Kongregate has been sharing great data on web games for years - now they're launching mobile games and are ready to start sharing iOS and Android numbers as well. Learn how monetization differs by platform, country, and device, what type of CPIs to expect for different kinds of games, and what's most important in making your free-to-play mobile game a success.
GDC 2014 Core Games, Real Numbers: Going Cross-Platform
1. Hi
there,
thank
you
for
coming.
My
name
is
Emily
Greer,
and
I’m
the
co-‐founder
&
CEO
of
Kongregate,
the
browser
games
plaDorm
and
now
a
mobile
games
publisher
as
well
1
2. That’s
what
I’m
going
to
talk
about
today,
this
is
going
to
be
something
of
a
post-‐
mortem
on
our
first
12
months
as
a
mobile
pubisher
with
the
good,
the
bad,
and
as
much
data
as
I
can
share.
2
3. My
brother,
who
is
my
co-‐founder,
had
previously
co-‐founded
a
small
game
studio
and
had
a
bad
experience
with
a
major
publisher
that
failed
to
market
their
game,
then
refused
to
release
the
rights
to
do
the
sequel,
ulMmately
leading
to
the
studio
dissolving.
We
wanted
to
open
up
more
opMons
for
developers
so
they
DIDN’T
need
a
publisher
to
make
money
on
their
game.
There
were
a
ton
of
generic
browser
game
sites
out
there,
with
no
persistence
between
games
and
no
connecMon
beyond
leaderboards
between
players.
7
years
later
we’ve
got
more
than
80,000
games
on
our
plaDorm.
Most
of
our
web
revenue
comes
from
the
400
free-‐to-‐play
games,
primarily
MMOs
and
CCGs,
and
the
business
conMnues
to
grow
strongly
despite
dire
predicMons
of
the
death
of
web.
Our
web
traffic
grew
about
30%
in
2013,
and
our
revenue
grew
55%.
GeYng
into
mobile
is
an
expansion
of
our
current
business,
not
a
replacement.
3
4. It’s
not
like
we
just
started
thinking
about
mobile
last
year.
We’ve
been
thinking
about
it
since
the
launch
of
the
iPhone.
But
the
app
store
was
such
a
closed
ecosystem
that
we
didn’t
see
what
sort
of
role
an
social
plaDorm/distributor
could
really
play
without
being
shut
down
by
Apple.
When
Android
came
out
we
thought
there
was
more
of
an
opportunity
since
it
was
a
more
open
system.
So
we
built
a
light
version
of
the
web
plaDorm
for
Android
called
Arcade.
But
it
turned
out
that
Android
isn’t
so
open
to
anything
it
perceives
as
an
alternate
app
store,
and
Arcade
got
yanked
the
first
day
we
launched.
A_er
some
discussion
we
reworked
some
features
and
were
allowed
back
up,
but
we
were
quite
limited
because
we
couldn’t
a)
sell
anything
or
b)
give
the
appearance
of
downloading
games.
UlMmately
those
limitaMons
fundamentally
blocked
any
real
revenue
opportuniMes
for
the
Kongregate
Arcade,
and
we
stopped
development
in
late
2011
when
Adobe
announce
they
were
killing
mobile
Flash.
But
you
learn
as
much
or
more
from
your
failures
as
your
successes
and
we
got
a
lot
out
of
the
experience
of
building
Arcade,
including
a
fair
amount
of
tech.
Some
other
things
we
learned
is
that
despite
the
promises
of
HTML
5
and
Flash
mobile
the
performance
of
browser
games
on
mobile
was
consistently
bad.
On
the
good
side
we
learned
that
we
were
able
to
drive
a
lot
of
traffic
from
Kongregate
&
Gamestop,
and
are
now
up
to
2
million
installs
despite
mediocre
game
experiences
and
zero
UA.
4
5. A_er
we
stopped
development
on
Arcade
we
stepped
back
and
reconsidered
our
strategy.
In
mobile
Apple
&
Google
are
the
game
plaDorms,
but
the
problem
we’ve
always
striven
to
solve
with
our
plaDorm
–
developer
friendly
distribuMon
–
sMll
exists.
