Artificial Intelligence and Business Strategy
In collaboration withBCG
Over the course of her nine-year tenure at LinkedIn, Ya Xu has held technology roles with increasing responsibility. Today, she heads the data function for the online professional networking platform.
Ya joins hosts Sam Ransbotham and Shervin Khodabandeh in this episode of the Me, Myself, and AI podcast, where she discusses AI’s essential role in helping LinkedIn create the best “matches” — content creators with content consumers, job seekers with employers, and buyers with sellers — within its three key marketplaces. Ya also describes how the company has fostered a data-first culture from the top down, and how its vast amount of economic activity data is helping governments and policy makers worldwide.
Read more about our show and follow along with the series at https://sloanreview.mit.edu/aipodcast.
Give your feedback in this two-question survey.
Sam Ransbotham: Most of us have used LinkedIn to search for a job or to make new professional connections, but how can AI help facilitate all the many ways users engage with LinkedIn? Find out today, when we talk with Ya Xu, LinkedIn’s head of data.
Welcome to Me, Myself, and AI, a podcast on artificial intelligence in business. Each episode, we introduce you to someone innovating with AI. I’m Sam Ransbotham, professor of information systems at Boston College. I’m also the guest editor for the artificial intelligence and business strategy Big Idea program at MIT Sloan Management Review.
Shervin Khodabandeh: And I’m Shervin Khodabandeh, senior partner with BCG, and I also colead BCG’s AI practice in North America. Together, MIT SMR and BCG have been researching AI for six years now, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and deploy and scale AI capabilities across the organization and really transform the way organizations operate.
Sam Ransbotham: Shervin and I are excited today to be talking with Ya Xu, head of data at LinkedIn. Ya, thanks for taking the time to join us. Welcome.
Ya Xu: Thank you for having me.
Sam Ransbotham: Let’s start with your current role. What do you do for LinkedIn?
Ya Xu: I am part of the engineering organization. I lead a team that’s called “data.” I know it’s very confusing to people outside LinkedIn to realize what “data” really means, but, essentially, if you think about all the data science, AI, and privacy engineering, it’s all happening in my organization. So we [work] all the way from research to production, so it’s a pretty big organization that is really helping the company realize our ambition in data and AI.
Shervin Khodabandeh: And what is that ambition?
Ya Xu: It is really using AI, using data, to help create economic opportunity for the global workforce.
Shervin Khodabandeh: Comment a little bit more about that. I like, by the way, how you said, “Look, it’s all about creating opportunity and economic value,” and you didn’t say, “It’s about bringing the most advanced technologies and building sophisticated algorithms,” which I’m sure is all part and parcel of what you do, but you started with a very high meta framing and positioning. It’s really encouraging to hear that. But, just generally comment on what that is for LinkedIn and what role AI plays in that.
Ya Xu: First of all, I mean, I think that creating advanced technology — to me, that is “how.” The end goal is not just creating technology, but really, like I mentioned, creating opportunities. Maybe I would break it down a little bit. I don’t know how well the audience understands LinkedIn in general. We like to think of ourselves as having three key marketplaces.
We have our knowledge marketplace where … people [are] creating content on LinkedIn and people [are] consuming content and getting informed so that they can advance in their career and make the right connections to the right people. That’s what we call the knowledge marketplace. We also have the talent marketplace. This is when job seekers come to LinkedIn — and also employers and companies post jobs and recruiters find the right talent for companies.
So that’s our talent marketplace, which also includes our learning — thinking about individuals or professionals who are reskilling themselves so that they can continue to advance in their career or find new opportunities.
The third marketplace is really the product and services marketplace. This is where marketers come to the LinkedIn platform and then really try to identify the buyers for their products and services — sales individuals come to LinkedIn and really try to identify the buyers as well for them.
So if you’re thinking about the role that AI plays, hopefully with that context, it’s very clear that obviously in the knowledge marketplace, we are really trying to match the content creators with the right content consumers. In the talent marketplace, we really just try to match the job seekers with the right companies and with the right opportunities. And then, for the product and services marketplace, we’re just trying to match buyers and sellers, etc. When you put it in that context, I hope it is very straightforward to see that obviously data and AI play such an essential role, because how do you match them? It’s really through advanced technologies that we have in data and AI.
