The science behind personalising streaming experiences with AI

In this episode of StreamTime Sports, co-hosts Nick Meacham and Chris Stone are joined by Richard Kearney, Senior Director at MediaKind, to dive deep into the evolving world of subscriber acquisition and retention. Drawing on his experience at Paramount and Showtime, Richard shares how AI and machine learning are redefining content recommendations, user engagement, and platform growth.

The conversation dives into how sports and entertainment streamers can balance personalisation with discovery, reduce churn, and maximise subscriber lifetime value.

 

Key Points:

• How is AI revolutionising content recommendations and enhancing personalisation for streaming services?
• What role does improved content discovery play in boosting retention and engagement?
• Can too much personalisation actually harm user experience or business growth?
• How are sports platforms using AI to analyse content and enhance fan engagement in real time?
• What AI-driven strategies are helping platforms navigate market challenges and stay competitive?

Episode Transcript

(00:04):
Hello everyone, Welcome back to the next episode of String Time
Sports. My name is Chris Stone, I'm the
community lead, joined as alwaysby our CEO, Nick Beacham.
Today we're joined by Richard Kearney, who's the Senior
Director of Product Management at Mediacon.
And today we're going to be talking about customer
acquisition and retention. So the first question I'm going
to ask, and I'll answer because I haven't prepped you guys for
this one, is we're talking about, you know, subscriptions

(00:25):
and churn and all that. How many services are you
currently subscribed to and whatis the most recent one that you
churn from? So I'll answer the question
first to give you guys some timeto think about it.
I'm currently subscribed to Disney Plus, Prime, NFL Game
Pass, Apple TV, Spotify and The Athletic.
And I think those might be the six subscription services that

(00:48):
I'm currently subscribed to thatI can think of on top of my
head. And the one I've most recently
turned from outside of the sports world, Nick, because I'm
expecting a child soon, is I have cancelled my Xbox Game Pass
subscription partially because I'm anticipating not having much
time to play games. They're also having a price
increase. So yeah, I think I'm currently
subscribing. I think I listed off about 6-6

(01:09):
there. You know some of them TV and
media, some of them, you know The Athletic for the editorial
side. You know Spotify for my music.
But yeah Nick, I think it's 6 services I'm subscribed to.
You know how many you're? Subscribed to, I'm not exactly
sure the number, but I think it's even less than that.
I would say it's about four or five.
Most of the ones you said, I don't think I currently have a

(01:29):
Disney one, but I have an AmazonPrime which replaces that.
I've got Spotify and I've got a couple of others.
I, I get actually the athletic, but I get it through randomly
through my Revolute accounts. I get a, there's a premium tier
that I get that for free, plus afew other things that make it
worthwhile. So that was an interesting and

(01:50):
random addition when I was when I was living overseas in Jamaica
that I had to use Revolute as the means of transferring and
exchange and oh, but happened toget access to all these awesome
premium subscriptions that I wasalready using.
So it was a great way to definitely, we're going to be
talking about retention here. It definitely kept me using them
as a result of that because it was too good an offer not to

(02:11):
refuse. So now about about the full
mark, I think. Yeah, yeah, yeah, Rich.
I think I've got from a, from a consumer streaming service
perspective, I think I've got pretty much all of them from
Prime Video to Paramount Plus toNetflix, HBO Max, which is now
HBO Max again and Hulu, yeah, basically all of them.

(02:34):
And then some of the more niche ones like I had master Class,
which my wife and I both had bigfans of.
And then on the music side, I'vegot Apple Music.
You know, I don't think I don't have any of the sort of like
virtual MVPD stuff, though I've considered that for some of the,
the sports piece. But you know, to, to your note
there, Nick, of like the ones that come included with certain
things. I have no idea which of those

(02:56):
I'm actually paying for directlyor I'm getting from my like
Disney Plus. Was it through Verizon or is it
not through Verizon now? I don't remember.
I might be getting double billedfor all I know.
And then some of them are ones that I got for free as an
employee at previous previous companies and they're not very
good at auditing and removing. That's it.
So in some case, I think I'm I'mstill gliding off of a

(03:18):
subscription I got for free 10 years ago.
They just never showed off. And I, and I've got a few where
I've kept them from what since I've moved two countries and
I've tried to change them to getat least in the, the local
currency and I couldn't, that wasn't going to happen for me
easily. So I, I've kind of given up on
that a little bit. At some stage I'll get the
wherewithal to commit to going the whole cancellation process

(03:39):
then trying to resubscribe. But it's, they make, they don't
make it easy for us, but there'sobviously a little bit of intent
in that obviously. So that's something we'll dig
into a little bit today. Yeah, one thing.
And even Nick, I have a story like yours.
I originally had Apple TV because it came for the first
year for free when I purchased anew iPhone.
And at the time it was only like299 per month.
And I was like, actually, if anyone's not on Apple TV, they

(04:00):
have great TV shows. But it's a bit like I had it for
free for a year through my phoneand then I just just let it roll
and just haven't touched it for the last four years.
Sounds like you've been caught in the the old, the classic
subscription tactic there as well.
All the things that were a few, I, I, we basically we, we
followed the trend that that allbasically good marketers want us

(04:21):
to follow. So there's maybe some proof in
the pudding after all. But let's let's jump into it,
Chris. Absolutely.
So yeah, Rich. I mean, today we're gonna talk
about subscriber acquisition andretention and we're going to go
through as well as some of the technology and the AI behind
that. And before we jump into that, I
think just a little bit of an interesting background from
yourself. I introduced you as working at
Mediacom, but before Mediacom, just for context, I think you've

(04:43):
only been there for about 3-4 months now.
You'd previously been at Paramount and Paramount of
obviously launched Paramount Plus and you've been around some
of those products. Maybe just give a little bit of
a background sort of what your awakening has been going from
working at, you know, the broadcaster, the distribution
side to now going to that service provider side and sort
of what some of that has been like that transition from a
career perspective or just a background perspective.

