Latest / The Yield Doctor with James Deaker / Ian Thomas building Analytics in Digital Media

Transcript

Ian Thomas building Analytics in Digital Media · The Yield Doctor with James Deaker

0:02
hi my guest today is Ian Thomas Ian has
0:06
worked in senior roles running data
0:08
organizations and most recently in a
0:11
series of Chief data officer roles one
0:14
of the things that is unique about Ian
0:17
is that he has worked in digital media
0:20
on both sides of the industry he has
0:22
worked on the CES side at companies like
0:24
Microsoft and on the buy side at indust
0:28
at companies like PU
0:30
he also brings different perspectives
0:33
having worked in both the United States
0:35
and where he is now in the United
0:37
Kingdom on digital media problems so
0:39
we're going to get into some of those
0:41
different perspectives today Ian thank
0:44
you so much for taking the time to talk
0:46
well thanks for inviting me Jame it's
0:48
great to have the CH to chat excellent
0:50
now Ian let's start with some of the
0:52
basics of the publisher side analytics
0:55
since that's where you and I first
0:57
crossed paths uh when I think about
1:00
publisher analytics I always think there
1:02
are two main work streams or categories
1:06
one is what are consumers doing on
1:08
publisher properties and then the second
1:11
different stream is how are advertisers
1:14
performing on those properties are those
1:16
the two main buckets you think about or
1:18
do you use a different organizing
1:20
framework when you're talking about
1:22
publisher side
1:23
problems yeah I think um those are two
1:26
very important areas for sure uh and
1:29
publishers particularly for you know for
1:31
a long time have been interested in the
1:33
consumption of content on their sites
1:35
and how people are navigating through
1:37
the sites and so on and as as you say um
1:40
understanding how they are monetizing
1:42
that that inventory and I think those
1:45
two things are coming together a little
1:46
bit at the moment with Publishers really
1:48
focusing on building a more holistic
1:50
understanding of their their audiences
1:54
and when we say the word audiences uh to
1:57
most people that kind of makes people
1:59
think of you know big cookie pools and
2:02
you know tracking people at the
2:03
individual level um and uh plenty of
2:06
thoughts about you know about the future
2:07
of that which we'll probably come on to
2:10
um but but there's also the opportunity
2:13
now for Publishers to understand their
2:14
audiences and kind of a more holistic
2:16
sense um joining together first party
2:19
data they may contribute around their
2:22
interests um and their own kind of
2:24
profile information the riseing
2:26
subscription based business models in
2:28
publishing I think is partly driven by
2:30
that um but then there's also Publishers
2:32
really understanding better the uh the
2:35
sentiment and kind of uh context of what
2:39
their audiences are actually consuming
2:41
and of course and that's not just or
2:44
it's moving on from the state of the
2:45
days of saying oh okay people you know
2:47
people were looking at um uh you know
2:50
current affairs content or information
2:52
about cars or information about
2:54
celebrities you know these sort of
2:56
simple categorizations into more
3:00
granular understanding of um of the
3:03
topics and the interests and starting to
3:06
use that to build new pictures and new
3:08
ways of thinking about their audiences
3:09
that they can then turn around and to
3:11
sell to advertisers I think it's really
3:13
interesting for to be a publisher and
3:15
hopefully a better time for be a
3:17
publisher than it has been for the last
3:19
five to 10
3:20
years yeah and we we'll get into that in
3:23
a minute because I think Publishers both
3:26
face a unique set of challenges and
3:28
opportunities around data and analytics
3:31
but but let's flip for a minute because
3:32
you then moved to the agency side after
3:35
you left Microsoft you spent time at
3:37
Publis and other agencies when you made
3:40
that transition from the sell side to
3:42
the buy side what did you see as the
3:45
main differences and sort of the
3:47
problems and challenges that the
3:49
agencies are facing around data and
3:52
analytics well yeah it was it was super
3:54
interesting to to move into the uh into
3:57
the agency world I got a little bit of a
3:59
taste that actually while I was still at
4:01
Microsoft because one of my my later
4:04
roles there was um building and running
4:08
larg scale customer day platform for the
4:10
Microsoft consumer business and so we
4:12
were working hand inand um with uh with
4:16
our agency partners on on that work
4:19
which is supporting a lot of
4:20
relationship marketing um and one of the
4:23
things that was interesting to observe
4:25
and to experience both at Microsoft and
