00:12my name is crystal so I have been at
00:16go-jek for two years now I'm gonna speak
00:20in English so I'm sorry but my Indonesia
00:25I can understand it but I have grown up
00:29in the States my whole life but I came
00:31back to Indonesia because of gojaks
00:33potential so when I first heard about go
00:36Jack there was go ride and this was a
00:39really interesting opportunity because
00:43there are so many opportunities for
00:45growth in terms of infrastructure
00:47mobilizing the the informal labour
00:51economy we had go son go ride go Mart
00:56and go food so I thought that my job
01:00here would be very easy there were only
01:024 products we had pretty sizeable demand
01:05but it wasn't anything crazy
01:07we had a monolith database and
01:10everything could easily be accessed in a
01:12my sequel dB and then a couple months
01:16went by and then we had three more
01:18services and I said that's okay I can
01:20handle three more services you know
01:23we'll have a couple more data points we
01:24have to add a few new dimensions but
01:26that's ok everything's still stored in
01:28one place let's just keep pumping the
01:30data into a single area and a couple
01:33more months go by and I said ok now we
01:37have quite a few more services expanding
01:40in our ecosystem suddenly there are new
01:43stores of data they're not all in a
01:46monolith service some of it is now in a
01:50some of it is in Postgres some of it is
01:53in a log stash that isn't even being
01:54stored yet and so today you have a
01:59couple more and you'll start to see even
02:03more coming pretty soon and because of
02:06this our philosophy at the very
02:08beginning was always about let's just
02:11make sure we store the data and create
02:14an environment where at least we have
02:16data because even bad data is better
02:19than no data provide an environment
02:21where at least people can make decisions
02:25and understand what's happening in their
02:27products even if it isn't as user
02:29friendly as it could be right because
02:32go-jek was moving so fast and so
02:34fearless that we had no choice it was
02:36either accept the data as it is and
02:38figure out the standardization later
02:40so as the micro-services grew and our
02:43kind of wealth of product offerings grew
02:46so did the scale and so we were pumping
02:50all of this data in a raw JSON format
02:53maybe it was unstructured somehow we
02:55didn't care we had clickstream data
02:58coming through our systems we had order
03:01management system data coming through
03:04we had driver locations and every time
03:07someone thought of a new data service or
03:09a new feature to add to a product they
03:11would build a new micro service and so
03:14with the level of data being created we
03:17were starting to have a really big
03:18problem our data ended up looking a lot
03:22like this just thrown into a single
03:26repository with no organization
03:28whatsoever we told ourselves ok we'll
03:31just take it from every micro service
03:33we'll store it in the database let
03:36people figure it out later at least
03:38everyone has access to something data
03:40scientists can tap in developers can
03:43look at their feature product owners
03:45will oh crap what about the product
03:48owner so product owners would actually
03:50go into our visualization tools or
03:53they'd go into even just the sequel
03:55front-end and they'd say well it's a lot
03:58of data but I don't actually know how to
04:00use this because you have so many micro
04:02services I get it there are products
04:04that we've built that are serving a
04:06particular function but I only know how
04:09to use what you guys have provided if I
04:12have a very specific question about a
04:13very specific feature if I want to look
04:16at the feedback ratings of all our
04:19orders I know how to find that but what
04:22do I do with that information how do I
04:24make better decisions about it so we had
04:27come into basically an area where we
04:31knew we had data and it had so much
04:33potential but no one was able to use it
04:35in the structure that it was
04:38we started to think about how we might
04:40want to organize our data how could we
04:42make it easy for people to discover new
04:44metrics and how could it we make it so
04:47that people could explore new data
04:50potentials rather than just asking the
04:52same questions over and over again
04:54because like we'll except to build
04:57growth you need to be extremely creative
04:58you have to find those nuggets of data
05:01and those insights that no one else
05:04would know how to find so now we had
05:08this data closet and I think the
05:12metaphor here is is that when you
05:14organize your closet you're trying to
05:16optimize for something you don't want to
05:18wear the same things every day you want
05:20to be able to mix and match different
05:22clothing items you want to be able to
05:24accessorize with all of the different
05:26things in your closet and not have a
05:29static outfit and so when you organize
05:32your closet you can organize it in
05:35several ways so you can categorize it by
05:39outfit so you can match all of the
05:41things that you know you wear together
05:42you can categorize by color so that you
05:45can understand okay these things match
05:47well together or you can categorize by
05:49clothing type putting all your t-shirts
05:51in one area all your shoes in one area
05:53and the way that we thought about how we
05:56were organizing our data at gojek was a
05:58bit similar so you could organize by
06:01outfit right so go food will have its
06:03perfect data set where it has all of it
06:06booking data has the drivers attached to
06:08go food it has the prices of the go food
06:11orders there are merchants in this tidy
06:14data set so that a go food person can
06:17just walk into the data closet and say
06:19oh there that's my go food data let me
06:21find out you know the answers to my
06:23question about Jessica food you could
06:26also organize it by business unit right
06:29so all the finance people will come in
06:31and say oh here's my accounting data
06:34here is the data that is just about
06:36surge pricing here is the data that
06:39tells me about my revenues or you can
