Building Products at Uber: How We Bridge the Physical and Digital Worlds

evening, everybody. My name is Greg Mittleider,
and on behalf of cs 50, I want to thank you once again for
joining us for a TED Talk in our series this fall semester. Tonight we have the pleasure of
having our friends from Uber with us. They’re going to talk
about the obstacles they have to overcome in what seems
to be a very simple task by us, the user, which is how the
car gets to you whenever you choose that you need a ride. Tonight we have the pleasure of
having Matthew, Yuki, and Alli. They’re going to be
speaking with you tonight. So without further ado, Uber. YUKI: Thank you so much. Cool. so today, Matt and I are PMs at
Uber, product managers at Uber. And we’re here to talk
about how we build products at Uber, how we bridge the
physical and digital worlds. At the expense of making
us seem really old, we wanted to talk a
little bit about what Harvard was like before the Uber days. Matt was concentrating in
English and busy doing pre-med in the beautiful house with Elliot. And I was concentrating in
computer science and TF and cs50 in superior house that is Mather. MATTHEW: You can do the slides. YUKI: Yeah, I did the slides. And in terms of how we
got around, of course, we used the T to get into the city. I think all of you still do that. But there’s always the inconvenience of
needing to come back before midnight. We took cabs to get to the airport. And of course, we took the Harvard
shuttles to get around campus, the Matter express being my favorite. And Matt and I liked using
this to visit our sad friends in the Quad who didn’t quite make
the cut for our blocking groups. But CS50, in terms of
the innovation that was happening in this
transportation space, CS50 was pretty eager to make
the shuttle experience better. And David built Shuttle
Boy back in 1999. And what we did was also
take this massive PDF that people had to parse to
understand settle schedules, and we have made these
CS 50 shuttle cards, which are these tiny little things
that you could put in the wallet. We printed thousands of them and
distributed it across campus. So that was the kind of innovation
that was happening at the time. MATTHEW: They were viral. YUKI: Yeah, they were kind of viral, but
we clearly weren’t thinking big enough. But it was really funny. I dug up some e-mails with
David back in the day. I was really excited about these things. And so I would ping him late at
night about the Matter schedule or something like that. So something I was
really passionate about, but that was the kind of innovation
that was happening on Harvard. MATTHEW: So that was kind
of what working at Harvard was like back when we were in school. But obviously, things
have changed quite a bit. We were in school six years ago. There was no Uber. Today that’s not quite the case. So if you look now at a map
of Harvard or Cambridge, Harvard Square, you can see
each of these yellow dots represents an Uber trip that’s begun. And this is just data
from the last week. So we’re moving quite a few people
around the city from point A to point B, which is pretty cool. And if you zoom in a bit, and you just
look at trips that begin or end at some of the Harvard houses– this
is also from like a weekend– you see that we’re doing some good work
bridging the Quad, River House divide, which again, as someone who lived
in Elliott, I think is really sweet. Because I honestly couldn’t tell you the
difference between Currier and Cabot. But apparently, we’re making that world
smaller, which is really wonderful. And if you just take a look at
the count of trips by house, it’s also really interesting
to see some of the trends here. So there are a certain set of houses
that people really like to visit, they’re being dropped off at. And there are other houses that
people seem to want to leave, and they’re being picked up at. So this is kind of the
relative ranking in terms of where people are going and coming. YUKI: We’re a little bit unsure
what’s going on in Winthrop. MATTHEW: Yeah, and not
[? happening. ?] And then Uber has another business called
UberEATS, which is food deliveries. You can get food delivered
from a local restaurant. If you look at the number of
deliveries made per house, it’s also kind of interesting to
see which dining halls may not be serving the needs of their
students quite as well at other homes. So [INAUDIBLE] what’s happening
at Mather, but maybe Yuki can– So Uber isn’t popular just at Harvard,
and hasn’t changed transportation just here at Harvard. We are in tons of cities
all over the world. And if you just look at in the US,
these are our five biggest cities as we rolled out five years ago. And you can see there’s lots
of interesting data here. One is that our growth has been almost
exponential, which is pretty cool. And if you look– this is in the order
of which we launched our bay city, so SF first, then New York, then
Seattle, then Chicago, then DC. That growth rate is increasing
as we roll out a new city. So as we launch in a
new city, the adoption is even faster than it was in
previous cities, which is pretty cool. And if you look at our
trend globally, we’re in tons of countries, which
we’ll get to in a minute. This growth is also really rapid. And we’re now basically
everywhere in the world you go, you can pretty much take an Uber. Latin America being our faster
growing market right now. And I’m actually going
to Colombia tonight to go see what’s happening
with our cities there, and see how we can make our product
experience better for our users. YUKI: And for those of
you who are observant, this is in a logarithmic
scale, and so we had to adjust this chart to make it fit. MATTHEW: All right. so Uber today,
2017, we’re at 600 plus cities. And we’re like 660 right now. We’re in 77 countries. And as of June 29, we hit
our five billionth trip, which is pretty amazing. But the cool thing is these
numbers are changing every day and growing everyday, or
expanding to more and more cities. Like just the other week, we
launched a new city in Croatia on some small islands. And these Uber’s were all boats. So they were contacting
us, and being like, can you change the