4:28
then later was was that a lot of
4:30
agencies still kind of operate on the
4:32
outside when it comes to data with their
4:34
clients so there's the kind of agency
4:36
world of building and running um you
4:40
know an adtech platform capturing data
4:42
from various sources trying to use that
4:44
to Target ads um and then using that
4:47
data to understand the performance of
4:49
that ad although you know those
4:51
campaigns um but a lot of that is still
4:53
very much kind of in its own little Silo
4:56
and agencies still struggle and
4:59
certainly you know two three years ago
5:01
when I was doing some of this work with
5:02
with with Publishers um they still
5:05
struggle to kind of cross that um the
5:08
Rubicon of actually getting real
5:10
partnership with their clients to share
5:12
the data that they need to really drive
5:15
performance and understand and
5:17
understand performance agencies are
5:18
still kind of held at arms length and so
5:21
you see a lot of that um uh in these uh
5:24
you know these hos these days of
5:26
agencies trying to extend their value
5:28
proposition
5:30
um I I the time I spend at Omnicom for
5:33
example I was working for omnicom's
5:35
in-house Consulting arm which was
5:37
explicitly about helping the agencies
5:39
cross that barrier and it was a real
5:40
challenge because the clients were not
5:42
used to buying that different kind of
5:45
service um and then the other thing
5:47
that's interesting about agencies is
5:49
they are um they're very audience-based
5:52
again um and that a lot of that is
5:55
driven by a lot of the buying behaviors
5:58
of their clients and so I was involved
6:02
with various um uh pitches at publicist
6:06
for example where essentially the major
6:08
client ask was like tell me how I can
6:10
find my
6:11
audiences uh out on the open internet
6:14
you know what tools can you bring um and
6:16
so you publicist created um the all
6:20
people cloud is now part of Epsilon post
6:22
that acquisition Omnicom has Omni a
6:26
similar offering which are really about
6:27
aggregating third party audience data TR
6:29
provide that targeting capability but
6:31
again with the way the industry is going
6:33
I'm not sure that's as an approach is
6:36
going to have sort of a lot of legs um
6:38
in the medium term unless some of that
6:40
audience methodology really really
6:42
changes um and agencies can kind of get
6:45
inside the tent with their clients which
6:47
I think some of them are achieving and
6:48
some of them are having less success at
6:51
yeah I mean obviously even though it's
6:53
the same industry very different types
6:55
of problems on the two different sides
6:57
of the industry hey um internet
7:00
companies are famous for pioneering the
7:03
use of of Big Data Technologies like
7:06
Hadoop And Hive simply because of the
7:08
volume of data that is generated in the
7:11
Internet space uh one of the concerns
7:13
that I had when I ran a data team was
7:16
that with so much data availability
7:19
teams can get distracted by pursuing
7:22
analyses that I put in the category of
7:24
interesting but not necessarily
7:27
impactful So to that end end how as
7:30
you've managed Big Data teams do you
7:33
ensure that with so much data the right
7:35
questions are being asked by the
7:37
organization
7:39
overall well um I mean I've seen the
7:43
same thing that that you have and it was
7:47
a major feature of uh of the kind of the
7:50
covid years that a lot of organizations
7:53
took the time to invest heavily in um in
7:57
Big Data platforms um lot of data Lakes
8:00
got implemented um a lot of data got
8:03
dumped in those data Lakes um and then
8:06
in many cases sort of data scientists
8:08
were hired and told to kind of just go
8:10
fish really in those in those lakes with
8:12
the kind of hope that they would just
8:14
sort of you know emerge from their
8:16
darkened cave at some point and go I've
8:18
come up with this amazing Insight or
8:21
algorithm that's going to you know make
8:23
us$10 million or something um and that
8:27
that really just doesn't happen
8:30
um and I think you the the the big Tech
8:34
firms uh not deliberately I don't really
8:36
hold them kind of accountable for this
8:38
but they've inadvertently set kind of a
8:40
terrible a terrible example um to the
8:44
rest of the world because they have
8:45
problems that they need to hire large
8:47
numbers of data scientists to do you
8:50
know highend work to solve um whereas
8:54
most organizations have more sort of
8:56
work a day needs not all of which have
8:58
to be solved with data science in the
9:00
first place and the you know the
9:03
starting point as you'll be unsurprised
9:04
to hear is sitting down and figuring out
9:07
what business problems you're trying to
9:08
solve in the first place and thinking
9:13
inclusively about what might solve um a