Demand
for
games
is
fundamentally
inelasMc,
and
the
race
to
the
bofom
on
pricing
in
the
early
days
of
the
app
store
has
put
developers
in
a
hole
it’s
hard
to
climb
out
of.
Prices
for
paid
games
would
probably
have
to
be
in
the
$9.99-‐$19.99
range
you
see
in
the
PC
market
for
it
to
be
more
viable
We
know
because
we
got
most
of
it
wrong
when
we
launched
our
kreds
plaDorm
back
in
2008.
But
over
5
years
and
500+
games
doing
F2P
on
our
plaDorm
we’ve
learned
a
lot.
I
expect
a
lot
of
you
have
seen
previous
talks
we’ve
done.
So
who
helps
developers
with
distribuMon
&
moneMzaMon?
A
publisher,
of
course.
5
6. We
announced
a
$10M
publishing
fund
last
February
Since
that’s
what
we
do
well
with
on
the
web
Not
every
game
is
going
to
succeed
but
we
want
every
developer
to
feel
we’ve
treated
them
well
work
with
us
again
To
put
our
skin
in
the
game
along
with
the
dev.
A
publisher
should
share
the
risk
for
it
to
be
a
true
partnership
Cross-‐promoMon
from
Kongregate
web,
our
parent
company
Gamestop,
and
porDolio
games
6
7. 6
of
the
live
games
are
new
Mtles,
but
2
(Lifle
Alchemist
&
Dragon
Storm
Gold)
were
live
games
we’ve
taken
over
in
the
last
couple
of
months.
It’s
about
half
core
Mtles,
including
3
CCGs
(Tyrant,
Bloodrealm,
and
Lifle
Alchemist)
and
half
casual
Mtles
like
Sheep
Happens,
Run,
and
Endless
Boss
Fight.
7
8. Here’s
a
list
of
the
Mtles
we
have
out
and
what
plaDorms
they’re
available
on
–
our
goal
is
to
have
every
game
we
publish
available
on
Kong
web,
iOS,
and
Android
to
maximize
success,
and
several
of
our
iniMal
games
are
properMes
already
successful
on
Kong
web.
Tyrant
and
Sheep
Happens
have
been
out
for
at
least
six
months
on
all
plaDorms,
but
most
of
the
rest
of
the
live
games
are
only
on
one
plaDorm
or
have
launched
very
recently,
usually
both.
Because
of
that
my
talk
is
going
to
concentrate
most
heavily
on
Tyrant,
which
has
been
out
the
longest
and
has
the
best
quality
data.
8
9. Since
I’m
going
to
spend
so
much
of
the
talk
talking
about
Tyrant
I
thought
I’d
give
it
a
larger
introducMon.
Tyrant
is
a
CCG
with
a
grify
sci-‐fi
theme
built
originally
as
a
Facebook
game
by
Synapse
Games,
a
small
studio
in
Chicago.
They
brought
it
to
Kongregate
three
years
ago
and
it’s
been
one
of
our
most
popular
games
ever
since
with
excepMonal
retenMon
and
good
if
not
great
moneMzaMon.
When
Synapse
started
working
with
us
to
publish
the
mobile
version
moneMzaMon
was
one
of
the
things
we
focused
on
since
we
felt
that
there
was
a
big
opportunity
to
improve
the
depth
of
spend
possible,
mostly
by
adding
in
some
deeper
card
progression
and
fusion
systems,
but
also
by
increasing
the
card
prices.
Those
changes
were
quite
effecMve,
and
though
retenMon
decreased
somewhat
relaMve
to
the
original
version
the
overall
moneMzaMon
improved
significantly.
9
10. It
hit
the
top
60
in
the
gross
ranks
and
has
conMnued
to
grow.
10
11. I
menMoned
paid
UA
in
relaMon
to
scaling
so
let’s
start
with
what
everyone
is
worrying
about.
Here
are
the
CPIs
we
by
paid
by
ad
network
for
Tyrant
in
Decemer:
they
ranged
from
$2-‐$8,
with
highly
targeted
campaigns
and
video
on
the
high
end.