(05:06):
Yeah, No, of course. So I have been in product
management for almost 20 years now.
So essentially my entire career and all of that has been in
media and entertainment and the vast majority of it has vast
majority of that has been in video streaming and a lot of
that and subscription video streaming.
But all of my career prior to media kind was on the consumer

(05:27):
side of that. So for I worked for a bunch of
startups early on. But in terms of more household
names, people would know, I worked at AMC Networks for a
number of years where I launchedtheir first direct, direct to
consumer screening service yearsago.
And then for the last eight years Prior to joining
Mediaconda, I worked at the Paramount Corporation as you
mentioned there. And for the first from six years
of that, I worked at Showtime which was AUS based streaming

(05:50):
service. I know the brand name's known a
little bit more internationally,but as a some streaming
streaming service people subscribe to US based for those
six years. I was VP of product there for a
number of years running both theShowtime direct to consumer OTT
service as well as the Showtime Anytime TV everywhere service
where you authenticate through your MVPD or MSO subscription.
And then we made the decision tofold Showtime into Paramount

(06:13):
Plus about two or three years ago now.
And so I worked very heavily on the sort of trying to migrate
all those subscribers over so wedidn't lose revenue for that.
And then for the last two years,I worked at Paramount Plus where
I was a VP of Product in the content discovery and engagement
group. Now that is very relevant to
this conversation because that'svery much about both the two

(06:34):
sort of high level levers here. But one is personalization of
constant recommendations to end users, but also merchandising of
your most important IP that's most relevant to people.
And then finding a healthy balance between those things in
order to drive, drive content engagement, which reduces churn,
extends subscriber life. And now in my role at Media
Kind, I'm focused on our Mkio platform product, which takes

(06:58):
our sort of flagship MKIOAV processing pipeline product and
then builds on top of that as a,you know, direct to full on,
direct to consumer streaming platform like a white label
streaming service essentially. And what I'm bringing in my
focus there is all these lessonsI learned and all these tools
that I built places like Paramount to drive engagement
and reduce churn and trying to make that extensible to our

(07:20):
entire customer base. So just on, on that, Rich, could
you just explain for those that'd be interested to hear on
the NKIO standpoint, how does that plug into someone's perhaps
existing ecosystem? What?
How does that all play out when media con come to the come to
the full? Yes.
I mean, there's many different ways that can happen.
And sometimes it's a Greenfield situation where maybe it's a

(07:41):
sports league that has some rights and they're interested in
launching from scratch a new streaming service.
And certainly our product maps did very well to that use case.
Sometimes they have you know, what existing encoding and
process, you know when I say thefull sort of pipeline there,
right, like encoding, packaging,CDN, the DRM, all the aspects of
the video stack is that sort of core mkio piece.

(08:03):
And then when I say the platformwhere you're sort of adding on
top of that is there'll be entitlements to the content.
So we understand who users are and what they have access to in
terms of content in order to enforce that.
But that also drives the recommendation aspect and then
also content discovery. So the AP is that power, your
search experience, things that power when you go into your
homepage, what rows you're seeing, what the content is

(08:24):
inside of those rows. So all of those tools come
together to form what we're doing.
And we do that for both OTT customers and also we have a
long legacy of doing that for pay TV customers as well.
Yeah, I mean, that's sort of thethe the stack and it's either
people who are migrating from somewhere else or like I said,
Greenfield situations. And you know, like you just
talked about Greenfield situation, building it from

(08:45):
scratch. So maybe that's where we'll
start this podcast talking about, you know, you're building
a, you're building a streaming platform and you're kind of
going through. And we, we've had a similar
conversation with the previous guest who built a new platform,
talking about the acquisition phase and sort of what are some
of the important metrics or considerations you have,
particularly maybe when you're trying to define what success is

(09:05):
on that Because you're obviouslytrying to bring in new
subscribers, but also once you get them, you don't want to lose
them either. So maybe you, you think about
building that new platform, sortof the prioritisation, like you
got to get new people on to evenget it off the ground.
But also if you don't keep them on, well, you're going to have a
lot of work finding new subscribers.
So just sort of that initial phase when you're launching a

(09:25):
new product where you're balancing that at?
I have launched from scratch a tonne of products in my life.
I mean a lot of the acquisition piece is more of a more of a
marketing concern I would say. But whether you're launching
something from scratch or whether you're talking about a
mature service with free trial users, for example, the
challenges are largely the same,which is that you have very

(09:48):
little data about these users coming in right beyond maybe if
you have your, your, if you're instrumented correctly,
hopefully you understand what the acquisition piece of content
was that brought them in. Like what was the ad campaign?
What, what piece of content do they specifically see an ad
somewhere for that they click through to your service, for
example, and then making sure that follows them through their,

(10:08):
their onboarding experience in order to drive and, you know,
recommendations, things like that.
And I think of all of this, you know, just establish at a high
level my learnings, whether we're talking about conversion
or retention, is that really theonly way to reliably increase
those numbers is through increasing content engagement?