9:17
business problem so you know we we were
9:20
talking just a minute ago about the you
9:22
advertisers and Publishers and what
9:24
they're trying to solve you advertisers
9:26
are trying to solve the problem of
9:29
driving the best response rate or Roi or
9:34
or roas on the money they're spending on
9:36
Advertising that's what they're trying
9:38
to solve for they're not trying to solve
9:40
for you know particular kind of
9:42
targeting or a particular kind of
9:44
algorithm or whatever they're trying to
9:46
drive they're trying to drive value um
9:49
and um and so really sitting down and
9:53
asking those questions up front of what
9:56
kind of problems you're trying to solve
9:57
and then thinking okay what's the
9:59
easiest way we can solve that problem
10:01
you know one of the challenges uh with
10:03
this sort of you know this rush to Big
10:05
Data and science is that um somebody in
10:09
the room would be like oh I could you
10:10
know I could I could use a compilation
10:12
neuron Network to solve that say yeah
10:14
sure you could um but you know I've got
10:16
here I've just got a little piece of you
10:18
know piece of basic business logic which
10:20
will solve it you know 80% as well for a
10:22
hundredth of the cost um and so it's
10:25
really about sitting down thinking about
10:28
what business problems are we trying to
10:29
solve and
10:31
setting goals and committing to those
10:34
goals ahead of time um at Microsoft
10:38
actually we spend quite a lot of time
10:41
working on exec education to help our
10:44
senior execs frame problems in a way
10:47
that could be helped with data um a lot
10:49
of um you know a lot of senior execs
10:52
still kind of basically do what they
10:55
want to do and then just set data teams
10:58
to you know kind of find the answer that
11:00
the thing they did solve the problem
11:02
they thought they were G to they were
11:03
going to get and so senior executive
11:06
need to be really honest with themselves
11:08
about what they're trying to achieve and
11:10
actually commit to those outcomes Ian I
11:12
could not agree with you more it's one
11:14
of these things where everybody feels
11:16
that they need a data team let's go
11:17
quickly hire one and work out what
11:18
they're going to do later and so
11:21
actually just as you've been talking I
11:22
want to go slightly in a direction I
11:24
hadn't intended which is where a data
11:27
team should be should sit within an
11:29
organization let's let's stay on the
11:31
publisher side because we talk big data
11:33
and Facebook and Google and others are
11:35
famous for their use of Big Data teams
11:38
in in your experience is it more
11:40
successful to have one big centralized
11:44
data team where you centralize all the
11:46
skills and knowledge and then have them
11:49
act as a service provider to the rest of
11:52
the company so that there's actually
11:54
some eyes on the consistency of approach
11:56
Etc or is it generally more effective to
12:00
have sort of um decentralized teams
12:03
where your you know marketing team might
12:05
have their own data team data and
12:06
analytics you might have your consumer
12:09
insights team have their own data and
12:11
insights team Etc and maybe you have an
12:13
overlay team uh who sort of manages the
12:17
processes amongst all these different
12:18
federations but how I mean what have you
12:21
seen I mean you could argue that either
12:23
of those models works but what have you
12:24
seen be more more successful in
12:27
practice well yeah it's really good
12:29
question um and it very much depends on
12:33
the level of data maturity within the
12:37
organization so the the conversation I
12:40
most often have around this topic is um
12:44
is to explain to people that that this
12:48
whole thing is a journey and so an
12:51
organization if somebody if somebody
12:52
came to me you know client comes to me
12:54
and says you we don't really have any
12:57
kind of data function and we don't
12:59
really know what what we're doing with
13:01
this stuff I will probably recommend to
13:04
them that they create a fairly
13:05
centralized data team because they need
13:09
to create the shortest path possible
13:11
between building out platform
13:14
capabilities not just technology
13:15
platform but you know data platform
13:17
capabilities building out you know a
13:18
good um data foundation for doing
13:21
analytics and so on um and you know a
13:25
short path between that and the people
13:27
who are actually going to use that data
13:29
so having a data team which is
13:32
combination of you know some data
13:33
Engineers you know Architects some you
13:36
some analysts maybe a data scientist at
13:38
some point etc etc um all working
13:41
closely together is a really great way
13:43
to get started as the organization
13:46
matures then it becomes I think it's