app to work for boats? And I was like, OK, we
can do that, I guess. So there’s a lot of land
still to be grabbed. And it’s really exciting times for us. YUKI: Cool, so this all
sounds really simple. And we’re doing billions of rides. And it seems really simple to get
someone from point A to point B. But it turns out that
a lot can go wrong. And specifically, the
thing that can go wrong is [? always ?] alluded to
earlier, which is actually getting picked up by your driver. And so if you think about the journey
for a rider, you open the app, you request a car, and
you get into the car. But actually, this particular
part is the most difficult part. And this is where we need to bridge
the physical and digital worlds. So I’m going to first talk a
little bit about this period right here, which is just the period
between the opening up of the app and requesting a car and all the tech
that goes into making this happen. So the first question we ask when you
open up the app is, where is the rider? And this sounds like a really, really
simple question, but figuring it out isn’t always easy. And one of the reasons for this is
because GPS is often inaccurate. And so there are many
reasons that can cause this. This can be because there are
tall buildings surrounding you, and they block the line of
sight with the GPS satellites. There can also be buildings that
cause reflections that misguide us. And therefore, I think all of us have
had this experience where we open up our app– whether it’s the
Uber app, Google Maps– and the blue dot isn’t
really where we’re at. And these are the kinds of
problems that can produce this. We have another kind of problem,
which is that reverse geocoding can be inaccurate. Reverse geocoding is the act of taking
a geocode, which is the lat, lng, these coordinates that represent
where you are, and then turning it into an address. So these are a lat lngs for this space. And when you reverse geocode it,
hopefully what it produces is 67 Mt. Auburn street. But actually, if you look at some of
the examples I’m going to show you, this can also be an inaccurate process. So as an example, I’m not sure if
any of you have been to Tea Luxe in this square. If you have, if you happen to put
your pin right in the middle of it, then it produces the correct
address, which is 0 Brattle street. It turns out Brattle street is
zero index, which is pretty cool. And so that’s correct. But if you move your pin slightly
away from it, then all of a sudden, we think you’re in 1 JFK street. And for whatever reason, it
may be because the outline of the POI, Tea Luxe– the point of interest–
might be slightly inaccurate. It might be because it’s actually
a little bit closer to JFK. And this slight difference
produces a different kind of pickup because the driver is going to
show up on JFK street, which is very different than
showing up on Brattle street, especially if you consider
Tea Luxe, which I think only has an exit on Brattle street. And there are many kinds
of examples like this. So this is the Cambridge Side
Galleria for those of you who’ve been. So if you put a pin on
top of PF Chang’s, which is a Dave and [INAUDIBLE] favorite. I don’t know if it’s changed. Then what happens is we have a
selection of addresses to choose from. It could be PF Chang’s. You’re technically still inside
the Cambridge Side Galleria, so we could choose to show that instead. We could also show the street
address for Cambridge Side Galleria. We could show the street
address for PF Chang’s. And you can imagine if
we pick any one of these, it might produce a
different kind of pickup. And so this is our
fundamental problem, which is that your input, which is
the lt lng, could be inaccurate. And then the process of
translating that into an address could also be inaccurate. So your output is also erroneous. So this is what we’re
dealing with, which is a difficult situation to be in. So what do we do? Some of the ways we try
to solve this is actually trying to predict where you are based
on the information that we have. So if you really think
through the problem, we started with some hypotheses. So for example, let’s
say that you recently got dropped off at Currier House. And even if the GPS is telling
us that you’re in Pfoho, we might actually
believe that you are more likely to be in Currier because
you just got dropped off there. So that could be an
interesting heuristic to start using in terms of round trips. Or another hypothesis, and
this is a simpler example. If you’ve saved a location–
so our app has the feature of basically favoring a location. So Mather House is a saved location
obviously for me, both in the app and in my heart. And even though the GPS might
be telling you we’re not there, it’s very likely because
you saved it that you’re going to be in this location
or you care about it. So that’s a signal that we can use. And when you start thinking
about these, these hypotheses turn into what we call features. Features that inform our
pick up prediction models. So our pickup prediction
model takes a bunch of these different kind of
parameters, if you will, and basically starts to build models on top of it. So other features include trip counts. So, for example, I worked
at the Harvard Crimson. And I spent a lot of time there. So this is the hypothetical me,
because these trips I never took. And I spent very little
time at Adams House because they didn’t let me eat there. I don’t know if that’s still a thing. But in any case, if
the GP tells us that we think you’re somewhere in
between these buildings, we might actually favor the Crimson
because you take a lot of trips there or you take a lot of trips from there. Another feature might be destination. So this is a slightly
contrived example, but you have this area right
near Lamont where there’s the Department of
Comparative Literature, and then there’s the Barker Center. And depending on where you’re
going, if you’re going northbound because these are both one way streets,
we might favor the Barker Center. If you’re going southbound, we
might favor the other place. And another feature that’s really
interesting is time of day and day of week. So in the weekdays, you might