Shervin Khodabandeh: So “How do you match them?” then takes us to a variety of use cases and experiences, and I’m just curious, how do you come up with those? What is the process where you’ve got the business side and you’ve got the technology and engineering and data side working together?
Sam Ransbotham: Take us through the recipe.
Shervin Khodabandeh: Yeah.
Ya Xu: [Laughs.] I would actually maybe start getting to the tactical aspects of things. I think LinkedIn is a very unique place, because the way that I describe LinkedIn as three different marketplaces and the way that data and AI play a role, that’s how our CEO describes LinkedIn.
What I’ve described is not so much like, “Oh, this is our data or AI view of LinkedIn”; this is LinkedIn as a company’s overarching view of LinkedIn. So hopefully that sets the stage for how integrated data and AI are for the mission and the vision of the company, for how the company operates, for the collaboration, which we can talk more about, for how different product development processes work and collaboration happens across different team boundaries. That’s where I wanted to start, because it’s just one word: It’s very integrated.
Shervin Khodabandeh: Yeah.
Ya Xu: Tactically … maybe [I’ll] just talk a little bit about how we are organized. LinkedIn is very functionally organized. So if you’re looking at who sits at our CEO’s table, you have, essentially, all the functional heads: head of engineering, head of product, head of sales, head of marketing, head of legal, you know, HR, head of, obviously, finance. We have a very strong muscle because of that to work cross-functionally.
And so even though all the data science and AI functions are in my team, we have the muscle and the structure to enable really strong collaboration between all the other functions that we work with, in order to bring the AI solutions to production, to life, and to deliver that value to our members and customers.
So concretely, my team — I have leaders who are focused on different areas. And as an example, I have a leader who is focusing on our consumer experience in particular, so the way that he would operate is that he would work very closely, in a very embedded fashion, with other cross-functional leaders who focus on consumer experience. So, what does that mean? … They have many, many touch points, all the way from quarterly planning — “How do you set OKRs? How do you do reviews?” When it comes to a particular initiative, then you’ve got functional heads all coming together and then strategizing what needs to happen, who [should] work on them — planning our road maps. All [of this is] happening very seamlessly as a cross-functional team. And, as I said, because we as a company have always, ever since the existence of LinkedIn, operated in this way; everyone has that muscle.
Shervin Khodabandeh: It’s brilliant, because Sam and I have done a lot of research here and talked with companies across sectors, and it’s really interesting: You said the three key [things that] have really been challenges for most who aren’t getting any value from AI. You said “strategy and mission” — that’s where you started. You said “integrated cross-functional teams,” and then you said “collaboration.” And in fact, when you look at the data of 90% of companies [that] aren’t really getting as much value for their investments in AI [compared with] those 10% who are getting it, it comes down to these very three points. And it’s interesting that LinkedIn, since its inception, has been that kind of a place, where it’s been “data first” in an integrated way.
[This] completely resonates, and [it is] also very refreshing that in your role as the head of engineering and data, that’s also where you first go, before you go and you talk about everything else, which is all important, but it’s the “how”; it’s not the “what.”
Ya Xu: Yeah, absolutely, and I always look to the leaders on my team as well. They are not just like, “Hey, I’m just a leader of AI; I’m just a leader of data.” They really need to understand where the end goal is, right? The end goal is never just creating maybe the largest model or the best state of the art. It’s really about delivering the value.
When you understand that, and then when you’ve got the cross-functional team all having the same shared goal and purpose, then that also brings that collaboration to life really easily.
Shervin Khodabandeh: How hard was that to put that in place and align the incentive and the organization and keep — you said “different muscles”? How hard was that to get to that place? Was it always there? Were there some very hard decisions that had to be made?
Ya Xu: I have been at LinkedIn for close to nine years now. I want to say it’s pretty much always there, because this is how we’ve always been organized. The culture is so ingrained that I think the new employees will just very quickly assimilate into “This is how we operate.”