13:49
desirable actively desirable actually to
13:51
explicitly have the sort of orientation
13:54
in your thinking that you that the
13:56
particularly the analytics work will at
13:58
some point become some
13:59
decentralized
14:01
because a central data team and it's a
14:05
very common problem Central data team is
14:07
serving the needs of multiple different
14:09
stakeholders within the
14:10
organization those
14:12
stakeholders don't really care about
14:14
each other it's not it's kind of not
14:16
their job to care about each other
14:17
they've got their own jobs to do you
14:18
know Finance doesn't necessarily care
14:19
about marketing doesn't necessarily care
14:21
about product and so on um H except in
14:24
the broader you know sort of K are we're
14:26
all part of one big company thing but in
14:28
terms of what they need today they've
14:29
got their own needs and you know and and
14:32
so a central team is always eventually
14:34
going to struggle just prioritizing the
14:36
conflicting needs of different teams um
14:39
and you end up with ring fenced
14:42
resources and stuff and that can get you
14:43
so far but eventually that leads to be
14:46
able to say okay look you know marketing
14:49
if you've got you know if you want to do
14:51
more and more with this data how about
14:52
your highest manaly yourselves and we
14:55
will Pride the foundation we will
14:58
support you and train the analysts or S
15:00
maybe we'll lend you somebody for a
15:02
while um but you will get you to become
15:04
more self-sufficient in what you're in
15:06
what you're doing so the data team
15:08
becomes more of a foundational service
15:11
provider and kind of Center of
15:12
Excellence to the rest of the
15:15
organization um as they become you know
15:17
and and these this kind of
15:19
decentralization is typically highly
15:22
asymmetric so you know some teams are
15:24
quite self-sufficient while other teams
15:25
are very much not and there's a scale
15:27
aspect um
15:29
uh because you um you know there's some
15:31
small team and he say all you have to do
15:33
is hire a whole person to do analytics
15:35
they're like we a't got the budget for a
15:36
whole person so you might end up having
15:37
to kind of maybe support those you know
15:40
those people
15:41
centrally um but really thinking about a
15:43
federative model in in due in due course
15:46
you some
15:48
organizations have you know multiple
15:50
data teams because they have multiple
15:52
sources of data you know Microsoft was
15:54
crazy big example of that because the
15:56
big product
15:57
organizations were generating a lot of
15:59
their own data a ton of their own data
16:02
and have their own fully formed data
16:03
teams and so there you have a central
16:05
team or series of central teams which
16:08
would just try to kind of coordinate but
16:10
that's again Microsoft is not a great
16:12
example because it's one of you know the
16:13
big Tech firms most organizations are
16:16
much more towards the the sort of the
16:18
lower maturity end of the spectrum yeah
16:21
uh changing directions a little bit
16:23
there has been a lot of talk at the
16:25
moment about privacy and the death of
16:28
third part
16:29
cookies I mean this is a huge topic but
16:32
how do you think that this will impact
16:34
advertisers and Publishers both in their
16:37
ability to collect the data that they
16:39
need and their ability to to use that
16:41
data effectively whether it's in
16:43
targeting or optimization or in other
16:45
methods yeah I'm I'm I'm pretty excited
16:48
about it actually weirdly
16:51
um I don't know about you but I got
16:54
pretty sick of being followed around the
16:55
Internet by you know that sofa I looked
16:57
at on a website three weeks ago um and I
17:01
feel like the whole the whole rise you
17:04
it's it's a really interesting example
17:06
actually of you know there's some MBA
17:09
thesis somewhere in here about what
17:10
happens in a sort of unregulated Market
17:13
where you know the market is moving you
17:16
sort of uper trajectory at some point
17:18
but is that the right overall direction
17:19
for it to go in because the easy
17:22
availability of third party tracking
17:24
Technologies um plus the rise of you
17:28
know the Big W Garden providers of of
17:30
Google and
17:31
metsa has really kind of hollowed out um
17:35
a big part of kind of media planning and
17:36
buying over the last you know five to 10
17:39
years um you know as we were sort of
17:41
touching on earlier it's kind of it's
17:43
impoverished a lot of uh of of
17:46
Publishers in you know in the open
17:48
internet um have them you know seed a
17:51
ton of control over a a ton of knowledge
17:53
about Their audience to to third parties
17:56
like meta um and it's not created a
17:59
great experience for um for consumers
18:02