be going to CS50 lectures or whatever that may be. And so if your GPS tells
us that you’re in an MBRD, we might actually intuit that
you’re actually closer to Sanders. But maybe in the evenings, or on
Saturdays or something like that, you’re in Queen’s Head. So all these contribute to this model
of figuring out where exactly you are. And we’re experimenting
with a lot of them all the time to make
it smarter and smarter. MATTHEW: So we talked a
lot about figuring out where the rider is when they open
the app, which is pretty awesome. But in order to make a trip actually
happen, we need to get a car to you. So I work on routing. And one of the big problems
is that where humans are isn’t necessarily a place where
a car can actually reach you. So we have this problem whee
even if we know where you are, that doesn’t mean it’s the best
place to actually get picked up. So we need to translate
the place where you are, which we’ve decided now– where
we’ve helped to intuit– to a place where a car can actually go. So let’s take an example. This is an airport. This is actually the airport in Dallas. And this is an image
taken from two years ago. What this is depicting is a bunch
of trips where every trip is represented by two points and a line. One point is the location where the
rider was requesting to get picked up. So they get off the plane. They’re walking to the food court, and
they’re requesting to get picked up. I wand to get pick up at the airport. The second thought is where
the trip actually began. So what you can see is
that people were asking to get picked up all over the airport– at the food court, where they’re
picking up their luggage. Some people were even requesting
a car from in the plane on the tarmac, which is impossible. Cars can’t go to any of those places. So there is this coordination
that has to happen where the driver has to
call you, and be like, where do you actually want to get picked up? I can’t go to the food court. And the rider has to be like, oh,
I guess somewhere near departures, or maybe I’m not from here,
I don’t know where to go. It’s very messy, right? But you’ll see that most
of the actual pick ups will happen somewhere along this
road, which is where arrivals are, and you can actually get
a car into this space. So what we did is we said,
OK, this is problematic, and there’s a team in Dallas. Every Uber city has a operations team
that works there and they’re local. We go to them and we say, hey,
look, you guys live in Dallas. You know your airport. We don’t know you airport. Can you tell us where at the Dallas
airport people should get picked up? And they’re like, yes, of course we can. We fly here all the time. These are the five spots. And so what we do is we actually mapped
out all the places at the airport that someone can get picked up. And actually we indexed it by
terminal, which is pretty cool. And so we launched this. And what this did was
it cleaned up the space where people were actually requesting. So now when you land at the airport,
we figure out you are at the airport using what you just described. And then we say, OK, you can
only get picked up and request a ride from one of the select places. And that simplifies the
problem for the drivers, so they know exactly
where to go, and also for riders, because they
know exactly where they’re going to end up meeting their driver. So this works for airports. That’s great. But how can we make this model
work outside of airports? There’s lots of places that are
also difficult to get picked up. We wanted to scale this globally. So one idea is we could
just go to the city teams and be like, hey, can you go
to every building in your city, and just tell us exactly where
you should get picked up. And the answer would probably be no,
because that would be a ton of work. And that’s super crazy, and no
one human knows what every pickup location is for every building. YUKI: You could [INAUDIBLE] MATTHEW: We could do that actually. Good idea. Maybe we’ll circle back afterwards. So that’s probably not going to work. So what we also could do is we could
just take the address of a given place, and move the pen to that road. So, for example, Leverett House
is located on 28 DeWolfe street. You could say, OK, for
Leverett House, let’s just find the closest
point on DeWolfe street and just have our pick ups happen there. And for a lot, this actually would
be a pretty decent pick up spot. DeWolfe’s a pretty chill space. There’s a sidewalk. It’s pretty good. But this sometimes breaks. So you look at Grendel’s. Grendel’s is a bar on the square. It’s just down the street. May have been there if you’re over 21. And one of the problems
with Grendel’s is that their address is on 89 Winthrop
street, which is this road right here. But Winthrop street at this
point is not car friendly. It’s actually a walking path. So if we start sending
cars to 89 Winthrop street, we’re going to start
running over people, which is really bad for people and for Uber. And I’d probably get fired. So we needed something else that works. Well, the thing that
we have at our disposal is we have all this GPS data from
all these trips that are happening. So we can look at all
of our trips and say, let’s intuit based on where
people tend to get picked up, what’s a good location? And if you do live at
Leverett– this is Leverett– it actually looks OK. You can see there’s some clusters
here around the side entrance. There’s a cluster here
by the main entrance. It’s decent. You might be able to
intuit something from this. But you go back to
Grendel’s, our problem case, and if you plot the same
data– this is all from week– there’s just so many trips
happening here on Grendel’s. It’s such a popular area. There’s a ton of popular
bars and restaurants there. It’s kind of hard to gather
signal from all of this. So we tried another approach. We look and we filter points only
by trips that we’re requested from Grendel’s. So as before, we were looking at all the
trips that began around Grendel’s, we were only looking at trips of the person
was at Grendel’s when they requested. When you do that, the
data is much cleaner. There’s a much clearer cluster just