(10:29):
As obvious as that sounds like AB test after AB test across
years and years. The only way you can reliably do
that is by making your content recommendations better and or
having better content, obviously.
But that's a separate issue, getting people to watch more
content. And in the context of free
trials, that can be difficult because we have what we call the

(10:50):
old start problem, which is thatyou really know nothing about
what they what they'd watched previously.
So for example, one thing that we did at Paramount Plus is we
have what we called our content picker UX, which was a part of
the onboarding flow that quite literally asked people to
identify what programmes on our service they were interested in.
And obviously we lean towards pieces of content that were more

(11:11):
popular things on the service, you know, the Yellowstone spin
offs and and Halo when we had itand things like that, right?
Or different, you know, the the NFL on CBS, for example.
And then have them pick those things so that the very first
time that they hit the homepage,it would actually be a fully
sort of personalised homepage because it was driven by
constant recommendations relatedto the programming that they

(11:32):
literally just told us that theywere interested in watching.
And one of the good news things there is in which I think people
who might want to implement a strategy like that should think
about is that that can be a sortof annoying or difficult user
experience to go through on a TVbased user experience.
But it's pretty easy to do on a mobile app or on a mobile web or
like a web UX. And the good news is, is that

(11:53):
even though most of your consumption happens on TV based
devices, generally speaking, your acquisition, a lot of it
happens on that mobile and web interface.
And so it's entirely appropriateto just sort of have that
concept picker user experience there to have a good balance
between not slowing people down getting into the product, but at
the same time learning as much as you can in order to have
relevant recommendations once they get past that sort of

(12:13):
onboarding experience. So Rich, just I'm curious, the,
you talked about that journey between acquisition to be able
to then serve a personalised setof options.
How when I've talked to people, just general consumers, they,
they kind of get sometimes will get frustrated with like being

(12:33):
able to find the right content on a, on a, on a platform is how
robust is that, that capability now to be able to serve a
recommendation that is directly linked with what created the
acquisition in the 1st place andand how long has it taken,
sorry, how long has that been available, I suppose in working
effectively? So I think there's a couple of,

(12:57):
there's a couple of interesting topics here that you, you
touched on. You know, one in terms of a like
the user experience aspect. And I think this goes well
beyond streaming services. This is a general, a general
dynamic that people have complained about on the
Internet, particularly over the last few years.
We're, and I think this effects the tech industry on the whole

(13:22):
we're you know, as long as thereis growth to be had that is
somewhat organic, then it's fineto focus on the best user
experience possible without thinking too much about
optimising for juicing your key metrics as opposed to just
having noble ideas of great userexperience.
But as soon as you kind of like Netflix ran into this challenge

(13:44):
this way, you start seeing things like the password sharing
crackdown and things like that where and this affected the
industry on the whole. I guess it was almost three
years ago now where Netflix for the first time had you know a
quarterly they reported their quarterly numbers and they had
lost subscribers for the first time ever that quarter.
And that sort of kicked off the whole industry wide freak out
about focusing on revenue and profit and not just subscriber

(14:06):
growth because everyone have been juicing that taking it back
to our earlier conversation by including it in your Verizon
account, including it in your Revolut and everywhere else and
Wall Street kind of caught on tothat.
It freaked out a little bit. I remember that.
I remember that window of time very, very well where the whole
market turned on its head, Netflix price plummeted, thought
there was existential questions being asked across the entire

(14:28):
industry, and look where Netflixis now.
Yeah. No, they you know it, it pivoted
and I think a lot of the advertising growth with has had
a lot to do with the sports focus in the industry has been
about not just looking at that subscriber number, but looking
at that revenue number. And particularly sort of the
ARPU or like the average revenueper user became the metrics that
services started to look out more in with more intent

(14:50):
internally. Like certainly we were
monitoring that constantly a paramount plus of like, you
know, should we be more pushing people towards our premium ad
free tier or should we push themtowards that cheaper tier that
includes ads? Because depending on the the
season of the year or whatever we're tracking, sometimes that
ads plus the smaller subscription fee is actually
more average revenue per user than that premium ad free plan.

(15:12):
So people are constantly lookingto optimise that.
But the reason I brought that upin the context this user
experience question is because once you get to a certain level
where you're not going to get like everybody in America has
heard of Netflix at this point, It's not, it's not, it's not a
question of identifying net new people who have never had
Netflix. It's about deriving more revenue

(15:33):
from the existing population of people out there who know about
it all. And that's just an optimization
game. And a lot of that is this
recommendation stuff. But you know, what ends up
happening, and this is often a asubject of fierce debate
internally at companies, especially between the user
experience and design folks, is when you do things in the
product that seem like they're kind of anti user in a way.

(15:56):
Like a good example, this is thecontinue watching or resume
watching row that every service has, right?
Yeah, I think most people that Italked to, what they would
prefer is that that is the number one row on their homepage
all the time. They don't have to go looking
for it because of an end user perspective.
You're trying to get in there and get into a piece of content
and stop fiddling with your remote as soon as possible.

