Inside AI Product Management: From Infrastructure to Innovation

Introduction to AI in Product Management

Rina | 00:00
Hi everyone and welcome to Product Side Stories, the podcast where we reveal the very real and raw lessons learned from product leaders and thinkers all over the world. I’m your host, Rina Lexin, CEO of Product Side. And today I’m joined by Digveh J Singh, the Senior Product Manager at YouTube.

Digvijay is working on a team using AI and machine learning models to optimize the ad experience. He also has a data science background and has held product manager roles at Swift Financial, soon later acquired by PayPal, and Google. In this episode, we’ll be unpacking the evolving role of AI in product management and implications on the product manager career. Welcome, Digvijay.

Excited to have you.

Digvijay Singh | 00:54
Thank you, Rena, for the introduction. I’m glad to be here.

Rina | 00:58
Well, I always like to start with your story. I—that’s why we named this series, Product Side Stories. How did you end up in product management?

From Data Science to Product Management

Digvijay Singh | 01:09
Great question. So I started my, I guess my career in engineering undergrad in India. And post that I got into data science and data science consulting initially, where I was basically advising clients in the credit card and insurance industries about machine learning models. Post that I went to a startup called Swift Financial, where I was basically building machine learning models for digital loans for small businesses. And once I was done with that, I was like, OK, what’s next? So I wanted to be able to bring together my data science skill set with some of the business skills that I had learned. Got an MBA at Columbia Business School, then got into product management at Google. But I’ve been primarily working on machine learning and AI products, first in the Google Play Store, more recently in YouTube ads.

First on ads targeting and now much more focused on ad optimization which is optimizing ad load experiences on YouTube as far as ads is concerned and yeah would love to get into more of those pieces if that’s helpful.

Rina | 02:25
So for everybody that’s not a premium user, are you the person trying to help advertisers make the most, like telling them when are the clicks happening, how to convert better, is that what you’re working on?

Digvijay Singh | 02:39
Great question. If I had a penny for everyone who starts with the premium part, I would be definitely in a different income bracket. So the way that sort of, think almost all ad systems work are, of course, they’re sort of what we call demand side signals, right? So that’s like more like clicks, conversions from an advertiser perspective. The part that I work on is also trying to optimize from a user perspective, right? So how many ads should you receive on YouTube? And if you’re a more ad-sensitive user versus a less ad-sensitive user, how should we bring that into our AI models and optimize accordingly? And there’s sort of obviously a lot of nuances that go along the way, but you can think of like demand and supply as the most abstracted version of what we work on on the ads tech stack, and I’m on the supply side.

And that’s sort of what our AI models are used for, is optimizing supply, if that makes sense.

Rina | 03:46
Yeah, so you help people get identified by like how many seconds can you put that skip button down in your screen, right?

Digvijay Singh | 03:57
That’s a very good guess. And you’re absolutely right. We’re trying to figure out whether you should receive a skippable ad or non-skippable ad and so forth.

The Evolution of AI and Its Impact on PM Roles

Rina | 04:08
Fascinating. So, and this is a great start to our conversation because it is true. One of the things that I’ve heard over the past few years is like everybody’s all talking about AI, but the reality is that AI has been in our life for quite a while, right? So many different products have it already in the product itself. And right now, I think obviously the capabilities are changing, the types of AI is changing, but it’s been a reality for us, hasn’t it?

Digvijay Singh | 04:42
For a long time, like I’ve been at Google since 2019. I’ve been working in data science before that. I’ve heard the term neural networks and deep neural networks for a long time. And of course, I think that it’s well documented about how large language models have come into the space and sort of certainly revolutionized big parts of our industry, like across the technology landscape. But I think, you know, the underlying technology or the underlying models beneath sort of those LLMs have been around for a long time. And I think there’s like so much literature that points to that, right? Which is that, you know, everyone at the frontier of AI today has been working on neural nets for the last 15 years. So that kind of gives you a sense that they’ve been around. And of course there have been like step change function, you know, step change, which is that have happened in the last few, few years. But a lot of the underlying technology has been around.

Rina | 05:42
Yeah, I know on our team Dean Peters likes to say he’s been working in AI since the 80s, right? So it’s been around. But maybe you can kind of clue us into what’s different now. What is different about the PM role landscape today? That’s, you as you said, there definitely is a step change right now that we’re going.

Digvijay Singh | 06:05
No, that’s a great question again. think maybe I’d say the way that I perceive PM roles on the AI side or across the AI stack are probably, I’d say, three broad categories. So if we start from how the AI models are used to begin with, obviously they use a lot of compute, they use a lot of the NVIDIA chips that everybody’s talking about. So I would call those sort of more of the infrastructure PMs that are building out a lot of the data centers. I know a few of them, I’m familiar with the work that they do… (continues verbatim from transcript)