in front of the bar, which is probably where you want to get picked up. And we can actually generate a
point for the pick up right here. And so if you use the app today, if
you got out to Grendel’s tonight, have a beer and request an Uber
home, you will see these points. And they’re available for all types
of locations all over the world. We go back to Lev, run the same filter. It also simplifies the data. Generates a much cleaner cluster. And if you see you in [INAUDIBLE] today
or in [INAUDIBLE] when you open up the app, you’ll actually
see we found spots that are in alignment with Leveretts’s doors. So it’s actually pretty cool. We went walking around the other
day because we hadn’t been back since this was launched. And we were like, oh, my gosh, our
spots are at the door at Leverett, which is so cool. We didn’t even tell the system. It just learned that based on
the trips that were naturally occurring there, which is pretty cool. Now this approach is actually really
important for what we call nested POIs. And these are really common,
especially in big cities. What this is, essentially
it’s a building that has multiple places
within that building. So for example, the building that
we’re in has multiple offices. The example here is the Charles Hotel. The Charles Hotel is a very
fancy hotel in Cambridge. Maybe some of you have been there. Your parents have stayed there. It’s a big complex. And within that complex
is the main hotel. And these are all the requests
points for that hotel. And you can see that most people were
requesting at the front of the hotel. And most the trips,
which are in green, are beginning at the front of the hotel. There’s a massive cluster here that
we’re showing you, which makes sense. Most trips at the hotel will happen
at the front door of the hotel. But also in this complex
is Legal Sea Foods, which is a little seafood
restaurant, pretty fancy. YUKI: Also a David favorite. MATTHEW: Also a David
[INAUDIBLE] favorite. Your professor’s might
take you there sometime. And it’s kind of nestled in
the back of this complex. Well, if you look at the trips
that start from this location where the requests happen, they
mostly start over in this section. So you probably want
to have a pick up spot over on university road,
which is actually where the front door of Legal Sea Foods is. But if you use both sets of data,
the number of trips at the Charles is a much bigger location,
much more popular spot. It’s going to totally outweigh
what happens at Legal Sea Foods. So you’re probably only
going to generate one spot, and it’s going to get
the Charles, which is bad if you’re eating at Legal
Sea Foods, and you’re old and can’t walk through
the whole hotel to get there. It’s just super unideal. So by filtering by
request location, we’re actually able to generate
different points for each spot. So this is actually a
sample from the live app. If you’re at Legal
Sea Foods, we’re going to send you to
University, which is great because that’s where the front door is. And if you’re at the
Charles Hotel, we’re going to send you to Bennett
street because that’s where their front door is. So essentially we’re able
to crowdsource and learn what are the entrances and exits
are based on user behavior. So that being said, there are so many
ways we can make these pick up points better and we’re working on. One of the things I
think a lot about right now is how to refresh these and
keep these new and up-to-date. Because as the real world changes, we
have to be dynamic and change with it. So as road construction happens or
a building goes under remodeling, we need to be able to adjust quickly,
so that our riders, too, can meet their drivers in a reasonable place. YUKI: So we just talked
just about the space of that time between opening
up, requesting a car, and all the things that happened. But as you can imagine
beyond that, there’s a lot more that needs to happen. So once you request a car, everything
that happens in between that and you getting in a car,
there’s many, many more problems. So as an example just to go
through some of them quickly, how do we compute accurate ETAs
that are aware of real-time traffic? This is a thing that
actually annoys riders the most when ETAs are either
wrong or they fluctuate. We have another problem of making sure
that the car positions are actually accurate and in real-time. And there’s a lot of
interesting problems around how to deal with stale data
when we lose connectivity or the driver loses connectivity. There’s another question of how
we get riders to be at the curb exactly when the driver arrives. And so we have a class of some
lazy riders who don’t actually leave their dorms or leave their
rooms until the car actually arrives. And that produces a
different kind of problem because the driver has to pull over. And there’s an entire side
of the driver experience that we haven’t even touched. And I now work on the driver experience. And this is definitely
the most stressful thing. It’s much more stressful for the driver
to actually get a rider picked up. And then there’s this
last 50 meter problem where riders and drivers
might not be able to find each other because it’s a really dark
area, or because it’s really crowded. And so as an example, this is
where we’ve developed technology like this where a rider can
choose a color of their liking, and a light that’s attached
to the windshield, which we call a beacon, lights up in that way. So in a dark venue, or when it’s
really crowded, this might be helpful. But there are a lot of these different
problems around the pick ups. And there’s a lot more
problems outside of it too. We’re going to stay and go
through some Q&A at the end, but really encourage you to
ask us a lot of questions, and come work with us on these problems. And for that, I’m going
to turn it over to Ally, who is going to talk a little bit
about what it’s like to work at Uber. ALLI: My name is Alli Reedy. I’m a university recruiter at Uber. I’ve been with Uber for a
little bit, about 2.5 years now. Started out the university in
general is fairly new at Uber. It’s only been around
for about 2.5 years. So we helped build this