you know you the number of times I've
18:04
been on what I consider to be a
18:06
reasonably prestigious publisher site um
18:09
you know and I scroll down to the bottom
18:11
of of a of a a of a article and by
18:14
myself in you know kind of tabua hell
18:16
where there's all these you know low
18:18
relevancy ads that apparently are
18:20
targeted to my cookie which you know
18:22
which are not adding to the experience
18:25
so let me sorry let me interrupt you
18:28
because you're actually going in a
18:29
different direction than I thought and I
18:31
think I would have a very different
18:32
perspective so I want to understand
18:33
yours I had this discussion with someone
18:36
you and I both know actually yesterday
18:38
where we were talking about who was
18:39
likely to be the beneficiaries of
18:42
cookies going away and my perspective
18:45
which I think sounds different than
18:46
yours is the people who always win will
18:48
win again meta Google Etc World Gardens
18:52
they're big they've got the scale and I
18:54
use the analogy that I think this is
18:57
just like gdpr was in Europe 3 four
19:01
years ago where well-meaning Regulators
19:05
put these rules in place not realizing
19:08
that many midsize Publishers wouldn't
19:11
have the capabilities the scale to be
19:15
able to compete with the the big guys
19:18
and so I think you what I'm hearing from
19:21
you is you actually think that this
19:23
could be something that gives a
19:24
competitive advantage to some of those
19:27
midsize Publishers against the wall
19:30
Gardens and clearly I'm coming at it
19:32
from a different place so help me um
19:35
understand the rationale of why the W
19:37
Gardens may be more negatively impacted
19:41
than say a midsize
19:42
publisher yeah well I wouldn't I mean I
19:44
wouldn't count them out for sure um I
19:48
think uh you know there's been a lot of
19:50
you know a lot of chatter about what
19:53
Google for example is doing with with
19:55
privacy sandbox and would it just you
19:57
know just the crew more
19:59
um uh you know more power to Google one
20:02
of the big champions in that fight
20:04
actually is is the competition and
20:06
markets Authority here in the UK um who
20:08
are you know very closely scrutinizing
20:10
what's going on there to make sure that
20:11
it doesn't create additional competitive
20:13
Advantage for Google so I I think that
20:17
the um the world Gardens will still be
20:20
strong where I think there's an
20:23
opportunity uh for Publishers is to work
20:27
with third party that are able to
20:30
leverage um deep contextual
20:33
understanding of publisher content to
20:36
create value that the the publisher can
20:41
know potentially own themselves it's
20:43
it's their content the semantic
20:44
understanding of it to be something that
20:45
they they you know they create to
20:49
enable uh things like contextual
20:51
targeting and sort of synthetic
20:53
audiences based upon behaviors that
20:57
historically publishers haven't been
20:58
able to do because the technology hasn't
21:00
been there and hasn't really been the
21:01
incentive it's been it's been too easy
21:04
for uh advertisers just to you know
21:07
spend their money with the world Gardens
21:10
um and and buy you know you know and buy
21:14
sort of you know cheap
21:17
undifferentiated uh inventory that just
21:19
happens to be kind of you know be well
21:21
targeted particularly sort of publisher
21:24
networks for for things like meta so as
21:26
I said I don't think it's gonna I don't
21:29
think you know say I wouldn't count
21:30
Google and meta out I'm just seeing some
21:32
interesting emerging
21:34
players um you know players like you
21:37
know teds and CAG that are really
21:39
offering some interesting targeting
21:42
capabilities that I think would be very
21:44
interesting to third parking Publishers
21:47
um and this you know this is relying a
21:48
lot on the use of you know our new best
21:51
friend you know generative AI um to
21:54
create um that understand that
21:56
contextual understanding and then do
21:57
other things like bring in um campaign
22:00
optimization and so on which you know
22:02
which really benefits the advertisers so
22:04
that's actually where I I wanted to go
22:06
let let's talk about that a bit which
22:09
when cookies are getting deprecated you
22:12
know identify as a potentially being
22:15
impacted you see generative AI as one of
22:18
the main techniques for sort of
22:21
offsetting some of the decline that
22:23
advertisers might see in their um aoas
22:28
or or that Publishers may see in the
22:30
yield you you're optimistic it sounds
22:32
like about the ability for AI to create
22:37
new techniques or new segmentation
22:40
methodologies U can you talk a little
22:42
bit about that yeah I think there some
22:44
there's some interesting things so
22:47
