. So I think productivity was one big team at LinkedIn that productivity and education, two big thing that was led for like couple of years how everyone can be productive in self-serve. So, so was that thing that got introduced at LinkedIn and obviously now it’s a very common thing.
Madhumita: [00:17:03]
But when I was at LinkedIn, that’s the journey I’ve seen. And besides productivity and these things at PayPal, my experience was more around they want to catch up because LinkedIn, Netflix or other big data companies, they have done a decent job and they caught up with the trends. But PayPal was something being in fintech space. They were a little behind and they want to catch up with the speed and how they can. Then they came up with this hybrid model of like you go with commoditized solution. They were completely against of it like earlier everything build, but they kind of started switching the gear to catch up with the trend and move to cloud. So so not many companies did that in the data space early on and these guys took that leap. So I guess that’s another learning I saw over the period. Like, if you want to be really agile, quick and catch up, then move to the cloud and how to kind of take away the ops part from you so that your team can just focus on the core data solving the core data problems and you can be agile the other. Big theme I have learned is data quality and observability. It was a huge problem because any KPI you use for driving business growth or any key decisions you are making, data quality is always at the bottom. Same thing for AI ML models. Also, we are building models and recommendations for individuals. So that’s another big challenge I saw in data management governance overall serving data at scale.
Madhumita: [00:18:44]
That’s for the bigger company that I mentioned about in terms of trends that I’ve seen. I think based on these challenges, there are a couple of teams I can see three bigger things if I have to summarize. One is the dynamism or diversity, like how you can activate both and build adaptive AI system with this growth of data. I think that’s a strong trend that is really standing out and also coping with this fluctuating market, global market. So it’s a very strong trend I can see and the challenges that are being seen at the company will become even more problematic if teams are not handling in a timely fashion. The second one is augmented intelligence. I can totally see that growing that area becoming stronger and stronger because everybody wants to know where is my data and where can I find my answer? And it’s also not data specific, but people oriented. Also a lot of context driven analytics and real time analytics. That’s another upcoming trend. And then and also like how you can derive value from data at scale. That’s another bigger thing. These are the three bigger buckets. And if I have to kind of take one level down, then the teams could be how we can every company can manage their data mass effectively, data productivity, like how everyone can be more productive in handling data like it could be self-service. The more and more people can do on their own, that will be a lot more easier. Data observability, obviously another thing and then you just facing analytics and AI and productivity.
Your Thoughts on Data Mesh?
Brett: [00:20:38]
So there’s a lot to bite off there. It sounds like you’ve had you’ve seen challenges that different organizations face based on where they’re at in terms of cloud transformation, lots of different team structures, lots of different data silos. So I, I want to get a little bit sidetracked here and talk talk about data mesh, which is something that you brought up and talk about self service as well. I’m curious, what is your opinion on data mesh as it relates to these enterprise wide problems that you’ve seen in the past? You know, this is something that I’ve mostly only read about. It makes a lot of sense in terms of taking a microservice approach towards data infrastructure. It’s something that we’re currently striving towards at my current employer. But based on what you’ve seen, do you think that data mesh is a solution for some of these problems? Is it more aspirational?
Madhumita: [00:21:37]
It’s more aspirational at this moment. I can’t say it’s a solution and it’s more of a concept to me than a product or a solution or a platform to me. It’s just that the world we are in. As an example, what was happening. Like traditionally the data team would develop something bottoms up and most of the time it won’t meet the needs of the individual, like who are not the direct users of your platform but end user of the product. The way I see it, like you are a data producer to begin with, you’re just producing the data, but you do not know how the data is being used. So what happens is the middle person who is analyzing and deriving insights from it, they will derive in certain way like they are again, not the actual data owners or the end users. And what when told to them, they will derive the information that we and this actual end user who has actually the business logic and understanding. So the gap between the three different layers is causing a problem across the board. And the sooner we minimize the gap, the better. Then we can operate efficiently in many different ways and how we can close the gap. One way to solve for that is the solving for data mess and making it more self-serve self-service product and also how data can be more discovered. Data management is done effectively and data observability is effectively implemented. I think these are three different areas. Plus self-serve part would help the end user who actually has the business logic or understanding of the who are the true owner of the data can be more independent and they can define and move quickly with their product experience. So I think these are that’s what in simple terms data mesh is about. And yeah, if, but if you talk about mapping the solutions, there are different things to be solved here. One is the data discovery problem, another is the data quality in observability. And then third is the self-serve aspect.
Brett: [00:23:55]
Thanks for that. It’s helpful to hear, based on your experience, what you think of this. And I think we kind of share a similar perspective that data mesh is kind of like a paradigm, an aspirational paradigm and how it can be implemented as still to be seen, especially by me at the least.
How Do You Measure Success in Data Product Mgmt?
Brett: [00:24:34]
Okay. Great. So another question here is you’ve brought up so many different challenges in the data space. And I think that you have a lot of you have a very strong grasp on what the challenges are like. You brought up self service and data observability, data quality. You know, I don’t think we’ve talked about time to market yet for insights, but we did talk about how do we scale, how do we scale knowledge of our data across the organization. So I think it’s really important to first understand what are these different areas, what are these different challenges as it relates to data strategy and data products? But one thing that’s that’s kind of hard to do once you’ve listed these all like all these components is how do you measure your success? Let’s say, if you could imagine that you’re a chief data officer or you’re somebody who’s starting a data team, you’ve probably been building a data team, you probably invested a lot of the company’s money in order to hire these data scientists in order to set up a data warehouse.
How do you succinctly measure your success?