program from the ground up. So I’m going to tell you a
little bit about Uber in general, some of the opportunities
that we do have available. As I mentioned, we will
have a Q&A at the end, as well as we’ll stay after and
answer any questions that you may have one-on-one. You can have a huge impact,
obviously, if you join Uber. More people earn income from Uber
than any other privately-held company other than McDonald’s or
Wal-Mart, which is pretty crazy. You can improve cities and communities. So obviously, we have seen fewer cars on
the road, more decreases of congestion and things like that. It also frees up parking
garages, real estate, and in business and residential space. I would say one thing
that we’re extremely proud of is that we are
making cities safer. So 11% decrease in
DUI-related car accidents after it launched in Seattle,
which is pretty crazy. And I think we’ve been seeing it in a
lot of cities in other cities as well. There’s also been a 20% lower crime
rate related just to transportation, such as your traditional taxis or things
like that once we launched in Chicago. So same thing, we’ve seen that
in a bunch of other cities as well, which is awesome. I think the cool part of that
is if you are taking Uber, you know who you’re
getting in the car with. You know where your payment is
going, and you know that the trip that you’re taking. So it has reduced in
that type of way as well. How many of you guys have heard of
ATG, or Advanced Technology Group. Yeah? Awesome. So that is obviously new to Uber. We are investing in the future
with self-driving technology. ATG, like I said, is also known
as Advanced Technology Group. I would say most of our team sits
in Pittsburgh, but a lot of them have recently moved to San Francisco
to work on this type of technology, which is awesome. UberEATS– how many have used UberEATS? Yeah? Awesome. I use it everyday. It’s amazing. But we’ve expanded to
31 more college campuses since August, which has been crazy. It says with a 1 billion
run rate in October 2016. And we do hope to get to about 10
billion within the next year or so. Some engineering locations
that we do have– our headquarters is
obviously in San Francisco. We do have three offices there, actually
all about 15 minutes walking distance from each other. We recently have opened a
new office in Palo Alto, which has been around for
about six to eight months now. We have 1,700 engineers and
growing total in those locations. Seattle, New York, and Colorado
have a little bit less. Seattle is actually one of our most
growing offices in the United States. There are about 130 engineers now. But we’ll have quite a few
in the next few months or so. New York is a little bit
smaller, same with Colorado. Our Colorado office is fairly tiny,
and they mostly work on our mass teams. Here’s just a general list of
some of our engineering teams that we do have available at Uber– business intelligence platform,
consumer products, core infrastructure, maps, marketplace, and real
time systems, security, and then Uber for business. Like I said, this is just a few
general teams that we have at Uber. We do have a lot more
within these larger groups. Obviously, you want
to belong to a place. And you want to make sure that
you are involved in other groups when you do join the real world. So here’s just a few of our
groups that we do have at Uber. If you were to do an internship
or even be a full time employee, you can join these groups. If you were an intern, we
do host ERG, which is also known as Employee Resource Groups. We do have networking nights with these
groups, so Los Ubers, Woman of Uber, Parenting at Uber, Shalom, Uber Pride. This is a photo of our Uber Pride
parade that was held in San Francisco. It’s a very fun event that Uber does. Yeah, we have a lot more. This is just to name a few. I would say we definitely have a
lot of great benefits and perks at Uber as well. So as a full-time employee, you
will be receiving full benefits– medical health, dental,
vision, et cetera. We do have a free breakfast, lunch,
and dinner Monday through Friday. I would say breakfast is more of a cold
breakfast where we have like yogurt and egg, cereal, bagels, et cetera. Lunch and dinner are catered. There’s different meals every day. You will be receiving a certain
amount of monthly Uber credits per month, which is awesome. It’s probably my most
favorite perk at Uber. And I would say living in
San Francisco, Uber is life. I don’t think we’d be able
to get over the big hills or walk anywhere in the city
without Uber, so using these credits is awesome. If you do run out of your credits,
you still receive a certain percentage each month as well. You will be receiving a monthly
stipend for your phone bill. We do have unlimited paid time
off, also known as vacation. I think that’s an awesome perk too. I don’t think you really realize
like how valuable a day off is once you hit the real world. So as long as you have manager
approval, you should be good to go. Paid time off during the holidays,
I think we’re about 11 or 12, I want to say, paid time off
holidays throughout the year. We do a lot of volunteer events just
in San Francisco, New York, Seattle. Austin I know has done a few. Atlanta has as well. So that’s some of our other offices. If you are interested in
applying, you can go to this link. This link will basically show you some
of the roles that we have available. If you’re interested in a specific role,
and it’s maybe not posted on that page, feel free to come talk to me afterwards. I’m more than happy to talk with anybody
about the interview process, what we look for, and things like that. So we can do a quick Q&A. And then
we will stay around for about 20, 25 minutes afterwards if
you have any questions. YUKI: Are there any questions about
either Uber, some of the material, whatever. We’re down to hear it. Ask us anything, I guess. MATTHEW: Tech in general. AUDIENCE: So this might not
really be what you guys covered, but I was doing research on Uber
and you were doing something about electric flying cars
in terms of Uber Elevate. And I was wondering how can
we get more involved in that if we wanted to work with Uber? YUKI: So the question, for those of