um there's you know there's there's two
22:51
ways or two major areas that that that
22:53
I'm excited about um in driving campaign
22:57
performance in this new world one I've
22:58
already touched on a couple of times
22:59
which is this idea of of deep contextual
23:02
targeting so really being able to uh to
23:05
to Target ads much more in a much High
23:08
much higher level of relevance to their
23:10
to their context uh that gives you you
23:12
know new tools in the in the in the
23:14
drive to have brand safety for example
23:16
so your targeting algorithm can tell the
23:19
difference between a story about um you
23:22
know a Boeing aircraft and another story
23:24
about a Boeing aircraft that's crashed
23:26
killing you know a whole bunch of people
23:27
so you don't have a time
23:29
vacations along time alongside the
23:30
second um the Second Story
23:35
um those those
23:37
techniques uh to do that deep contextual
23:39
targeting are fundamentally the same as
23:41
a lot of the core technology inside of
23:43
generative AI so it's around what we
23:46
call semantic indexing so going in and
23:49
um paing the content of those sites to
23:51
build a structured representation of the
23:53
content which you can then match to the
23:56
the sentiment of the um of the ads your
23:58
you're choosing to show the other side
24:00
of the coin though the the other big
24:02
thing is campaign and creative
24:05
optimization and there's been techniques
24:08
in this area for you know for quite a
24:10
while so you know dynamic creative
24:11
optimization has been a thing for ages
24:14
um and there's quite a lot of well-
24:15
evolved solutions for doing things like
24:19
inserting uh you know locally relevant
24:22
uh content into ads so that you know you
24:24
have one ad created but it has an
24:25
overlay for the you know the local store
24:28
that you're trying to drive people to
24:29
with you know hyperlocal targeted
24:30
advertising and that sort of thing and
24:32
you know using combinatorial techniques
24:34
to um uh to create variants and so on
24:37
and then test those variants to see
24:40
which which ad combinations or ad
24:42
creative variants actually work but one
24:45
of the big impediments to that
24:47
historically were has always been the
24:49
creating of the variance in the first
24:51
place and um I remember you know we we
24:55
ran a test with some technology to do
24:58
this again back when I was at Microsoft
25:00
um and we had a a platform we were
25:04
working with which said hey you know if
25:06
you can give us you know 10 times as
25:07
many ad variants we can drive
25:09
performance you know this sort of
25:10
iterative optimization algorithm that we
25:13
have this cool piece of technology theyi
25:16
we can drive you know better performance
25:18
so we turn around to uh to the agency
25:20
and say can we have 10 times as much
25:22
creative and they said of course it's
25:24
just going to cost you 10 times as much
25:25
money and we're like oh crap okay well
25:27
maybe we'll think about this differently
25:29
but now using using gen and we're not
25:33
talking by the way when you know people
25:34
think about gen created they're think
25:36
about you typing in you know happy young
25:39
person holding a bottle of Coca-Cola and
25:41
you know that the magic ad will come out
25:43
that is not what we're talking about
25:45
here what we're talking about is
25:47
dynamically recomposing ads putting in
25:49
new backgrounds changing you know time
25:52
of day lighting the color of a vehicle
25:55
the color of other elements in you know
25:57
in the in the the image um generating
26:01
lots and lots of different versions of
26:02
taglines and other you know other AD
26:04
copy and so on so you end up with
26:06
something that you can vary thousands of
26:09
ways and then treat that as a giant
26:12
experiment to you know delivering that
26:15
across the audience cross referencing it
26:18
with what you know about the audience
26:19
and other you know other environmental
26:21
or kind of targeting out or contextual
26:24
attributes and then picking the winners
26:26
on a continuous kind of iterative basis
26:29
um and so that can drive you can you can
26:31
achieve a lot of what I've just
26:32
described almost all of what I've just
26:34
described without using any kind of
26:36
identifiers for the audience um uh
26:39
certainly some of the tracking will need
26:42
to kind of pass through things like
26:43
privacy sandbox because that you know
26:46
even you know even even our tracking and
26:48
conversion measurement needs to be kind
26:50
of go through those environments now but
26:52
it's a there's a lot of opportunity
26:53
there and and the interesting thing is
26:55
that as I say it's not just the kind of
26:58
you you know the the the the sort of
27:00
high-profile like you know type in your