06:42organize by clothing type
06:45so in this it would be for every booking
06:47being stored in a specific location all
06:50of the bid data being stored in a
06:53specific location and all of the
06:56cancellation orders being in a specific
06:57location we ended up thinking based on
07:02these different categories that people
07:04were used to finding their data in how
07:08can we structure it such that people are
07:10using the data in a way that is
07:12efficient and effective for them to
07:15build on their product so it leads us to
07:18the North Star metric completed
07:21transactions so the North Star metric is
07:23something that you want to align the
07:25organization behind so that everyone in
07:27the company knows what the company's
07:29goal is and so at go-jek that would be
07:32completed transactions as we increase
07:34the number of completed transactions we
07:36are making our goal and every product
07:39owner wants to know how can I contribute
07:41to the North Star metric at go Jack so
07:44now you need to go into the metrics that
07:47matter so what metrics matter when we
07:50want to increase the number of completed
07:52transactions well you have to increase
07:55go ride completed transactions go food
07:58completed transactions we'll add to that
08:00North Star metric go pay p2p
08:02transactions we'll add to that metric so
08:05you start to break down the categories
08:06and the layers that the data needs to be
08:09aligned by now for go food completed
08:13bookings what are the metrics that
08:14matter for that because you can have the
08:16product owner caring about go food
08:19completed orders but what about all of
08:22who are working on specific features and
08:24they don't know exactly how their
08:27feature works towards the North Star
08:29metric well then you go into total
08:33bookings right if there are no bookings
08:36on go food obviously there can't be go
08:37food completed bookings there will be
08:41allocation for every booking that does
08:44happen how do we ensure that there is a
08:46driver to accept that order and complete
08:48it and then cancellations for every
08:52order that is placed and then gets a
08:55driver how do we ensure that a
08:56cancellation does not occur
08:58because when he goes to those restaurant
09:00or he goes to the store the item is out
09:03of stock so how do we prevent
09:05cancellations so now PMS can rally
09:08around these specific metrics that
09:10matter to the metric that matters to the
09:13metric of the Northstar metric and they
09:15understand how their features are
09:17helping either reduce cancellation rates
09:20improve allocation or increase total
09:23bookings now the metrics that matter to
09:26the measures that matter are things that
09:27we align the business units around as
09:30well as build our tableau dashboards or
09:33our visualization tools around so when
09:37you go into a visualization tool at
09:39go-jek it's always centered around one
09:41of these kind of concepts so when you
09:45look at total bookings what you need to
09:47have in order to increase total bookings
09:50are obviously active users on your
09:52platform to complete orders as well as
09:55merchants for people to book from for
09:58when you want to understand how to
10:00improve allocation you need to know
10:03where drivers are located and you need
10:06to know whether or not they are
10:08incentivized well enough to complete
10:10these orders because just it just
10:13because you have supply doesn't mean
10:15that these drivers are actually going to
10:17complete these orders and for
10:20cancellation rates we would need to
10:22understand how long it takes a customer
10:24to get a driver so that they don't
10:26cancel because they're being impatient
10:29and we need to understand when a driver
10:32goes to the restaurant and there's a
10:33stock out how could we have prevented
10:35that content quality issue so that this
10:39issue doesn't happen anywhere and so now
10:40you have all of these sub features of
10:43the product that different PM's and
10:46business owners can rally around because
10:48they understand exactly how what they
10:51are doing impacts the metric that
10:53matters to the metric that matters and
10:54so on and now there's always I guess on
11:00the technical side people are find it a
11:02bit harder to understand like oh how is
11:04my work you know helping complete
11:06transactions so for us on the developer
11:09side we're always focused on
11:11service of time driver supply hours so
11:15back to this now we knew what our
11:18Northstar metrics were we had aligned
11:20the organization around it the BI team
11:22had gone to each product owner and said
11:24hey this is our Northstar metric this is
11:27what we expect in terms of KPIs from
11:29each product how should we organize this
11:32so that we can efficiently enable
11:35product owners and business units to
11:37find their own aha moment on their own
11:39so that we don't have to constantly go
11:41to them and say hey are you looking at
11:43allocation rates are you looking at
11:45cancellation rates and instead they
11:47would be able to go into this and
11:49understand it a bit better so what we
11:52decided to do was a couple of things we
11:54decided to organize it in different ways
11:57but use the same data so we weren't too
12:01concerned about duplication of data
12:03being represented in our data warehouse
12:05because we're mostly on a cloud platform
12:07and because of that storage is cheap you
12:10can duplicate data as long as it
12:11improves the references so we decided to
12:15look at one style of organization where
12:19we consider what product owners were
12:22commonly coming into the data warehouse
12:23for they would say oh I want to
12:25understand something about my customer
12:26or I want to understand something about
12:28my drivers but they would often miss out
12:30on what was in between those two things
12:32which are you know feedback ratings that
12:35tie them together the bookings that they
12:37complete together or things as silly as
12:40the weather which everyone probably
12:42noticed this morning so on this what we
12:46would have a product owner do is they'd