you online, was about our flying cars initiative, Uber Elevate,
and how to get more involved. It’s really still a nascent effort. And it’s just being part of the
narrative of how we move people around. And that is definitely
a future possibility. And so it’s for that reason that we’re
involved in those kinds of efforts. At the moment, it’s really
just trying to figure out who the players are in the world who
are in this space, talking to them, and figuring out what model is right. So it’s just super small right
now, and if you come to Uber, there’s certainly opportunities
to work with that team, to get to know that team. And I think that’s
the best way to do it. But I think we’ll find out more as
we learn more about the technology honestly. AUDIENCE: You guys talked about
from when someone requests a ride to whey they get in the car. There’s also this
other step of matching. There’s this two-sided matching
problem that you have to solve. Can you maybe talk about
features in figuring it out. Are there any interesting features
you guys used in that respect, besides distance from driver? YUKI: The question was
about matching and what are the things that we do to
intelligently match riders and drivers, as well as riders and riders
in the case of uberPOOL. I worked a little bit on
uberPOOL, so this can speak to it. I think fundamentally, it’s
one big optimization problem. And in the case of
uberPOOL, for example, we’re trying to find riders whose
journey overlaps most with you. And that’s what economically
makes sense for us. Because the greater the overlap,
that means that we’re actually being really efficient. And so the objective
function there effectively is keeping all car seats
populated at any given time. And so that’s what we’re striving for. And there are some
interesting nuances around. For example, how long if we increase
the window for matching, as an example, then it increases their likelihood of
being able to produce a better match. But that might mean that there’s a match
that’s comes in and that’s available, but you might predict that you’ll in
10 seconds, come up with a better match that you might just discard
it and wait for the next one. So it’s this game of probability. And those are the experiments that we
do on the rider and driver side as well. Sometimes we tell you in the app
today that, OK, a car is on the way, and you have a two minute ETA,
but you’re not assigned a car until a little bit later. And that’s to give us a
little bit of breathing room to find the most optimal thing. So there are some
tricks like that we use, and certainly many, many other
factors that go into figuring out what the best optimal route is. MATTHEW: I think the other major
component to how we’re going to actually perform that match–
and there’s tons of features that go into making that decision– one of the major ones is time. So we look and see what is
the time it’s going to take for the car to get to a given rider. Because we want to minimize that time
as much as we can for all the pickups, so your pickup quality increases,
and your wait time is reduced. We also want to minimize the time the
drivers spend going to pick you up. Because we want the driver
to have as much of their time as they are online at Uber, to
have their car be full of drivers, with passengers, that they’re making
money going from point A to point B. And so if they’re spending
more time driving to a pick up, they’re spending less time earning. So we’re also trying to optimize around
making most [INAUDIBLE] of that time. And then there’s a whole other
component on top of that time. There’s a lot of complexity with
the pricing of the marketplace itself, which is more outside
of both of our domains. There’s an entire team of just folks
that just takes these various inputs, and then performs an optimization,
basically the drivers time. And also the amount that we’re
paying the driver, and the estimated fare that we’re
receiving from the rider. So those are the components
that are at play. And it’s constantly changing how
the optimization is being performed. And it varies also by product. That’s a really good question though. AUDIENCE: I was wondering if you
guys could talk a little bit more personally to your roles
as product managers, and maybe how that works
on a day in, day out basis. Perhaps you could compare it
to a software engineering role if you guys have any
experience with that. MATTHEW: The question
was can we describe a bit what it’s like to
work with a product manager, maybe compare it with like an IC
role, like a software engineer. So my experience at Uber and
actually at other companies as well. I worked at Google before. A typical setup for a
team structure is you’re going have a product manager,
software engineers, data science, and, if you’re working on a product that
has some sort of experience component, a design designer and
maybe a user researcher. The project manager in this
sense is the glue of this group. You’re working to leverage the skills
that each person brings to the table, and try to effectively move toward some
shared vision utilizing those skills. So there’s a real tactical
component, just day-to-day, making sure that those folks
are all working together, thing are being unblocked
and they’re on time. They’re kind of like project manage-y. But then at a high level,
you’re also helping to develop what’s the thing that
we all should be working on. What’s that vision for this group. And that’s really where
you’re taking input both from your team, and from
outside of industry, from research, to help [INAUDIBLE] what
you’re doing long term. You have to wear both hats. And the best product managers
do both of those things. But it’s fun, because you
get to be in the weeds, but you also get to
be kind of high level. That’s my take on it. YUKI: Yeah, the way I think
about some of the distinctions is, ultimately, a product manager
is responsible for the why. Why are we doing this? Why should we invest in this? The what comes from many
parts of the company. There are ideas floating
around everywhere. And in terms of the how, that’s one
of the places where the engineers are really involved in and designers in