27:02
ad it pops out thing but it's the same
27:04
fundamental technology as used to
27:06
generate these things you just think
27:07
thinking it would support this kind of
27:09
experimentation based approach got it
27:12
thank you um look I know you spend a lot
27:15
of your time now advising companies on
27:18
how to build grow make more successful
27:20
their data
27:21
organizations when you think about the
27:24
current skills that are needed in a data
27:27
science team the these days talk to me
27:30
about the need to pull the data versus
27:32
structure the queries versus
27:34
sophisticated analysis versus soft
27:36
skills like working with other teams you
27:40
know what what's your sort of perfect
27:42
team mix or skill mix look like
27:44
now yeah well I I I frequently say to
27:48
people you data science is really a team
27:50
sport uh the yeah the sort of the Mythos
27:53
of this you know this data scientist
27:55
this genius data scientist who lives in
27:57
the dark and off this and occasionally
27:59
emerges you know waving a piece of paper
28:02
with an amazing algorithm on it is you
28:03
know that was never really a thing
28:04
outside the big Tech firms and we
28:06
already talked about the bad example
28:08
they've set um so in order to have an
28:12
effective data team you need people to
28:14
manage data through the whole of what I
28:17
would call the data value chain so you
28:20
need a a fairly you know a group of
28:22
people who essentially think and operate
28:24
like
28:25
Engineers who take care of of of
28:28
sourcing the data from wherever it's
28:30
coming from bringing it in um uh storing
28:34
it and and structuring it in a in a in a
28:37
in a scalable way in some kind of
28:38
appropriate platform which they also
28:40
manage um uh like a proper kind of
28:43
product platform um and then you need a
28:46
set of people
28:48
who um can think a bit like Engineers
28:51
but also think a little bit like
28:52
analysts and sometimes these people are
28:54
referred to as analytics Engineers um
28:58
and they are they're slightly unicorny
29:01
people you have to you know not get too
29:03
hung up on exactly the right mix of
29:04
skills and but these are people who are
29:06
really good on understanding business
29:08
context around um reporting analytic
29:11
scenarios and what might might be needed
29:12
for data science um but who understand
29:16
enough about the underlying data to be
29:17
able to structure the data in that in in
29:19
that way um or in a way that's going to
29:21
be able to serve you know to serve those
29:24
those requirements um and the whole
29:26
industry has grown up around analytics
29:27
engineers
29:28
um um the um there's an analytics
29:31
engineering platform called DBT which
29:34
now runs one of you know the biggest
29:35
conferences in the uh in in the industry
29:38
really focusing on that really essential
29:39
kind of um switchboard uh type of
29:42
activity and then you know you've got
29:44
then you've got data scientists and uh
29:46
analysts
29:48
who who are
29:50
servicing sort of day-to-day servicing
29:52
stakeholder requirements and building
29:54
capabilities and um and uh thinking
29:58
about that in a kind of productized way
30:00
so one of the things I preach a lot is
30:02
the idea of data product management um
30:06
back when you and I were working
30:07
together at Microsoft the the Microsoft
30:09
internal tool that I was running we ran
30:12
it like a product we had a release
30:13
schedule for it we had release notes we
30:16
had documentation we had onboarding um
30:19
if we had you know if we had a
30:21
disruptive change we would carry people
30:23
you know through that change and so on
30:25
and a lot of that rigor around providing
30:28
as a product internally is still very
30:31
poorly involved within organizations so
30:33
some you know some team will be like oh
30:34
we made a database you can go query up
30:36
for cool stuff and then you go in and
30:37
it's completely different and to the
30:39
last time you used it and they're like
30:40
oh yeah we changed it and you're like
30:42
well did you plan on telling anybody so
30:45
that kind of thing thinking about it
30:46
like a product um is uh is really
30:49
important and then with you
30:51
know one one big thing that analyzed all
30:54
of this is this idea
30:56
of um
30:58
data Ops and ml Ops so the data products
31:03
and machine learning models and so on
31:04
this whole thing again if you think
31:06
about them as products they have a life
31:08
cycle you know they're in development
31:09
they're in testing they go into
31:11
production they go into you know they're
31:13
they're being maintained they need they
31:15
might need to be retired particularly
31:17
machine learning models might need to be
31:18
retired and replaced with new ones so
31:20