12:48come in and they say ok I want to look
12:49at you know the customers who are using
12:51go food or I want to look at the
12:53customers who are using go points
12:55vouchers and from here they would be
12:58forced to almost to find those
13:00relationships between every single
13:02possible shared property between a
13:04driver because on our platform were
13:07really interested in not just promoting
13:11the customer experience but also the
13:13driver experience with it because the
13:14drivers are always our agents towards
13:17the customers so the other way that we
13:20wanted to organize this was by event
13:23type and shared properties now everyone
13:27at the company is kind of focused on
13:28different streams of work and in doing
13:31this we kind of forced different mmm
13:38shared properties so that they would be
13:41forced to not look at just a single
13:43problem when people were looking at go
13:48food allocation rates they noticed that
13:50the cancellation rate was very high so
13:52they would look into the bookings they'd
13:54only look at bookings and say oh wow
13:55constantly know we have a lot of
13:57cancellations but they weren't really
13:59looking at things like location or they
14:02weren't looking at things like driver
14:03incentives because that was very far
14:05from the concept of bookings so by
14:09combining you know things like a go-cart
14:13booking or a go food booking - driver
14:16statistics like his performance overall
14:19on the platform and then - the unit
14:21economics so what was the price that a
14:24driver was being given for every order
14:26that he completed and for canceled
14:28bookings how did that compare to
14:30completed orders now this was a feature
14:32that they could look at as a data metric
14:35and compared to all in one place one
14:40example of an aha moment that we looked
14:42at here was in looking at how go points
14:45vouchers were being adopted so we looked
14:48at all of the adoption rates of go
14:50points vouchers you can buy vouchers in
14:53our app and redeem them at a store for
14:57users who are using alpha Mart belcher's
15:00we didn't just look at how they redeemed
15:03them in store we don't just look at the
15:04time stamp but we link it to the user's
15:06profile level what was he doing at that
15:08time and by looking at his user profile
15:11level you could see that oh he had just
15:14completed an order so this person was
15:16literally at the store he bought a go
15:18points voucher for alpha Mart and
15:20redeemed it in the store and now this
15:23made us wonder couldn't we tie go points
15:26voucher data more closely with our go
15:28ride data go points is a completely
15:30separate team while go ride is
15:34focus mostly on transport but this data
15:36could be easily linked together you
15:38could give a more contextual experience
15:40to the users and these are two
15:42completely different product lines that
15:44most people wouldn't have considered as
15:46I grow the strategy that you could push
15:48and incentivize a user and contact them
15:51at the right time based on two
15:52completely different data points having
15:58the experience of setting up our data
16:01way that we could explore and be very
16:04creative has allowed us to do a lot of
16:07different data blog posts in a very
16:09short amount of time for us to use the
16:14data warehouse as it is now essentially
16:16we can just start with a single question
16:17what is go Jackson impact for Indonesia
16:20well a lot of people would look at that
16:22and say well we've uplifted a lot of
16:25people's income right we've created a
16:28lot of jobs and opportunities for
16:29drivers so let's look at that one
16:31feature drive our unit economics and
16:35let's see where that story takes us so
16:38for us it becomes an exploratory
16:40experience where you just start with a
16:41single question and it leads you towards
16:43other different unexpected relationships
16:48with data points that you hadn't even
16:49considered in the beginning but because
16:51the data was so linked together you're
16:54almost forced to realize that there are
16:56connections there that you hadn't
16:58expected before do you know how many
17:01more topics we have delivered in the
17:03past year yes so we've delivered three
17:07million more taluk in the past year and
17:10this was kind of a data point that we
17:11didn't even have to really search for
17:15it's just something that occurred when
17:17we were looking at basic go food data
17:23when we wanted to write our most recent
17:26blog posts on sudhir Mun it was not hard
17:29for us to find all of these different
17:31data points we didn't have to take a lot
17:34of time to expand on ok what does all
17:37the data in Sudan like how do we pull
17:39this data because all of our existing
17:43data points are matched with low
17:46station based data wherever they can be
17:48all we had to do was literally type in
17:50Sudhir maan as a location point and all
17:53of the data that was represented on a
17:56location based on a location based
18:00granularity would pop up so you didn't
18:03need to say I want specifically okay how
18:05many go food orders are being sent in
18:08Sudirman all you had to do was say
18:10what's happening in Sudhir maan I think
18:14on the traffic side it was quite easy
18:17for us as well so in understanding what
18:20happened in Sudhir maan and what its
18:22effect what the ban might have an effect
18:25on we initially looked at how many
18:30orders when that were in the area we
18:32took a look at how many drivers were
18:36being picked up and dropped off in that
18:37area and because we had standardized our
18:39data to the point where it's using these
18:41s2 ids which are a open source Google
18:45library product that translates latitude
18:48and longitudes into geographical cells
18:50we had standardized our data so that you
18:53could look up Geographic data across all
18:55different products across all different
18:59understandings and looking event levels
19:03like driver location pings these weren't
19:06even on a booking level but we could
19:09still identify that drivers were in that
19:11location at that time so to read more
19:15please go to our go check data blog