Sam Ransbotham: You mentioned getting better in collaboration, and I’m kind of thinking back on what you were saying about matching. You used the word matching in all three of your scenarios. What risk is there that you get down this path of a bunch of engineers ever-improving the existing matching algorithms and perhaps missing out on a fourth area that you need to be focusing on? Is there a tension that you’re feeling between ever-improving that matching process through better and better algorithms and data, and figuring out where to apply it in some of your newer products like your … I don’t know … newsletter, LinkedIn Lives, or some of those kinds of new things? Where’s that tension?
Ya Xu: There are nonstop new areas that pop up regarding how we can innovate and should be innovating, and even just starting with matching, it’s very simplistic to say, “Hey, we just can try to do better matching.” But what does “better matching” even mean, right? Let’s take our data marketplace as an example. Obviously, we want to match the best qualified candidate with the best company. But how do you even define that?
Sam Ransbotham: What is “best”?
Ya Xu: What is “best,” not only for the job seeker, but also, what’s the best for the other side of the marketplace, which is the companies? If I were trying to hire somebody, my ultimate ideal state is, I really only talk to one candidate, and that is the dream candidate I wanted to hire.
And then, same thing, let’s say, in the knowledge marketplace, as we try to connect to the content creators, to people who are interested in reading their articles: What does that even mean by “the right matching” as well? Is it that we want to maximize the engagement that people have on a particular post? How do we think about distribution of those engagements across … ? How do we think about … maybe I’m a new creator, first time writing a post on LinkedIn, and I didn’t get any response. I will be so discouraged and never post again. How do we think about that shorter-term trade-off and the long-term trade-off? There’s so much more, even just in this very simplistic framework of “We’re just doing matching.”
So now, coming out of the matching aspect of it, there’s again way more, like thinking about, how do we help people do content discovery? And … when advertisers come to us, how do we actually help them pace their budget? How do we help them utilize their budget better?
Shervin Khodabandeh: Given the progressive culture and the highly collaborative, integrated functional culture of LinkedIn, what makes a good candidate for your team? What are you looking for, in addition to the hard skills of technology and data and AI and data science? What do you think is the secret sauce there?
Ya Xu: I would say, number one, someone who — I think Satya [Nadella] was the one who said that quote, “Would you want to hire a learn-it-all or a know-it-all?” I am a strong believer in a learn-it-all, and what made a candidate successful in the past doesn’t necessarily mean that they will be successful in the future. But that attitude of “I’m going to learn; I’m going to adapt” — I think that’s so important.
So I would say learning — somebody who is really a big learner. And the second one is someone who is just curious. Because when you are curious, you have that drive, you have that “I’m going to get to the bottom of it.” And so much of the amazing progress we’ve made is because someone is like, “You know what? I’m not here for all the fluffy things. I’m just going to be here to really focus on this problem that I saw that I’m just trying to figure out how to solve.”
Shervin Khodabandeh: You know, it reminds me — I had a mentor who used to tell me, “There are no boring projects, only boring people.” And so every time I was like, “Adam, I’m not sure I’m crazy about this project,” he said, “You could make it interesting. You have the ability to learn; you have the ability to complain constructively.” So now you’re saying that; it brought back to my memory, there’s only boring people, no boring projects.
Ya Xu: Absolutely. I love that. I might start quoting that as well.
Sam Ransbotham: What’s interesting, and I think that ties into … you’re a bit of a hero in the academic community. I think maybe you’re attracted to those people — academics complain a lot; maybe you’re attracted to those academics that complain a lot. But I wanted to tie that to another way that you’re creating economic value through LinkedIn.
And I think one thing that’s really fascinating here is that you’ve got a platform that has insight into what’s going on into the economics of the invisible hand, making the invisible hand visible. You’ve got unparalleled insight. I was hoping you could talk a little bit about the things that you — I mean, I’m aware of some of the things you’ve done with code, and with your LinkedIn graphs projects with the academic world. What are you doing in that aspect to get insights into those things that we’ve just never had insight into before?
Ya Xu: Really, really good point. Because of the volume that we have on the platform and how much economic opportunity activity that happens on the platform, we particularly have that insight into [the] future of work, what skills are in demand, how different companies are hiring, which industry is hiring more and hiring less, and even just thinking about the equity aspect of it as well: Do women [or] men have a different rate of job changing, etc., [or] advancements in their careers? All that. So tons of insights that we have on our platform.