figuring out exactly how we’re solving this particular solution or problem. But it’s the product manager
to decide what problem we’re solving in the first place. So that’s one thing. And even in a pickup space,
a bad pick up is a problem. But maybe our jobs might be to
break that down into specific areas, like what part of that
pickup is the problem. And then apply our team to
figure out solutions for it. So that’s one way I think about it. MATTHEW: And comparing it to an IC role,
I did work in an IC role at Google. Sorry, individual contributor, working
basically as a component to that team, as opposed to being a manager. Some of the differences are when
you are an IC, you problem space is a little bit more defined. You spend more of your
time on your own, not in meetings, just
working on that problem, writing code, performing analyses,
or whatever you might be doing. So it’s a lot more like you’re
spending time with fewer people and more with an idea and tools. Whereas when you’re a PM, we spend
a lot more of our time with people, and talking, and writing
documents, and clarifying, roadmapping why you’re doing things. So you spend more time with ideas
and people, as opposed to with tools. So it’s just a different
way of spending time. They’re bot satisfying. And the skills you if you work
in an IC capacity as an engineer, those skills are totally transferable to
management and backwards and forwards. YUKI: And to answer your second
part of the question of how it’s different from a place like
Google, or I also worked at Microsoft. I would say the biggest thing is the
operational nature of Uber’s work. So for example, we experimented
a couple of years ago– and this is still part
of the experience today– in New York, getting people
to walk to the street corners. And that would be more efficient,
so cars can go up and down avenues, instead of driving all over the place. And so that’s the kind
of thing where we need to work with the local
team on the ground to make sure that everyone
understands what’s happening. They might have to change
their support flows, or they might be getting
a lot of inbound about it. And so there’s aspect
of this physical world and the team on the ground that you
have to be really thinking about, which is very different at Uber. I felt that at Microsoft or Google,
you could do this all remotely, and manage it, and it’s fine. So that’s one big difference. And another one is just the size. Uber is a big company now, but even
then, it still feels like a startup. And one of the things about
that is, as a product manager, you’re responsible for your
team, and how well they’re doing, and how they’re feeling, and all these
other aspects of even like recruiting and things like that. you’re given a much
larger responsibility. So that’s something I felt as I’ve
transitioned from different companies. AUDIENCE: So over the
last few years, I’ve just been transitioning from
a scrappy startup that’s disrupting a massive space to
a more established company, kind of as you were touching on briefly. I’m curious how things are changing
at Uber, like structurally, like relationship, how it flows
between engineers and management. What the individual’s role looks
like now for someone like me who is trying to find a
position right out of college. How has that changed
over the last few years? How is it projected to change? YUKI: So the question was
about Uber as a company, it’s becoming bigger and growing,
and how the dynamics of the workplace changed as a result
of that. and what it’s like as am individual
contributor working within that kind of environment. So I can take that part of it. So I think, yeah, it’s definitely true. When Matt and I joined, we were about
2000 people, and now we’re like 15,000. And so that definitely
changes a lot of things. And it’s certainly the case that we
have real-world responsibilities now. We can’t just launch something in
some random city and not tell anyone. And that kind of thing
happened a lot when there were a lot of these autonomous teams. So there is a certain aspect of we
do need to be talking with each other and communicating. And there certainly needs to be some
process in place, even a marketing campaign that one city
runs that actually people need to know because they
could affect many other people. You could have competing policies,
all these kinds of things. So there’s the reality
that as we grow bigger and we have millions of
millions of riders and millions and millions of drivers
whose lives we’re supporting, we have to put some process in place. So for that, I think
there’s a little bit more of the checks and balances
that are necessary. But I think parts of us
are growing that we’re really conscious of is making
sure that everyone still feels that sense of autonomy. And I think the way we’re
organized is actually a good way that facilitates that,
which is if you’re in a team, let’s say there’s a team
called the pick up experience. And their sole mission is
to make the pickup great. And so they actually
have a full stock set of engineers across both Android and
iOS, as well as deep inside the mapping stack. There is a person an ops, product
manager, designers, data scientists. And they basically have
the expertise to make this happen without creating
too many more dependencies. So that’s how Uber is
organized in terms of teams around these missions that
can execute on their own. And so the setup has this autonomy. And it’s really then just
a matter of making sure that there is a process around
that in terms of communication where we’re all talking,
so it’s not always silent. So it’s definitely a
delicate balancing act. And I would say that one of the
interesting things right now of joining is that you’re in this
kind of like gray space where we’re trying to figure out what
the most effective way to do this is and contributing to that process. For example, at a place like Google,
they have everything figured out. So there’s this launch process
that’s quite sophisticated. But maybe that’s not the
right approach for Uber, and we’re still trying to figure out. And that’s part of what it