These
are
prefy
inMmidaMng
#s,
but
with
a
high
LTV
game
like
Tyrant
you
can
sMll
do
it
profitably,
though
not
every
campaign
listed
here
is.
However
with
an
extended
revenue
curve
like
Tyrant
has
(only
~17%
of
the
6
month
ARPU
comes
in
the
first
month)
it
takes
both
paMence
and
deep
pockets
to
wait
for
the
UA
to
pay
off.
11
12. But
that’s
not
the
whole
story:
CPIs
wary
wildy
by
game.
We’ve
used
just
one
network
to
get
an
apples-‐to-‐apples
comparison
there
so
here
our
average
CPIs
since
September
for
most
of
our
games.
Games
in
hard-‐core,
compeMMve
genres
like
CCGs
and
MMOs
have
much
higher
CPIs
than
more
casual
games,
but
if
the
theme
is
appealing
and/or
the
graphics
are
very
strong
the
price
comes
down
a
lot.
12
13. Now
take
the
daily
#s
here
with
a
grain
of
salt
because
the
sample
size
on
a
lot
of
these
days
is
small
and
there
were
a
lot
of
moving
parts,
but
here
are
the
CPIs
on
Facebook
for
Lionheart
over
the
last
90
days,
all
but
the
last
few
of
which
were
in
test
markets.
As
you
can
see
CPIs
are
both
higher
in
test
markets
than
they
are
globally
and
that
in
each
market
they
start
lower
and
rise
over
Mme.
This
is
a
natural
result
of
saturaMng
a
market
and
a
source,
but
was
exaggerated
in
this
case
because
we
weren’t
tesMng
and
refreshing
creaMves
at
the
rate
that
we
would
for
a
game
in
wide
release.
To
fight
the
CPI
creep
you
need
to
be
tesMng
and
refreshing
creaMves
at
least
every
6
weeks.
13
14. A
lot
of
the
reason
we
can
make
high
CPIs
is
that
we
credit
paid
campaigns
with
some
value
from
organic
users.
In
our
office
there
are
two
camps
on
what
mostly
drives
organics:
one
is
that
paid
installs
are
helping
to
drive
chart
posiMon,
and
chart
posiMon
then
drives
organic
installs,
the
other
that
players
are
telling
their
friends
about
the
game.
I
think
organics
are
a
complicated
phenomenon
with
elements
of
both,
but
long-‐term
I
think
word
of
mouth
is
more
important,
especially
if
you’re
not
in
the
upper
parts
of
the
charts
that
get
extensive
exposure.
This
chart
shows
our
organic
and
paid
installs
for
Tyrant
on
IOS
since
launch,
along
with
our
retained
DAU,
ie
our
DAU
less
new
installs
to
just
show
returning
users.
Early
on
there’s
a
prefy
clear
relaMon
between
UA
and
organics,
parMcularly
noMceable
where
we
spiked
UA
around
8/28
and
again
in
October.
But
long
term
the
relaMonship
has
gofen
weaker
and
weaker,
to
the
point
now
where
organics
are
higher
than
installs
from
paid
UA,
and
follow
the
pafern
of
our
retained
DAU
much
more
closely.
I
think
that
while
people
are
most
likely
to
share
a
game
when
they
start
playing
it
but
there’s
sMll
some
chance
of
sharing
it
at
any
point
while
they
conMnue
playing
it,
so
the
value
of
paid
UA
to
installs
is
long-‐term
and
cumulaMve
and
likely
cross-‐plaDorm,
since
the
friend
of
the
iPhone
user
might
have
an
Android,
and
vice
versa.
Also
changes
in
your
game
can
help
drive
word
of
mouth.
Anyone
who’s
been
14
15. It
was
prefy
clear
from
the
last
chart
that
while
UA
is
important,
the
app
store
feature
we
got
for
Tyrant
along
with
the
Gamestop
promoMon
were
just
as
important
in
building
our
DAU.
The
effect
is
much
more
dramaMc
on
Android
since
we
were
able
to
get
more
and
larger
features
there.