(16:17):
But there's a bunch of ADAB tests that you've been running
that show that if you push that thing down to row #5 and you
force them to be exposed to the content you want to recommend
and rows one through 4, the numbers bear out that that
person ends up watching more average minutes watch of content
per month than somebody you didn't do that too.
And so you end up in this thing where the priorities are a

(16:38):
little bit mixed there of what the user ostensibly wants versus
what is driving the business. What what they think they want
versus what actually, yeah, makes makes the the right
impact, which I think is a constant challenge in this
world. So what, Chris?
I'll let you take the reins. No, Nick, I love it.
You're you're in the zone right now.
You do your thing. I can see you're intrigued.

(17:00):
Well, I'm just curious because Rich, that example you gave is a
great example of how how deep wecan go into sort of curating
that experience. Just talk me through about, you
know, what's the role of AI playing in that today and how
you're seeing that impact Because everyone is everyone.
In the last 12 months alone, what we've seen across the
entire industry, everything we've been talking around with
sports media is AI is being brought up in almost every facet

(17:24):
of sports media. And actually I think only really
in the last few months I've seenit being talked about publicly
anyway from some of the big streaming players of its role in
creating a personalised experience.
Just talk us through what that looks like.
Yeah. So at A at a high level, I think
an important distinction to makethat really colours the way that
I look at all of this is that right?
AI is very much an umbrella term.

(17:44):
And a lot of people nowadays will use AI interchangeably with
generative AI slash large language models, slash
transformer models. But you have more, it's funny
now to call it traditional because it's only like 1015
years old, but more traditional machine learning driven models
for recommending content becauseyou know, services like Netflix.

(18:05):
And I mean, I launched machine learning driven personalization
on Showtime and I think like 2018 or something like that.
I don't think there were, you know, many years before anything
like ChatGPT existed. So when I look at the
opportunity with AII, look at itin the context of how it can
improve upon the traditional machine learning approach that
has been being optimised for many years by the Paramount

(18:28):
Pluses and Netflix. So the global streamers, let's
call them. And an important thing to
understand is how AI impacts it.There's sort of four different,
like we're talking specifically about content recommendations.
So content recommendations, there's basically four different
high level approaches to that without getting, I'm not going
to get into like specific types of models and algorithms or

(18:50):
whatever. But at a high level you've got,
you know, what we call content based filtering, which is
saying, OK, this piece of content is similar to this piece
of content. Maybe they're both westerns or
something. And so the person watched this
Western recommend this one. Then you have more collaborative
and well, the challenge with that with content based
filtering is it requires really good metadata about your content
because the better the metadata is that you have, the better you

(19:12):
can be an understanding similarity between content.
The second one is what we call collaborative filtering, which
is more users who watched this also watched this.
We don't necessarily care whether or not there's an
explicit tie between those things based on metadata, but
we've just seen patterns that users who watch this also watch
this. And then there's the hybrid

(19:32):
approach which combines both, which pretty much every global
streamer does some mix of that. And then sort of the 4th 1 is
more sort of global popularity for black of a better term.
Like what is everybody watching?What's trending content
essentially? And so now that we bring LLMS
in, they're able to and it's just specifically makes sense
for sports too. But it really helps sports in

(19:54):
certain ways that it's not that that sports has always been
traditionally a bit more challenging to understand.
So like that content similarity piece for example.
That enrichment of metadata years ago, you know, when I was
at Showtime, you know, we have whatever metadata about our
movies that we get from, let's say we're licencing a bunch of
movies from Warner Brothers or something like that.
Like they would licence us thosemovies and provide us whatever

(20:16):
base level metadata, what actorsare in a short description of
the movie. But there's a lot more metadata
about like the mood and all these different things that the
concept has that we would go offto a TiVo or a grace note or
somebody and lots of people do this and licence all that
additional rich metadata from them and then feed that into our
recommendation system. Nowadays with LLMS, you can do

(20:36):
metadata enrichment that's like,let's go out to the hive mind of
the Internet basically and enrich and pull in all this
metadata. Additionally, that traditionally
you'd have to get from, you know, a grace under a TiVo like,
and be able, it's not going to be as structured as that data,
right? But you're able to have better
metadata enrichment of the content.
So that's just like one example of how you can start to identify

(20:56):
and enrich your metadata and understand, let's call it like
semantic meaning. Cuz like traditionally when you
look at 2 pieces of content, youcan really only be like, OK,
other keyword matches essentially between the
descriptions, between the list of the actors or between the
genres. But this is able to say, OK,
what does this content really like mean?

(21:17):
What is it the story about? And then find similarities that
you would not find between content with traditional sort of
keyboard based matching and sortof just general metadata.
That's just one example, but I could go on for a very long time
of different AI use case examples.
Well, I think one of the things to be interesting you spoke a
little bit about this you and I beforehand, you know, AI kind of
the the engine behind personalization.

