How to use data to make a hit TV show Sebastian Wernicke
Roy Price is a man that most of you
have probably never heard about,
even though he may have been responsible
for 22 somewhat mediocre
minutes of your life on April 19, 2013.
He may have also been responsible
for 22 very entertaining minutes,
but not very many of you.
And all of that goes back to a decision
that Roy had to make
about three years ago.
So you see, Roy Price
is a senior executive with Amazon Studios.
That’s the TV production
company of Amazon.
He’s 47 years old, slim, spiky hair,
describes himself on Twitter
as “movies, TV, technology, tacos.”
And Roy Price has a very responsible job,
because it’s his responsibility
to pick the shows, the original content
that Amazon is going to make.
And of course that’s
a highly competitive space.
I mean, there are so many
TV shows already out there,
that Roy can’t just choose any show.
He has to find shows
that are really, really great.
So in other words, he has to find shows
that are on the very right end
of this curve here.
So this curve here
is the rating distribution
of about 2,500 TV shows
on the website IMDB,
and the rating goes from one to 10,
and the height here shows you
how many shows get that rating.
So if your show gets a rating
of nine points or higher, that’s a winner.
Then you have a top two percent show.
That’s shows like “Breaking Bad,”
“Game of Thrones,” “The Wire,”
so all of these shows that are addictive,
whereafter you’ve watched a season,
your brain is basically like,
“Where can I get more of these episodes?”
That kind of show.
On the left side, just for clarity,
here on that end,
you have a show called
“Toddlers and Tiaras” –
(Laughter)
– which should tell you enough
about what’s going on
on that end of the curve.
Now, Roy Price is not worried about
getting on the left end of the curve,
because I think you would have to have
some serious brainpower
to undercut “Toddlers and Tiaras.”
So what he’s worried about
is this middle bulge here,
the bulge of average TV,
you know, those shows
that aren’t really good or really bad,
they don’t really get you excited.
So he needs to make sure
that he’s really on the right end of this.
So the pressure is on,
and of course it’s also the first time
that Amazon is even
doing something like this,
so Roy Price does not want
to take any chances.
He wants to engineer success.
He needs a guaranteed success,
and so what he does is,
he holds a competition.
So he takes a bunch of ideas for TV shows,
and from those ideas,
through an evaluation,
they select eight candidates for TV shows,
and then he just makes the first episode
of each one of these shows
and puts them online for free
for everyone to watch.
And so when Amazon
is giving out free stuff,
you’re going to take it, right?
So millions of viewers
are watching those episodes.
What they don’t realize is that,
while they’re watching their shows,
actually, they are being watched.
They are being watched
by Roy Price and his team,
who record everything.
They record when somebody presses play,
when somebody presses pause,
what parts they skip,
what parts they watch again.
So they collect millions of data points,
because they want
to have those data points
to then decide
which show they should make.
And sure enough,
so they collect all the data,
they do all the data crunching,
and an answer emerges,
and the answer is,
“Amazon should do a sitcom
about four Republican US Senators.”
They did that show.
So does anyone know the name of the show?
(Audience: “Alpha House.")
Yes, “Alpha House,”
but it seems like not too many of you here
remember that show, actually,
because it didn’t turn out that great.
It’s actually just an average show,
actually – literally, in fact, because
the average of this curve here is at 7.4,
and “Alpha House” lands at 7.5,
so a slightly above average show,
but certainly not what Roy Price
and his team were aiming for.
Meanwhile, however,
at about the same time,
at another company,
another executive did manage
to land a top show using data analysis,
and his name is Ted,
Ted Sarandos, who is
the Chief Content Officer of Netflix,
and just like Roy,
he’s on a constant mission
to find that great TV show,
and he uses data as well to do that,
except he does it
a little bit differently.
So instead of holding a competition,
what he did – and his team of course –
was they looked at all the data
they already had about Netflix viewers,
you know, the ratings
they give their shows,
the viewing histories,
what shows people like, and so on.
And then they use that data to discover
all of these little bits and pieces
about the audience:
what kinds of shows they like,
what kind of producers,
what kind of actors.
And once they had
all of these pieces together,
they took a leap of faith,
and they decided to license
not a sitcom about four Senators
but a drama series about a single Senator.
You guys know the show?
(Laughter)
Yes, “House of Cards,” and Netflix
of course, nailed it with that show,
at least for the first two seasons.
(Laughter) (Applause)
“House of Cards” gets
a 9.1 rating on this curve,
so it’s exactly
where they wanted it to be.
Now, the question of course is,
what happened here?
So you have two very competitive,
data-savvy companies.
They connect all of these
millions of data points,
and then it works
beautifully for one of them,
and it doesn’t work for the other one.
So why?
Because logic kind of tells you
that this should be working all the time.
I mean, if you’re collecting
millions of data points
on a decision you’re going to make,
then you should be able
to make a pretty good decision.
You have 200 years
of statistics to rely on.
You’re amplifying it
with very powerful computers.
The least you could expect
is good TV, right?
And if data analysis
does not work that way,
then it actually gets a little scary,
because we live in a time
where we’re turning to data more and more
to make very serious decisions
that go far beyond TV.