We’ve
been
able
to
secure
features
on
for
almost
all
of
the
games
we
have
live
globally
on
each
plaDorm
so
are
starMng
to
get
a
feel
for
how
many
installs
you
get
from
parMcular
types
of
features.
Apple
generally
features
games
primarily
at
launch,
prefers
indie-‐feeling
and
more
casual
games,
and
keeps
their
categories/placements
fairly
consistent.
Google
is
more
open
to
non-‐launch
features
and
refeatures
of
games
with
good
metrics,
which
is
nice,
but
the
feature
value
is
much
more
variable
because
their
arrangement
changes
more.
15
16. And
will
have
similar
qualiMes
as
long
as
the
control
scheme
&
playability
are
similar.
Run
is
a
fast-‐paced
endless
runner
with
instant
respawn
that’s
been
a
viral
traffic
juggernaut
on
Kongregate
web.
When
it
launched
on
Android
on
December
we
seeded
it
with
some
promoMon
on
Kongregate
and
it
went
viral,
hiYng
1.5M
downloads
without
any
further
markeMng
or
features.
There’s
no
easy
way
to
tell
how
much
was
exisMng
fans
picking
up
the
Mtle
or
new
viral
spread,
though
I
suspect
a
bit
of
both.
But
I
think
that
building
a
fan
base
on
web
and
then
using
that
to
help
launch
a
game
on
mobile
is
a
very
underrated
strategy.
Bloons
Tower
Defense
5
has
been
in
the
top
100
grossing
charts
for
more
than
a
year
with
no
markeMng
fueled
by
the
huge
audience
they
built
for
that
series
on
the
web
over
the
last
5
years.
16
17. Bloodrealm
is
a
CCG
from
Making
Fun,
with
the
same
gameplay
and
features
between
the
web
version
and
mobile
though
with
a
lifle
bit
of
lag.
A_er
a
substanMal
beta
period
on
Kong
(which
allowed
them
to
make
substanMal
improvements
in
metrics)
the
game
was
pushed
broadly
on
both
Kongregate
&
iOS
in
November.
While
day
to
day
there’s
variaMon
the
average
ARPDAU
between
the
two
plaDorms
is
idenMcal
at
$0.19.
Now
there
are
two
caveats
to
this
data:
1)
developers
have
generally
reported
Kongregate
LTVs
as
2-‐3x
that
of
other
plaDorms
such
as
Facebook,
so
comparison
between
the
FB
version
and
a
mobile
version
would
probably
reflect
a
similar
difference.
2)
ARPDAU
on
its
own
can
be
a
decepMve
stat
because
two
games
with
similar
ARPDAUs
but
different
retenMon
rates
will
have
very
different
LTVs.
In
this
case
however
I
can
confirm
that
the
LTV
by
cohort
is
also
virtually
idenMcal
between
the
two
plaDorms,
and
thus
retenMon
as
well.
And
finally
if
you’re
wondering
why
I’m
using
a
stat
I
don’t
care
for
like
ARPDAU
it’s
because
I
can
safely
talk
about
it
and
downloads
without
revealing
our
total
revenue,
which
as
part
of
a
public
company
I
have
to
be
careful
about,
while
sMll
managing
to
share
relevant
informaMon.
17
18. In
fact
more
than
half
of
our
revenue
is
coming
from
Android
right
now
because
our
top
performing
game,
Tyrant,
has
done
so
well
there.
18
19. Here’s
Tyrant’s
breakdown
of
revenue
by
device.
Overall
our
revenue
is
closer
to
50/50
because
other
games
have
done
befer
on
iOS.
One
factor
besides
the
heavy
Google
features
for
Tyrant
may
be
that
it’s
more
popular
on
phones
than
tablets
for
some
reason,
perhaps
related
to
genre.
Lionheart
TacMcs,
which
we
launched
two
weeks
ago
a_er
an
3
months
in
test
markets,
gets
50-‐55%
of
it’s
revenue
from
iPad,
and
Tyrant
has
always
ranked
lower
on
the
iPad
grossing
charts
than
the
iPhone.