So, what we have done, and, Sam, you kind of alluded to [it], is we — by the way, we call all these activities, this vibrant activity on our graph, what we call economic graph — and we have started a particular effort on economic graph, probably seven … at least six years ago, where we essentially stand up a team that includes a bunch of folks on my team with our policy teams, with our comms teams, with our editorial teams, that really try to share and bring some of the data and insights to the external communities.
And we have been very successful, as a matter of fact. For example, last year we sent a report to most of the [U.S.] congressmen/women on what the labor marketplace looks like for their region. We have a collaboration partnership, for example, with Singapore’s education department to help them figure out what skills [are] in demand and [are] lacking so that they can change their educational curriculum to help.
Sam Ransbotham: That’s just huge.
Ya Xu: And then we have worked with obviously a lot of other institutions, either much more directly with a particular local government or with some like the World Economic Forum, the G-20; we share a lot of our reports with them to help influence some of the policies they have.
Another simple example is, we are really helping the broader community know what the green skills [are] — people who are either hiring for green skills, people who are like, “Where is that talent going?” so that as we invest more in green energy, both the governments and those industries can be more guided in that from a talent perspective.
Sam Ransbotham: That’s huge.
Shervin Khodabandeh: You’ve been featured in Fortune’s 40 Under 40, you’ve written a book, you’ve given numerous speeches, you’re a very successful practitioner in a very successful company. I’m curious, what would your advice be to your peers in other organizations that are leveraging AI as the how to achieve whatever the what of their company is? What would be the two things that you think might not be obvious to others?
Ya Xu: The first thing that came to my mind is maybe not so provocative. It’s really just, have the best talent. It’s so important. And I am such a strong believer that when you bring the best people, all you need to do is to get out of the way, and then help them to be successful, and then wonder just happens. Many of the folks on my team can do their job way better than I can do their job, especially in a field that is constantly innovating.
I always joke about how the pace that this field, this domain, is changing is like 300, 400, 500 miles per hour; it’s crazy. I mean, I got my Ph.D. in this domain, and I was in this workshop that was talking about graph neural networks and just graph learning in general, and what the practitioners are doing today versus when I was doing my thesis 10 years ago — it’s entirely different. And what was state of the art 10 years ago is nowhere to be found in today’s practice. So I think that’s, again, just really emphasizing how important it is to bring the best people and the talent, especially in a very innovative space.
And the second thing I want to say is — maybe this is a little less obvious — is to make data and AI work in our company, the way to do it is not to build a wall between folks who know data and AI and people who don’t know data and AI.
And, by the way, this is a general pitfall, either in the mindset of people or how companies are organized. … Let’s say you’ve got an expert team that is world-class in data and AI, and then you just expect, “Hey, you know what? The rest of the company knows nothing about data and AI.”
Shervin Khodabandeh: I think what you said about talent is probably not obvious to many, and I also think it really corroborates your earlier point about curiosity and learning. I mean, if those are the ingredients, then of course talent matters a lot. So thank you for that.
Sam Ransbotham: We’re just seeing that over and over again, or maybe it’s just the kinds of people that we’re, perhaps, attracted to on the show. But it does seem to be showing up a lot.
Ya Xu: Absolutely.
Sam Ransbotham: OK. So …
Shervin Khodabandeh: Is this time for the five questions? Should we do that, or —
Sam Ransbotham: Sure. Yeah. Do that.
Shervin Khodabandeh: Do you know about this?
Ya Xu: I do not know about this, but fire away.
Shervin Khodabandeh: Oh, we didn’t tell you? All right.
Sam Ransbotham: Surprise!
Ya Xu: I like surprises.
Shervin Khodabandeh: We have this thing where we have five questions. You could just riff, give an answer. So, what’s your proudest AI moment?
Ya Xu: I would probably go back all the way to when I was in grad school, and I was taking this class, and it was probably the very first time I really saw AI in application in the way that you can feel and you can touch. I was taking this class with Andrew Ng where we were supposedly building an algorithm that was able to, given a stream of video, identify objects in the video. I worked really hard with a classmate of mine and a close friend of mine.