means to come and work at Uber. MATTHEW: And one other thing to consider
too, especially as a recent grad, is the kind of mentorship
you’re going to get and experiences you’re going to have
when you’re taking on that first job. And I think one of the benefits of
Uber growing over the past few years, and now having doubled and tripled
in size since we’ve joined, is that there is more people to
actually mentor and help people grow. In the earlier days, you showed
up, and it was like, all right, figure out how to make this thing work. And there’s no one who’s
really going to hold your hand. And it was a lot to figuring it
out on your own, which is find. That’s a way to grow. But from my experience,
having come from Google, it was really nice to
be in a place where there was a layer of management who
were people who were a few years older than me who would teach
me how to do x, y and z, and help me hone my own skills. I think as we’re maturing as a
company, we’re getting to a place where we have those people. People have time to be able to take
on recent grads and help train them and onboard them an stuff. And that is a really important
part of any first job, just having that mentorship. And I think that’s
something that we now are able to offer, which is really cool. AUDIENCE: How does
internships work at Uber? And what do you look for in an intern? ALLI: The question was, what
is an internship program like, and what do we look for in an intern? Correct? AUDIENCE: Yeah. ALLI: I’ll tell you a little bit about
our internship program in general. So typically, it’s about a
12 to 16-week internship, depending on which school you go to. We do hire interns year round. I would say our biggest intern
class is obviously in the summer. But then we do have a fall intern class. I believe our fall intern class
is about 35 interns right now. And then in the spring, it’s
roughly about the same too. Over summer, we probably
have about 180 and growing. So there’s obviously a lot
more intern events and things like that during the summer. So that’s the program itself. There’s a few different
internships that we have available. So if you are looking for a
software engineer internship, basically what will happen is you
will interview as a generalist. And what will happen is
that we will allocate you a team, most likely about
three months prior to starting. And the reason we do that is just
because Uber is constantly changing, organizing team, things like that. So we’ll allocate you to
a team three months prior. You actually get to
choose your team too. We typically give you a survey of about
20 to 25 that you can choose from. I would say most interns did
get their top two this year. If they didn’t get their
top two, it was maybe because like they wanted UberEATS,
and they wanted Palo Alto. UberEATS isn’t in Palo Alto, so
they had to work in San Francisco. So things like that. Everybody loves having an
intern, which is awesome. Yep, you’ve had them. YUKI: Yea, I’ve had them. I’ve had the the maps routing team. And we have had the most phenomenal
interns every quarter, every semester. You’d be amazed. So I used to work at Google, so
I’ve had interns at other companies. You’d be amazed the things at
Uber that the interns build. Things that you’ve all used in the
app, interns have totally built and they’ve launched within their
internships, which just blows my mind. Because other bigger companies, it’s
just hard to ramp up in three months and actually ship something. But we have interns contribute
really amazing features to the app. So I’ve been blown away
with some of our interns. Most of them have come back now. Some of them are still in school. But yeah, pretty cool. ALLI: That’s one of the things I
always tell our interns as well. If you’re looking to make
an impact at a company, and you’re looking to
work on meaningful work, and you’re looking to work on a
project that really interests you, then Uber is probably the place to be. If you’re looking for an
internship that’s like, I don’t really feel like working
today, or I want fun all day long, that probably
isn’t the place for you. And we’ve done a pretty
good job at making sure that each intern works on
different projects that impact the company like Matt was just saying. We’ve had a few interns work
on projects that have been implemented into the entire company. I think we’ve have an intern
speak at our All Hands meetings about the type of project that they
worked on within a 12-week span. So I wouldn’t say that we
look for any specific things. I would say we really
are looking for passion and looking for somebody
who wants to learn and who wants to grow their career. From most of my interns that
I’ve had over the past 2.5 years, I would say majority of them have said
this is the most challenging internship they’ve had. Because they have never learned
so much within a 12-week span. MATTHEW: You’re basically treated
like a full-time employee. It’s pretty cool. AUDIENCE: I’m asking this
for my little brother. Do you guys do freshman
interns after freshman year? ALLI: We do– freshman friendly. AUDIENCE: Do you guys
draw a clear distinction between project managers
and individual contributors, or is there an opportunity to do both? YUKI: The question was about product
managers and individual contributors. I think individual contributors in that
context is a little bit of a misnomer. Product managers are individual
contributors, too, in some ways. It just so happens that you’re
working with many people. And so that’s the primary distinction. But there are definitely
managers of engineers, managers of product managers, and things
like that, especially off the bat. Everyone is essentially
an individual contributor. All right, we’ll stick around. But thanks for showing up.

5 thoughts on “Building Products at Uber: How We Bridge the Physical and Digital Worlds

  1. Frankly speaking those "uhhmmms" seen to be very common in the USA. Is the educational system not teaching the children at the very tender age to be able to flow in their sentences rather that using conjunctions, they most of the time want to use hmmms

Leave a Reply

Your email address will not be published. Required fields are marked *