19
20. So
let’s
dig
into
the
comparaMve
#s
by
plaDorm.
Here’s
a
chart
of
Tyrant’s
ARPDAU
since
global
launch
on
each
plaDorm.
On
average
Android
ARPDAU
runs
10-‐15%
lower
than
iOS
except
in
the
mid-‐November
to
early
January
periods
when
there
was
a
big
influx
of
new
traffic
on
Android
from
Google
features.
But
the
overall
#s
are
actually
masking
a
lot,
as
the
traffic
mix
is
quite
different
between
plaDorms.
The
majority
of
our
traffic
on
iOS
has
come
through
paid
user
acquisiMon
and
Gamestop
promoMon
in
North
America
&
Western
Europe
while
on
Android
our
traffic
is
more
geographically
mixed
since
the
majority
came
through
features.
20
21. Here’s
a
breakdown
of
our
installs
and
revenue
by
country.
Installs
are
fairly
diverse
but
revenue
is
completely
dominated
by
English-‐speaking
countries,
Western
Europe,
and
Scandinavia
though
Russia
is
also
fairly
strong.
21
22. Here
is
the
ARPDAU
by
plaDorm
for
just
US
traffic
–
outside
the
heavy
feature
traffic
periods
Android
ARPDAU
is
very
close
and
o_en
befer
than
iOS.
22
23. In
Germany
Android
ARPDAU
is
usually
befer
than
iOS,
parMcularly
in
the
last
few
months.
23
24. Whereas
the
moneMzaMon
of
our
Russian
Android
traffic
is
quite
low
relaMve
to
both
Russian
iOS
ARPDAU
and
US
or
German
Android
traffic.
24
25. What’s
going
on
becomes
clearer
when
you
dig
into
the
devices.
Most
of
the
German
devices
are
high-‐end
Samsung
Galaxy
phones,
whereas
Russian
devices
are
mostly
fragmented,
lower-‐end
devices.
25
26. Devices
are
a
prefy
good
proxy
for
demographics:
people
with
new,
high-‐end
devices
tend
to
be
more
wealthy
and
willing
to
spend
than
those
with
older
or
cheaper
devices
–
the
iPhone
5s
has2x
the
ARPDAU
of
the
4s.
Children
are
more
likely
to
be
using
iPods.
But
aside
from
the
demographics
of
the
users:
a
lot
of
the
older
devices
may
just
not
play
the
game
very
well.
26
27. Everyone
talks
about
Android
fragmentaMon,
and
it
is
very
real.
But
it’s
been
7
years
since
the
iPhone
launched
and
4
since
the
iPad
and
as
you
can
tell
from
this
chart
there
are
a
lot
of
different
iOS
devices
out
there,
too,
with
different
screen
sizes,
resoluMons,
and
levels
of
processing
power,
and
since
nearly
every
model
sold
well
they’re
sMll
a
lot
of
the
older
devices
in
use.
It’s
easy
to
hit
performance
problem,
crashes
and
instability
from
too
many
high-‐res
assets
or
3D
effects
on
lower-‐end
devices.
This
chart
is
the
tutorial
funnel
for
Lionheart
TacMcs
early
in
its
test
market
period.
It’s
a
beauMful
3D
game
that
looks
fabulous
on
reMna
devices
but
as
you
can
see
from
the
chart
had
very
sharp
drop-‐offs
on
older
iPhones
and
iPod
touches.
The
developer
(Emerald
City
Games)
was
able
to
opMmize
for
the
lower-‐end
devices
by
using
different
menus
and
textures
there,
and
removing
lightmaps
and
other
visual
effects
during
bafles
but
managing
all
these
different
version
has
been
more
challenging
and
Mme-‐consuming
than
they
expected.
27
28. While
the
device
fragmentaMon
on
Android
is
worse
than
iOS
the
tools
they
give
you
to
manage
it
are
much,
much
befer.
Detailed
crash
&
freeze
reports
are
extremely
helpful
in
diagnosing
and
fixing
issues
and
if
there
are
devices
that
you
just
can’t
support
you
can
specifically
blacklist
them.