(21:39):
I think whether it is retention or even that initial acquisition
phase is the degree of personalization you can provide
for an individual user. But one of the things you kind
of discussed is the scalability of that personalization.
Like you could turn it all the way up and be incredibly
personalised, but is that necessarily kind of in your best
interest to do so? Sort of kind of like what is the

(22:02):
the level of, you know, letting people still organically
discover stuff versus, you know,turning up, you know, the AI too
high, you know, there any risk of being too personalised with
it. It's an interesting question,
and I think there's a. I'd like to know how far you go
with that. Like how how far is 2
personalised? Like yeah, but I'm intrigued to

(22:22):
see how you tackle that one Rich, because I'm a bit of miss
as to what 2 personalised might be.
I think there's people there's there's comfort level.
First, if you're asking my gut instinct here, my gut instinct
is, is that if you look at the numbers on things like TikTok or
Instagram, I think it, it's safeto say that you really can't go
too far. The personalization angle in

(22:44):
those contexts now greater thesecontexts are a little bit
different, right? We're not talking about solving
the problem of. So I'd say that at a high level,
I don't think you can go too personalised, but the reality is
you don't have an infinite pool of content in your content
library, right? So it's a, it's a different, the
more content you have, the more you're gonna get out of your

(23:06):
personalization approach, right?Cuz you can create a much
different service basically for each end user.
But when you have a limited amount of content, that's not
true. And also there's always the
challenge of, you know, something always came off.
I think of all the global streamers is that you've got

(23:26):
content that are your, your original pieces of content you
produce yourself and then you'vegot content that you licence
from somebody else. And so while the algorithm
traditionally won't care about the acquisition method of a
piece of content, it's just trying to find people, things
they want to watch. It might be more in your
interest from a financial perspective to get people to
watch content for the economics a little bit better for you than

(23:48):
the content that you licence from somebody else.
So like that's one dynamic that you need to manage.
There's also something that I'veheard anecdotally from like like
Netflix, for example, when they started doing original
programming for the first time, which came on later in their in
the life of their service. You know, you have fully
optimised home pages. But let's say some movie star,

(24:09):
I'm not going to name names, I've heard stories.
Some movie star has a brand new movie that was $100 million
budget that they've made for Netflix and they log in on the
Friday night, but the movie cameout and it's not on their
homepage. And they're like, what?
You know, like because that's just the nature.
They don't show things to peoplethey don't think they're going
to want to watch. So the data suggests that

(24:29):
whatever. So I think we've all gone
through that where we hear abouta show on Netflix or something
from a friend and you're like, that sounds perfect for me.
I've never even seen that on thehomepage.
And then you go look it up. You're like, that's weird, you
know, so you do need to balance,as I talked about very beginning
itself, but my role is ParamountPlus balancing sort of
merchandising versus personalization.

(24:50):
And two of the key tools that wehad for that there and which are
things that I'm I'm have on the road map now at at media kind is
you have sort of a pinning, A pinning approach, which is to
say, OK, for example, let's say you think the content resume,
the resume watching where I talked about before.
If you believe that that should be the top row always.

(25:13):
Or if you are a live sports person and you want to make sure
that any live game right now is always at the top of the page,
you can pin that and say, OK, let let the algorithm decide the
order of the rows for each individual user based on their
projected affinity for that content.
But make sure, regardless of therecommendations, it's not
allowed to not put that as the top roll on the page.
And the other thing we do is sort of a Max min, which is to

(25:35):
say like a maximum minimum position on the page.
So let's say your editorial team, you know, they put
together a whole thing of like movies related to Halloween.
And then we have a bunch of movies that we licence that have
to do with Halloween and they want people to watch them.
So let's put that row up and let's say that the floor for
that is 10. So it's not allowed to be lower
than row 10 on the homepage for any individual user because hey,

(25:58):
we're spending a bunch of marketing dollars out here
promoting and driving people in to watch that stuff.
We want to make sure they don't miss it when they come in,
right? So part of it is like that's one
way to make sure that your marketing spend is aligned in a
way that you're recommending system is not going to be good
at understanding. And so you have that ability to
override that in order to bettermerchandising of your content.

(26:18):
So, Rich, I mean, the, the idea of this whole ability to
personalise the experience people come on is obviously
really appealing. We're talking, we've been
talking a lot and using a lot ofexamples of the big global
streamers. Let's talk about how far can
that go in terms of the ability for someone who may be a
challenger platform or a sports property to be able to bring

(26:39):
that personalization to the the level that perhaps we're
talking. Is there a, is there a threshold
where that's there's only a limited amount of
personalization available at thelower end?
Or is it really transferable at all levels and worth doing I
guess, and worth the resource commitment to it as well?
So I sort of teased this out a little bit earlier, but didn't

(27:01):
speak to it at great length. But when I was making that
distinction of machine learning versus the new large language
model versions of AI, the most important thing, I gave metadata
enrichment as an example and that's one.
But there is multiple other waysin which AI tools now make it
more scalable and give an opportunity for companies to

(27:23):
kind of shortcut where they did not spend the last 10 years fine
tuning and machine learning algorithm about their content in
order to whatever. So, and the three different ways
I think that happens are 1 is curation.
So the curation of what you havethere, right and where curation
meets personalization because those things are fully hybrid at
this point. They're not separate concerns

(27:45):
necessarily of like here arose on the home page that an editor
came up with manually versus these are ones that an algorithm
came up with. There are, Netflix is a good
example here in terms of work they had to do manually that is
now accessible and scalable for other people, right?
Like the homepage is not just saying, hey, there's 20

(28:06):
different rows on the homepage. Every user sees the same 20 and
they're just in a different order.
It's instead saying we have a pool of maybe 1000 different
rows that could show up on the homepage ostensibly.
And then we're going to figure out the right subset for this
user based on what they've watched before.
But in order to have that pool, which not everybody's sitting on
what Netflix did 10 years ago, whatever was go through and

(28:27):
basically manually tag this content with what the mood was
like. These are shows that are set in
Paris. These are shows that are set in
Berlin. These are like every level of
little bit of data about the thing in order to then generate
like what I'll call micro genresessentially, But that was done
manually in LLM. Can look at the content in your