Madhumita: [00:25:51]
Yeah, that’s a great question and also a difficult question. I think the success will vary from company to company. In my opinion. It all depends on the problem context. You have at hand what problems you are trying to solve, and every company has the team, but they might be solving different things. As an example, if you’re like more into advanced data space, then you’re doing AML and advanced analytics, then your success would look very different. And if your company is focused more towards innovation and quick time to experimentation, then your success would look very different than a Fintech company who is want to be really careful in everything that they are handling with transactions. Anything goes wrong in their transaction data or fraudulent activity. They want to make sure the quality is the big component there.
So based on the context, I think your success could be very different at one company. It could be quicker time to experiment and other company everything like how to do it right and yeah, so not sure I give you a straightforward answer, but but based on I usually use this framework and obviously another big part, it all starts with the problem statement and where you want to go and you define your success criteria based on that. Another big component is also not forgetting the individual and you just centricity in mind who are your users? It comes down to finally that like for example, are you a B2C business versus a B2B business, or B2C in a very high risk, highly regulated environment, etc .
Madhumita: [00:27:42]
The data strategy could be very different for each of these companies, so the success would very look very different. But I think if you talk about the commonality in common challenges and what are the some of the success factors over the period that I’ve worked at different companies… I think the first thing is 1) definitely as a company we need to see if we can prioritize our effort and be focused — that is very, very important. When building a strategy, there should be a focus or theme. And then second is, 2) do you have the right team in place? If not, then what is your plan B? What does your Plan C look like whether you build-vs-buy or outsource something, and mainly the data agility is something everyone should be considering. And the third thing is 3) obviously keeping future in mind, because the in the data space we are super reactive — this is changing a bit, but I think being proactive is key. And the fourth one is 4) don’t forget overall data management, all the governance compliance things that obviously were not there before GDPR. And yeah, that’s an ongoing thing and I’ve seen security being a bottleneck at every company, data security, data management.
What Advice Would you Have for People Looking to Get into Data PM?
Brett: [00:29:30]
Got it. Yeah, that makes sense. So just moving to some concluding questions here, I think on my free zoom, I only have 30 minutes or so or 40 minutes, so I only got 9 minutes left. So. Thinking about data, product management and all the things that you’ve learned. What would you what would you recommend to people who are interested in data, product management? I suspect that there in the future will be a lot more people in this role who are either product managers who are looking to get into getting into the data space, or maybe they’re data practitioners looking to kind of transition into products. Short of having the incredible experience that you’ve had, is there anything that you can recommend?
Madhumita: [00:30:14]
Yeah. I mean, I have written an art study newsletter on this topic and I had done a clubhouse session with Quin Joe, who is the Senior Director of Product at PayPal. And she had explained very nicely and both of our conversation I recorded in there. But to summarize, I think data, product management, is still in its infancy. For example if I go and search on LinkedIn for Data Product Manager, there are barely any results. But you have to realize there are many jobs where you can be a data product manager, even if the title isn’t explicitly there. I’m sure that will change in future. So I think in terms of job search, people need to understand like what that role demands for and that’s not quite yet there. So I guess first thing is needed to understand these opportunities, problems and challenges in the data space. And is that something you are passionate about solving for? My journey started from there when I started my career. My first assignment that I worked on is building a data driven map for IMS Health to help measure supply and demand, and manage their whole segment using data. I really got passionate about it. So I think for me, the way I get attracted to a problem is trying it out and seeing what you learn along the way.
Madhumita: [00:31:46]
Next step would be if that is something really I’m interested about, I’m passionate about, then look at my experience and where I want to land on. Like if I’m a technical person, then maybe a data scientist or data engineer, or if I am non technical or I’m technical, but want to pivot to a non technical role. So then there is different roles just starting from customer success sales, product management, program management, all those areas one can explore. But in everything I think there should be basic literacy needed. So there is some skillset gap there needed to be done. Either you can attend boot camp these days. I’ve seen Carnegie Mellon University. I think they are giving some special courses for first grads to even do data plus product management together. And I’ve seen like two mentees at a people, they both landed in very good jobs, one at Microsoft after the internship and both were from Carnegie Mellon University. They were able to read data and product management together. So yeah, that’s another way. And there are different parts like not one part, straightforward part, but first of all, one needs to identify is that something interesting to them? And that’s the problem they want to go after.
Brett: [00:33:05]
Got it. So if you’re interested in data, product management first try to understand whether you’re interested in this problem space. You mentioned like taking courses, boot camps. I know that there’s a lot of kind of data projects out there in terms of Kaggle and things like that as well. There’s also your newsletter, that’s the Product Management Digest that you publish on Linkedin
Madhumita: [00:33:31]
Yes, that’s correct.
Brett: [00:33:33]
And then the third one that you threw out there is the Carnegie Mellon courses on data management. I’ll have to look that up. Thank you for sharing that. I think just as a last question for people who want to learn more about you and your career and the writing that you do, how can they find out more?
Madhumita: [00:33:53]
Yeah. You can follow me on LinkedIn and I have a medium handle. I’m not writing as much on Medium these days, but I have written one interesting article in Medium, which is once you land on a product management job, what you need to do in 90 days to be successful. And I’ve tried that recipe. It’s a recipe book playbook and I’ve tried it. My previous companies have been successful and I’ve seen a lot of good traction, so you can find me there. I’m also on Instagram. You can find me on Instagram and Twitter. These are my social handles.
Brett: [00:34:27]
Great. Okay, well, I can’t wait to read that. How to be successful in your first 90 days. Maybe there’s still time for me to do that, but thank you so much for the time. It was great chatting with you and I hope to stay connected.
Madhumita: [00:34:45]
Yeah, same here and hopefully build the data beam community and love to stay in touch and be part of this community.