You
can
also
block
downloads
based
on
other
criteria,
such
as
screen
size,
which
is
very
helpful
in
blocking
the
proliferaMon
of
low-‐end
devices
in
Southeast
Asia.
Being
able
to
push
beta
builds
through
the
developer
console
and
then
test
them
as
if
they
were
live
is
invaluable,
second
only
being
able
to
push
a
build
live
to
users
whenever
you
want
and
need
to.
28
29. Unfortunately
the
google
transacMon
APIs
are
not
as
user
friendly.
A
good
example
is
that
item
prices
are
returned
with
a
currency
symbol
rather
than
a
currency
code,
so
it’s
impossible
to
disMnguish
something
charged
in
US
dollars,
Canadian
dollars,
or
Mexican
pesos,
which
all
use
the
same
sign.
We
work
around
this
by
puYng
the
price
in
the
item
name
and
parsing
it
from
there
but
it’s
a
bit
of
a
pain
to
manage.
What’s
more
than
a
pain
is
that
Google
is
very
slow
to
verify
purchases
and
overzealous
in
their
idenMficaMon
of
fraud,
especially
on
higher-‐priced
items.
This
causes
customer
frustraMon
and
forces
you
to
ping
their
servers
to
check
all
the
users
transacMons
every
Mme
they
enter
the
game
to
make
sure
they’re
granted
what
they’ve
bought.
29
30. The
receipt
verificaMon
I
menMoned
on
the
last
slide
is
really
important
and
this
graph
shows
why.
It’s
the
first
two
weeks
of
iOS
revenue
reported
by
Sheep
Happens,
a
wacky
endless
runner
we
launched
last
fall,
which
at
the
Mme
was
not
checking
Apple’s
servers
that
a
purchase
was
valid
before
granMng
the
currency
and
many
users
with
jailbroken
phones
took
advantage
of
the
omission.
It
turns
out
when
IAP
is
free
demand
is
very
high!
While
the
game
did
decently
actual
revenue
was
of
course
a
fracMon
of
the
numbers
shown
here.
I
promised
you
real
numbers:
these
are
simultanously
real
and
very
inaccurate.
Nearly
all
of
our
developers
had
some
degree
of
trouble
implemenMng
receipt
verificaMon,
either
in
the
actual
receipt
verificaMon
like
Sheep
Happens
or
incorporaMng
it
correctly
in
analyMcs
calls
and
it
wasted
a
bunch
of
Mme
for
everybody.
We’ve
baked
it
into
our
SDK
now
so
developers
don’t
have
to
deal
with
it
but
it
sMll
causes
legacy
problems
for
some
games,
like
Sheep
Happens,
because
players
aren’t
forced
to
update
so
old
client
versions
conMnue
to
send
bad
data.
30
31. I
did
a
talk
at
GDC
Next
last
fall
about
data
problems
and
piDalls
and
used
this
image
to
talk
about
how
under
the
surface
data
is
o_en
a
steaming
pile
of
corrupted,
inaccurate
shit.
That’s
true
for
web
data
but
it’s
twice
as
true
for
mobile
for
a
several
major
reasons
that
are
interrelated:
connecMon
issues,
client
unreliablity
and
client
fragmentaMon.
On
the
web
there
are
just
a
few
major
clients,
aka
browsers,
that
provide
data
in
a
standardized
way,
events
nearly
always
occur
online
so
you
can
rely
on
your
own
server
for
Mmestamps,
and
your
biggest
headaches
come
from
idenMfying
people
uniquely.
On
mobile
you
need
to
rely
on
the
client
much
more,
both
because
important
events
occur
offline
and
because
the
client
itself
is
crucial
informaMon.
But
even
something
as
basic
as
the
Mmestamp
can
be
reported
very
differently
from
plaDorm
to
device
to
region.
We
recently
had
problems
with
the
data
from
Lionheart
TacMcs
because
a
bunch
of
clients
from
Southeast
Asia
were
reporMng
the
date
in
the
Buddhist
era
calendar,
where
it’s
2557
instead
of
2014
and
if
you
don’t
have
the
right
Mme
for
events
you’re
screwed.