(28:47):
content database, identify that level of structured data of
different sort of fields about that content, and then suggest
an entire pool of different waysto curate your content, which
could essentially then just sortof review and go, yeah, I like
that one. And then there's that little bit
of control. Taking it back to your part
about how far can you go at personalization, people do think

(29:08):
want some degree of control. You know, you're a, in some
cases a public company with shareholders.
You're not OK with some relationships between content
that the LLM might might figure out and present that are a
little bit taboo from your perspective, you're not
comfortable with that, right? So it's about the curation piece
that AI is helping with to make it scalable for smaller
companies that don't have the same, you know, level of people

(29:30):
internally. And then also sort of the
context piece, which is an improvement over this is helpful
for sports content as well, where traditional models of
machine learning were not very good at understanding a real
time context. It's 7:00 PM on Saturday night.
Or they just went to the search page and typed in baseball 5

(29:51):
seconds ago. Or, you know, or like they
literally just finished washing this game.
But you know, the the this NFL football game with these teams
or whatever, right? Traditional models are more OK,
let's look with the user watch and every 24 hours, let's
increment the list of recommendations for them and on
the service versus in real time.Let's augment that data with

(30:12):
what do they literally just do. And it's very helpful, I think
for sports content where it's more of a real time aspect to it
and also where that that that content is not as structured,
meaning like, you know, movies and series, I'll sit in some
database somewhere where they have unique identifiers that
everyone understands across multiple system.
Sports lacks that sort of structured metadata regardless

(30:34):
of league thing, right? Like there's some database of
players across every sport that you can reference and have good
data around. So it helps with that problem as
well. I'm just thinking now just a
little bit outside the box and Nick got me thinking on this is
when we do talk about sports, weget more of these direct to
consumer platforms where they aren't just all about content,
they're also selling merchandiseor they're selling additional

(30:57):
products. So that just curious, you know,
from your perspective, the how, how much can this data then be
connected across like your broader ecosystem, You know, in
terms of wanting to be able to connect what my D to C platform
is, you know, I've acquired a customer and then you talked
about earlier, like actually themetric is probably ARPU.
Are there ways for me to be ableto, you know, kind of connect
one to do from a content side? If I'm, you know, I'm the I'm

(31:21):
the zone, I've subscribed to NFLGame Pass, you know, are they
going to be able to help the NFLdrive me to buy tickets for the
Wembley game? You know, are you able, how
easily is it to connect some of those learnings and data to
potentially Dr ARPU to other places that just isn't strictly
content if you're building that kind of one stop shop platform?
Yeah. And that's certainly a major
focus of many of our customers in the area that we think about

(31:42):
and are focused on a lot from a road map perspective.
So, and it also ties nicely into, you know, at a high level,
and I sort of was talking about the different ways in which AI
creates new scalable opportunities versus traditional
personalization. And, you know, gave that, that
metadata enrichment example, thecuration example, sort of the
real time context example. But continuing from that real

(32:03):
time context, which is so important for live is this is
more the I'd say the the next frontier of these things if
everyone's figuring out the right model for this.
But what I'll call sort of content understanding, which is
saying AI analysing the video file itself during the encoding
process in order to understand what is happening in the video,

(32:26):
whether that is that a home run was just hit or whether someone
just scored a goal. Because those tend to be the
types of triggers people want totie to this person at a home
run, present an ad to buy the jersey of that player.
Right? And there are some, there are,
let me wrong, their existing companies that provide data
feeds around these types of things that we have integrated
previously into our platform, which then gets into, I mean,

(32:50):
not to get too technical, but for this audience perhaps like
Scuddy markers and that kind of thing where like there is that
sort of sidecar feed of events that are happening, which you
can then react to. But you can also use this for
things like this is again, a frontier or something we're
we're working on, but I'll put more of this sort of
experimentation phase and figuring out how the best
product dies. But things like highlights, for

(33:11):
example, right, we have this like the highlight catch up
highlights that you see around alot of services today.
You know, you tune into an NFL game, an NBA game, maybe you
joined in progress. Let's show you sort of that
highlight reel of the videos that happened earlier.
You know that with AI can becomemuch more scalable to include
things like fantasy football. Let's say there's some specific
player that you're interested in.

(33:32):
You're not interested in necessarily just watching all
the field goals and touchdowns that happened so far in the
game. You want to see every time that
person touched the ball, clippedout.
And one really cool thing that we, and this is one thing I'd
love to say about mkio platform and the fact that it uses our
mkio processing pipeline is thatso much the innovation and
flexibility is in that video processing layer.