At the end of the day, they had this competition of the accuracy and precision recall, etc. In a pretty large class, we won second place — so not the first place, so still room for improvement — but [I’m] very proud of that, especially [since it was the] second year in my Ph.D., and before that, a lot of my experience was a little more contrived examples.
Shervin Khodabandeh: Very good. What worries you about AI?
Ya Xu: On one hand, you know, obviously I’m super excited for the potential, but what worries me about AI would be in the responsible AI space in particular. Obviously, I’m really glad the attention that responsible AI is able to get in the public and in the research community, and in industry as well. But at the same time, it’s not just a buzzword. We’ve got to really put it into practice and make sure that we are [continuing to] research how we are able to identify biases that AI systems can bring, and it’s a super challenging space. I’ve been working in this space for, I want to say, like, you know, extensively, a couple years now, and just know how challenging this is, so my call to action to your audience is to definitely lean into the space and research. Continue to push the boundaries on what’s possible.
Shervin Khodabandeh: Very well said. What is your favorite activity that does not require technology?
Ya Xu: That’s easy. It’s skiing.
Shervin Khodabandeh: That requires technology.
Sam Ransbotham: [Laughs.]
Ya Xu: It does not require technology!
Shervin Khodabandeh: The ski.
Sam Ransbotham: Oh, oh, come on.
Ya Xu: By that argument…
Sam Ransbotham: You can expand that to anything; come on!
Ya Xu: About four years ago, I really got into skiing. And the reason I love skiing is it’s a forced meditation. It’s like, you go down the hill, and I’m not nearly a good skier, but all I think about is, how can I get down safely?
Shervin Khodabandeh: Yes.
Ya Xu: Nothing else comes to mind. So that —
Sam Ransbotham: I was sure you were going to say “gradient descent.”
Ya Xu: [Laughs.] That’s funny. But I also like that every time when you go out, it’s different, because the snow conditions, the weather, the slope all add variability to how you actually do when you’re skiing that day, and I love that, just because, you know, it’s never status quo.
Shervin Khodabandeh: What was the first career you wanted, like in your childhood? What did you want to be when you grew up?
Ya Xu: Well, if you say way back, I joke about it still today. Like I told my mom, I wanted to be president then.
Sam Ransbotham: Oh, I will vote for you.
Shervin Khodabandeh: What’s your greatest wish for AI in the future?
Ya Xu: It’s just really help. AI should serve the people and help everything that we are doing to be more efficient and better. And I would say that in general about technology. So many things that we were not able to do, now we can do because of AI and technology, so I continue to be very bullish about that, and I’m very excited that I can be part of it.
Shervin Khodabandeh: Thank you.
Sam Ransbotham: Ya, it is absolutely wonderful talking to you. You know, I was thinking back to when you were talking about muscles, and when you first said “muscles,” I have to say, I was thinking biceps. You know, I’m thinking about big muscles. But as we’ve talked, I’m now thinking more like eye muscles — like focus, like getting the granularity — because what you’re talking about is the data that’s going to let us see things that are happening in the world that we just have not been able to see before, and it’s fascinating, and we’ve really enjoyed talking to you today. Thank you so much.
Ya Xu: Thank you. Thank you for having me.
Shervin Khodabandeh: It’s been truly, truly insightful. And thank you for making time.
Ya Xu: Of course.
Sam Ransbotham: Thanks for joining us today. On our next episode, Shervin and I talk with Nitzan Mekel-Bobrov, eBay’s chief AI officer. Hope you can join us then.
Allison Ryder: Thanks for listening to Me, Myself, and AI. We believe, like you, that the conversation about AI implementation doesn’t start and stop with this podcast. That’s why we’ve created a group on LinkedIn, specifically for leaders like you. It’s called AI for Leaders, and if you join us, you can chat with show creators and hosts, ask your own questions, share insights, and gain access to valuable resources about AI implementation from MIT SMR and BCG. You can access it by visiting mitsmr.com/AIforLeaders. We’ll put that link in the show notes, and we hope to see you there.