31
32. We’ve
built
our
own
SDK
for
developers
to
use
on
games
we’re
publishing
to
take
advantage
of
Kongregate
logins
and
badges
and
other
features,
and
have
now
rolled
analyMcs
into
it,
but
Swrve,
Leanplum,
Kontagent
and
many
other
commercial
services
will
do
a
much
befer
job
than
you’ll
do
on
your
own.
32
33. It’s
tempMng
to
instrument
your
game
from
the
start
with
everything
you
think
you
might
want
to
know
but
you’re
not
likely
to
get
it
all
right,
it
will
be
overwhelming
to
QA,
and
expensive
to
store,
which
is
especially
frustraMng
if
you
know
it’s
garbage.
We’ve
moved
to
a
staged
approach
with
analyMcs
implementaMons
where
we
have
developers
implement
the
most
crucial
stuff:
player
idenMficaMon,
sessions,
and
transacMons
(crawl).
Once
that
is
QAed
move
on
to
basic
game-‐specific
informaMon
like
tutorial
compleMon,
level
progress,
win
rates,
and
pvp
parMcipaMon
which
are
the
most
acMonable.
Finally
once
that’s
solid
it’s
Mme
to
run
with
the
really
detailed
informaMon
if
the
game
is
deep
enough
to
warrant
it.
Endless
runners,
ironically,
probably
only
need
crawl/walk
whereas
that
last
stage
of
data
is
very
important
in
most
mulMplayer
games.
33
34. SomeMmes
the
hardest
thing
to
know
is
that
there
IS
a
problem,
but
if
you’re
cross-‐
checking
across
mulMple
sources
they
are
easier
to
find.
We’re
constantly
cross-‐
checking
our
internal
analyMcs
against
AppAnnie
and
Ad-‐X,
which
we
use
for
markeMng
tracking,
and
will
soon
add
Swrve
which
will
add
another
data
point.
34
35. This
is
something
that
came
up
for
us
in
part
because
of
the
gap
between
a
publisher’s
commitment
to
analyMcs
and
the
developer
–
some
of
our
developers
are
as
into
analyMcs
as
we
are,
but
some
have
been
implemenMng
the
schema
more
for
us
than
for
them.
The
result
is
skimpy
tesMng
on
their
side
and
in
our
hurry
we
haven’t
always
QAed
enough
either.
On
several
games
we
launched
into
test
market
with
fundamentally
bad
data
and
it
was
a
waste
of
both
Mme
and
markeMng
$s.
Everything
doesn’t
have
to
be
perfect
yet
but
you
need
at
least
the
basics
for
retenMon,
revenue,
and
player
progress
tracking.
35
36. A
game
with
bad
data
is
a
black
box.
You
might
have
a
great
game,
and
it
doesn’t
mafer.
More
likely
you
have
a
game
that
could
be
befer
but
without
data
it’s
hard
to
know
even
what
direcMon
to
go.
Now
I’m
not
in
any
way
in
the
school
that
thinks
you
should
be
A/B
tesMng
every
bufon
color
–
for
games
to
have
soul
and
to
innovate
you
need
to
look
beyond
data.
But
data
is
crucial
in
diagnosing
what
is
and
isn’t
working,
generaMng
theories
of
why,
and
eventually
confirming
if
your
changes
had
the
effect
you
expected.
Sheep
Happens
was
a
black
box
–as
you
saw
earlier
the
data
from
the
game
was
quite
messy,
and
we
launched
with
them
only
parMally
fixed
and
a
hazy
view
of
the
metrics
and
issues
with
the
game.
We’ve
done
our
best
guessing
at
what
needed
improvement
but
with
so
much
corrupt
data
in
the
system
it’s
difficult
to
tell
if
it
has
had
much
effect.
Unfortunately
the
answer
is
probably
not.
36
37. And
you
can
really
improve
your
game
over
Mme.
Here’s
Tyrant’s
ARPDAU
chart
again
which
you
can
see
increased
dramaMcally
over
Mme
even
with
conMnuing
influxes
of
new
traffic.