(33:54):
Because if we're already encoding the content, we can do
the AI understanding of the content as a part of that flow
rather than egressing that videoout somewhere else for content
understanding, increasing your latency while bringing it back
in and all that. So we try to make it sort of a
part of what we're we're doing and a really cool thing that we
do, which I mean, it, it does get into technical, but I feel
like it's too cool not to mention here is that so many of

(34:16):
these things are about, you know, you generate a bunch of
odd assets basically that then need to be stored somewhere and
played back to the user. But a lot of these AI things are
based around more like manifest manipulation where it's about
finding out what the relevant segments are in the actual, just
not a separate video stream, butthe video stream of the event.
And all those segments of the video sitting on the CDN and

(34:38):
figure out what are the correct ones to show to somebody as
highlights. And then pull them out of the
existing, basically out of the existing video that's already
stored there. And then just sort of repackage
it for the user to watch it, right?
Cuz you need to figure that out if you're gonna scale to truly
personalise someday where somebody goes.
I just wanna watch all the highlights of Messy and all the

(35:00):
goals that he scored in the 2018season of Whatever, right?
And like to generate content sort of on demand for users in a
personalised way, which I think is sort of the Holy Grail of the
future here. So figure out how to make that
scalable is certainly something we think about a lot.
I'll just say Rich, we, Nick andI have been hosting this podcast
for four years now. I've been working in sports for
over six years now and I've I'vesaid it multiple times at

(35:22):
events. I've sent it multiple times a
podcast. If you wanted to just literally
take the money out of my pocket.If some if someone just built a
platform where I can upload my fantasy team and then in on
every Monday morning. I just woke up to a 20 minute
package of just the carries of my running backs of just the
reception. Take my money.

(35:44):
You can have it included into FPL like I'm begging for rich.
If someone can do that, talk about what services would I
subscribe to Just take all of mymoney.
Good to know. Yeah, No, I mean, it's, it's, I
don't know, it's, it's an interesting challenge because I
think, you know, live is definitely where it's at, don't

(36:05):
get me wrong. And like I it's, it's health.
It's healthy to have a little bit of cynicism or scepticism, I
should say, about the value of that library of VOD assets of
all the old games and things like that.
But if there is ever technology that I saw that like maybe this
is the thing that cracks the nutof figuring out how to repurpose

(36:26):
that content in a very entertaining to the individual
user way. These technologies are the ones
that I look at and go, man, if you've got that, especially as a
sports, maybe you licence the games during seasons, but
generally you retain long term your library of all those past
matches or games or whatever that perhaps that's a way to
leverage it from you user experience perspective, make it
relevant to users in a more modern way is to be able to

(36:48):
repackage it again in a scalableway from a tech perspective.
But you know, it's all, it's sort of scary, but like, scary
is not the right word. It's a brave new future, a brave
new world of are we moving from a paradigm that traditionally
was, OK, I've got all of this content and now my goal is to

(37:08):
figure out to use personalization to get people to
watch my content Library 2. Do I actually basically generate
the content based on what they want rather than this matching
game, right. Because it's like if somebody's
describing a show, they want to watch every highlight of my
fantasy team and then we are generating, it might be just a

(37:29):
playlist at first, but we're literally generating the content
based on their demand. And we're figuring out ways to
make that scalable so that each user doesn't just get a
different homepage, but each user can have literally content
that's generated on the fly for them.
Essentially. It's a massive, I mean, it's,
it's the name of the game and the opportunity you're going
forward, I think in general for video.
So rich in terms of all the stuff we've been talking about.

(37:51):
I mentioned it before about delineation between the big
players years versus the maybe the the start-ups or the
challenges of someone who's maybe having to be really
focused on. They want to build a really
efficient and effective businessthat is not throwing the kitchen
sink financially at some of these ideas and opportunities.
Just how impactful have you seenthe the personalization part of

(38:16):
it play a role? Can you give any examples of
look, if you, if you don't, you've seen since implementing
exit had this sort of impact into conversions or into
engagement times, what sort of guidance can give up but just
the scale of the impact it can have?
I mean the scale, I can't think of any.

(38:36):
I mean the numbers have changed.I mean, the numbers are so
different from test to test of what you're doing, but things
that I can say proved out time and time again that they made a
material impact on subscription metrics of like a subscription
service, for example. And things that I, I mean, I'm
leaning into the most things that we're doing at Media kind
because my belief is that they will drive content engagement

(38:57):
for our customers, which will then increase their subscription
metrics, which will also drive them to consume more video
content, which is also healthy for our business of more video
being processed and everyone wins.
But specific things that have a material impact, certainly
personalising the order of the rows on your homepage certainly

(39:18):
does that. Personalising the sort order of
content within the rows. I mean, as a general principle,
people tend to stay to the left side of the page and go down.
So like you need to optimise forthe content.
The top left corner is where youwant the content to go that is
most personalised to a user. And the first item in each row
going down the page is going to have greater engagement than

(39:40):
anything. That's people.
People scroll vertically before they scroll horizontally.
And that's proven out of time and time again.
So affecting those two things, which is something we're working
on in both cases for our customers, that's going to drive
material impact. Another one that always probably
the biggest lift of anything I'dsay I've done is that sort of

(40:00):
watch next experience inside of your video player.
I know people tend to think of it is obviously when you're
watching a series sort of autoplay to the next episode of
the show. But leveraging that to surface
correct content, particularly where the AI comes in for the
additional context piece servingup the correct next links to you
can keep people in the video player and naturally juices

(40:21):
engagement there. And it's not just like, oh, they
watched a movie, get them to watch the next movie.
A lot of times people are watching the newest episode of a
show and so there is no next episode to bring them into.
So you're identifying a piece ofcontent and getting them into
there. That is always a huge driver of
of further engagement there. And then also around sort of
push notifications, things like that, but like, and it could be

(40:42):
literally push notification, it could be a marketing e-mail.
But something we've seen good has been a good driver for
customers of ours is identifyingwhat somebody is watching and
then sending the push notification stuff that are tied
to that. Oh, they started watching the
show, they abandoned it, Maybe they forgot to finish that
movie, try to get them to to watch that.
So all of those are things that drive and then things that we