Does anyone here know the company
Multi-Health Systems?
No one. OK, that’s good actually.
OK, so Multi-Health Systems
is a software company,
and I hope that nobody here in this room
ever comes into contact
with that software,
because if you do,
it means you’re in prison.
(Laughter)
If someone here in the US is in prison,
and they apply for parole,
then it’s very likely that
data analysis software from that company
will be used in determining
whether to grant that parole.
So it’s the same principle
as Amazon and Netflix,
but now instead of deciding whether
a TV show is going to be good or bad,
you’re deciding whether a person
is going to be good or bad.
And mediocre TV, 22 minutes,
that can be pretty bad,
but more years in prison,
I guess, even worse.
And unfortunately, there is actually
some evidence that this data analysis,
despite having lots of data,
does not always produce optimum results.
And that’s not because a company
like Multi-Health Systems
doesn’t know what to do with data.
Even the most data-savvy
companies get it wrong.
Yes, even Google gets it wrong sometimes.
In 2009, Google announced
that they were able, with data analysis,
to predict outbreaks of influenza,
the nasty kind of flu,
by doing data analysis
on their Google searches.
And it worked beautifully,
and it made a big splash in the news,
including the pinnacle
of scientific success:
a publication in the journal “Nature.”
It worked beautifully
for year after year after year,
until one year it failed.
And nobody could even tell exactly why.
It just didn’t work that year,
and of course that again made big news,
including now a retraction
of a publication
from the journal “Nature.”
So even the most data-savvy companies,
Amazon and Google,
they sometimes get it wrong.
And despite all those failures,
data is moving rapidly
into real-life decision-making –
into the workplace,
law enforcement,
medicine.
So we should better make sure
that data is helping.
Now, personally I’ve seen
a lot of this struggle with data myself,
because I work in computational genetics,
which is also a field
where lots of very smart people
are using unimaginable amounts of data
to make pretty serious decisions
like deciding on a cancer therapy
or developing a drug.
And over the years,
I’ve noticed a sort of pattern
or kind of rule, if you will,
about the difference
between successful
decision-making with data
and unsuccessful decision-making,
and I find this a pattern worth sharing,
and it goes something like this.
So whenever you’re
solving a complex problem,
you’re doing essentially two things.
The first one is, you take that problem
apart into its bits and pieces
so that you can deeply analyze
those bits and pieces,
and then of course
you do the second part.
You put all of these bits and pieces
back together again
to come to your conclusion.
And sometimes you
have to do it over again,
but it’s always those two things:
taking apart and putting
back together again.
And now the crucial thing is
that data and data analysis
is only good for the first part.
Data and data analysis,
no matter how powerful,
can only help you taking a problem apart
and understanding its pieces.
It’s not suited to put those pieces
back together again
and then to come to a conclusion.
There’s another tool that can do that,
and we all have it,
and that tool is the brain.
If there’s one thing a brain is good at,
it’s taking bits and pieces
back together again,
even when you have incomplete information,
and coming to a good conclusion,
especially if it’s the brain of an expert.
And that’s why I believe
that Netflix was so successful,
because they used data and brains
where they belong in the process.
They use data to first understand
lots of pieces about their audience
that they otherwise wouldn’t have
been able to understand at that depth,
but then the decision
to take all these bits and pieces
and put them back together again
and make a show like “House of Cards,”
that was nowhere in the data.
Ted Sarandos and his team
made that decision to license that show,
which also meant, by the way,
that they were taking
a pretty big personal risk
with that decision.
And Amazon, on the other hand,
they did it the wrong way around.
They used data all the way
to drive their decision-making,
first when they held
their competition of TV ideas,
then when they selected “Alpha House”
to make as a show.
Which of course was
a very safe decision for them,
because they could always
point at the data, saying,
“This is what the data tells us.”
But it didn’t lead to the exceptional
results that they were hoping for.
So data is of course a massively
useful tool to make better decisions,
but I believe that things go wrong
when data is starting
to drive those decisions.
No matter how powerful,
data is just a tool,
and to keep that in mind,
I find this device here quite useful.
Many of you will …
(Laughter)
Before there was data,
this was the decision-making
device to use.
(Laughter)
Many of you will know this.
This toy here is called the Magic 8 Ball,
and it’s really amazing,
because if you have a decision to make,
a yes or no question,
all you have to do is you shake the ball,
and then you get an answer –
“Most Likely” – right here
in this window in real time.
I’ll have it out later for tech demos.
(Laughter)
Now, the thing is, of course –
so I’ve made some decisions in my life
where, in hindsight,
I should have just listened to the ball.
But, you know, of course,
if you have the data available,
you want to replace this with something
much more sophisticated,
like data analysis
to come to a better decision.
But that does not change the basic setup.
So the ball may get smarter
and smarter and smarter,
but I believe it’s still on us
to make the decisions
if we want to achieve
something extraordinary,
on the right end of the curve.
And I find that a very encouraging
message, in fact,
that even in the face
of huge amounts of data,
it still pays off to make decisions,
to be an expert in what you’re doing
and take risks.
Because in the end, it’s not data,
it’s risks that will land you
on the right end of the curve.
Thank you.
(Applause)