Since
they
launched
last
August
they’ve
pushed
nearly
20
builds,
nearly
all
with
various
fixes
for
UI
issues,
performance
across
devices,
and
bug
along
with
constant
tweaking
of
the
matchmaking
algorithm,
all
working
to
improve
the
base
player
experience.
Along
with
that
every
month
or
so
Synapse
has
done
a
release
of
a
major
new
system
to
increase
the
depth
of
the
game,
parMcularly
the
late
game.
Those
features
don’t
necessarily
drive
moneMzaMon
themselves,
but
they
increase
player
engagement
and
retenMon
without
which
there
can
be
no
moneMzaMon.
What
drives
the
revenue
spikes
you
see
is
the
regular
release
of
new
content,
parMcularly
of
limited
Mme
events
with
rewards,
which
are
Med
to
the
release
of
new
gacha
boxes
and
occasional
special
offers.
37
38. Last
summer
when
we
first
started
launching
games
we
were
expecMng
to
spend
4,
maybe
6
weeks
in
test
markets
and
we
pushed
our
first
two
games,
Tyrant
and
Sheep
Happens
in
that
Mme
frame.
We
had
some
internal
deadlines
that
we
needed
to
hit
–
the
games
needed
to
be
out
on
both
iOS
&
Android
by
the
beginning
of
September
to
get
significant
promoMon
from
Gamestop
because
once
GTA
V
and
the
new
consoles
started
launching
there
wouldn’t
be
an
opportunity
again
unMl
a_er
Christmas.
For
Tyrant
this
was
the
right
call
–
while
the
game
was
missing
some
features
that
we
knew
were
going
to
be
important
long-‐term,
like
guilds,
and
that
the
UI
and
matchmaking
needed
more
tweaking
the
game
was
in
fundamentally
good
shape
with
solid
data
and
metrics.
We
didn’t
get
the
fullest
possible
value
from
our
early
Apple
feature
but
that
was
more
than
made
up
for
by
the
value
we
got
from
Gamestop
promoMon.
But
Sheep
Happens
was
a
black
box
that
wasn’t
ready.
The
game
did
okay
anyway,
and
has
been
profitable
for
both
us
and
the
developer,
but
the
value
of
the
heavy
Apple
features
we
secured
for
it
at
launch
were
likely
much,
much
less
than
they
could
have
been.
Holding
a
game
longer
in
test
markets
definitely
has
some
costs,
and
the
pros
and
cons
need
to
be
weighed
carefully.
If
you’re
not
going
to
get
a
launch
feature
or
38
39. Staying
longer
in
test
markets
fits
in
with
something
I
like
to
say,
which
is
that
things
are
a
marathon,
not
a
sprint.
It’s
definitely
true
that
free-‐to-‐play
games
are
a
marathon.
Good
games
can
grow
and
maintain
revenue
for
years,
as
long
as
the
systems
keep
the
players
engaged
and
the
developer
can
keep
adding
content
and
events.
Tyrant
started
solid,
but
a
lifle
slow,
and
at
launch
both
Synapse
and
us
were
a
lifle
disappointed
since
we
thought
the
game
was
capable
of
more.
Over
8
months
the
game
has
come
to
meet
and
exceed
our
expectaMons,
hiYng
top
50
grossing
in
the
US
iPhone
charts
just
last
week.
A
crucial
part
of
that
long-‐term
growth
story
is
the
relentless
pace
at
which
Synapse
worked
pushing
content
and
changes,
making
mistakes,
learning
from
them,
fixing
them,
pushing
for
the
kind
of
LTVs
that
make
paid
UA
profitable,
even
for
a
niche
game
with
a
grify
theme.
You
can’t
sprint
forever,
and
since
January
Synapse
has
sefled
into
a
steadier,
more
sustainable
rhythm
but
the
type
of
drive
and
speed
they’ve
shown
is
something
we
now
look
for
in
every
team
we
sign,
because
the
rate
at
which
you
improve
a
game
mafers
as
much
to
its
ulMmate
success
as
the
iniMal
quality.
39