(41:03):
have, they're not just working on, but we've launched and seen
material impact are basically improvements to our search, our
search recommend like our our search algorithms and
increasing. Let's talk about this earlier,
but using AI to understand semantic meaning of content in
order to drive better content similarity and identify
connections between pieces of content you might have missed

(41:23):
otherwise. And then show those that higher
up in search results that has driven material impact for
customers of ours that are on that version of our search API.
And that's something we're, you know, rolling out with more
customers over time. I'm wondering about the the just
to that just the do you have anything that stands out to you
over your time, whether it's in your previous days or now with

(41:46):
media con that are some perhaps random but outliers that a trend
you can say that, hey, if this person is doing X or not doing
X, they're way more likely to churn than someone who's doing
YS. The example I actually have that
stuck with with the first times I heard talked about this was
actually years and years ago when the old NBA app existed

(42:08):
before Mediacon actually got involved and there was a a stat
that was something to do. I can't remember the exact
number was basically if we saw them consuming non live content
on the platform, that was the best indicator that that were
they were going to stay in and stay on if they weren't just
what's coming on for the live. Do you have any interesting
outliers that can really point to someone either churning or to

(42:29):
staying on as a long time customer other than just
watching lots of stuff? More Grant, I mean still in the
in the realm of watching stuff, but a little bit more granular
on this front of and whatever the patterns are that exists
with our customers, which is a little bit varied from customer
to customer. So I don't know, it's not a
great answer as relates to our current customers of media kind.

(42:51):
I haven't been here long enough to identify those patterns, but
certainly in my previous roles where original series was such a
big part of what we were doing that original series and
particularly users watching morethan what like free trial
conversion, the magic numbers getting them to watch that
second series that they come in and they start often they come
in, they watch an original series, they try to fit the

(43:12):
entire thing into the free trial.
They'll often sometimes I'm surewe won't done this.
You like wait until the entire season is done and up and then
you sign up for the free trial. Watch the entire season in churn
before leave, before the free trial ends thing.
If you can get somebody to watcha second original series or any
original or not get somebody to watch a second series besides
the one that they came in and initially watched.

(43:33):
That has a dramatic Inflexion point in terms of the number of
people. So we were always, for example,
setting that as like an OK R forthe quarter or whatever to be
like try to get that number fromone series per user to 1.5 or
whatever on average. That was one that was always a
huge one. See, I'd say that's probably the
best example of a, a magic thingthat would work as a lever to

(43:54):
get people in. Actually, one other one I'd
mention here, which I think it'sreally important to take the eye
off the ball, which something wecare about a lot is sort of the
the reliability of your service.Certainly.
And this is where I love sellingMKL platform with MKL under it
because it is a rock solid bulletproof streaming solution,

(44:15):
which is why the disowns of the world and people in an MBA and
whatever love the our product isbecause scalability is through
the roof of what we're capable of doing and the video quality
that we provide. And one thing I've seen over
time, right, and some of the better, you know, quality of
experience and observability of what's going on in your, your
platform. And if you can figure out which

(44:36):
we did over time, certain sort of metrics around start up time
in the video player and things like that, where you can, you
can often do sort of like churn prediction based on certain
metrics around video quality, like video quality and video
playback and start up time and things like that.
Like it's not just like fix the bugs and make it better.
You can often, if you look in the data close enough.
And I've done this in some cases, like identified specific

(44:59):
metrics that do impact your churn, which is hugely helpful
because while quality the product, obviously everyone
understands sometimes people churn off of a product because
of poor quality, but often that is in App Store ratings or in
like NPS surveys and things likethat.
And it's hard to tie that type of data to individual users and
their behaviours at scale, like in your engagement data, because

(45:21):
App Store reviews are completelydivorced in the data you have of
like usage of the product. But this video quality and
quality experience is is a piecewhere you can tie it to user
cohorts and understand how the quality of your experience
impacts. Sure.
Yeah. Well, Rich, we're going to have
to wrap things up, unfortunately.
But I found this incredibly enlightening because for me, the
whole time you've been speaking about this, I've been sat in the

(45:43):
back of my head being like, OK, when I'm going through Disney
Plus and I'm trying to find a show, I've kind of just always
accepted it's one of those things like the Wizard of Oz,
don't look behind the curtain. Or as I sometimes say, you don't
need to know how the sausage is made.
But today, like I got to learn alittle bit about how like the
homepage sausage is made. And just I was set to being
like, you know what? I do kind of behave like that.

(46:03):
And so, you know, so I found it really interesting because it's,
I think it's something I've always kind of just vaguely
known, like there's a system that's trying to do stuff for
me. But I found this really
interesting to get more insight to what's actually going on in
the back and the considerations that people are taking into it.
So I think, Nick, I've always said anytime we can leave a
conversation with someone where I feel like I've learned from
it. That's a very good episode.

(46:25):
I'm glad to hear that it was that it was helpful and
informative and I appreciate youguys having me on and I am I'm
happy to come back anytime in the future people the topics we
want to talk about and and yeah,thank you so much for inviting
me. Great stuff, Rich.
Thanks very much. That was super interesting and I
look forward to seeing more about what me comes up to.
No, I think they'll be in a quite in a significant

(46:46):
attendance at the media Summit next month in Madrid as well.
So shout out to to that event ina couple of weeks time.
But thanks very much for joiningus.
Thank you. Cool.
Thank you so much.