there's a whole process piece there
31:23
which um again gets gets kind of missed
31:27
when people think about what a data team
31:29
does they kind of think us of smart
31:31
people writing SQL queries or writing
31:33
code in Python but there's actually a
31:35
whole sort of ml data Ops piece there
31:37
which which shares a lot of common
31:39
characteristics with with the deop
31:41
discipline that's really you know grown
31:42
up a lot in in engineering team so it's
31:44
a it's a really interesting mix of you
31:46
know mix of people inovated team
31:47
typically yeah um as we start to sort of
31:50
come to the end of our time together in
31:53
maybe this is a a summary type question
31:55
but when you think what are the sort of
31:57
top two or three things that companies
32:00
get wrong when they're trying to build
32:02
an analytics function within digital
32:05
media yeah I think um the first one I
32:08
would say is number one is starting with
32:11
the tech rather than the business
32:12
problem we we talked about this earlier
32:14
of of somebody going oh you know I'm I'm
32:18
you know I'm running a project to
32:19
implement snowflake like okay sure but
32:22
like why are you putting snowflake in
32:25
what's the you know what are the
32:26
business problems because they can
32:27
become very self um you know
32:30
self-defining uh where the outcome you
32:33
know people like what's the business
32:35
outcome and the business outcome ends up
32:36
being like Oh well the business outcome
32:37
is we've got
32:38
snowflake congratulations you know
32:40
you've got more technology to support
32:42
that's not a business outcome so that's
32:44
a a very common problem I see across
32:47
lots of businesses including digital
32:48
media businesses then the second one is
32:53
particularly in again particularly in in
32:55
a in a sort of digital digital media
32:57
cont text is um operating in s so
33:00
particularly marketing teams not
33:03
involving or not not working properly
33:05
with the the tech and data team um and
33:08
that results in people setting up you
33:10
know um um separate Stacks to do you
33:14
know to do to do work around particular
33:15
around customer data platforms um and
33:19
and other kind of you know campaign
33:20
analytics type things and it's I get it
33:23
you know if you're in a marketing team
33:24
and I've seen this play out so many
33:26
times that you're some in a team and
33:29
they can't articulate what they need in
33:31
a way that the the technology of the
33:33
engineering team understands and the
33:36
technology team can't ask questions of
33:40
the marketing team in a way the
33:42
marketing team understands and often
33:44
it's the role of the data team in my
33:45
experience to mediate between um between
33:48
those two organizations um so that the
33:51
data is made available to support what
33:52
the marketing team needs and the
33:54
marketing requirements are kind of
33:55
translated into ways that can be
33:56
implemented by by Engineers so working
33:58
inide is is a real Challenge and then
34:01
the third one I see people get wrong a
34:03
lot is just just looking in the rearview
34:07
mirror so a lot of campaign analysis is
34:12
still people sitting down at the end of
34:14
a campaign and looking at a bunch of
34:16
dashboards and going how did the
34:18
campaign do with the again with this
34:21
sort of faith-based expectation that
34:23
just by looking at some numbers about
34:24
the campaign that the next one will
34:26
somehow be better
34:28
and really you need to flip that on your
34:30
on its head and say okay look we're
34:33
going to be investing you know this
34:34
money in these you know these marketing
34:38
campaigns we we have decided we want to
34:40
drive this specific outcome like pick
34:44
one like is it conversions is it Expos
34:49
you know is it brand awareness is it is
34:51
it clicks whatever it is but pick the
34:53
you know pick the outcome you're trying
34:54
to drive and then stick to that um and
34:58
of optimized against it so that you're
35:00
doing what I would call kind of good
35:02
science so you know all the decisions
35:04
you take are against the outcome you're
35:06
trying to drive um and that enables you
35:09
to look
35:10
forward to make decisions about what
35:12
will happen in the future rather than
35:13
just looking back and looking at a bunch
35:15
of you know a bunch of numbers um but
35:17
it's still very very difficult because
35:19
unless people have a good understanding
35:21
understanding of what they they could be
35:23
rigorous about the about the outcome
35:25
they're trying to drive you can't look
35:28
forward because you don't know the
35:29
direction you're trying to you're trying
35:31
to go in so that's a real kind of data
35:33
maturity thing that I that I help a lot
35:35
of my clients with Ian it has been an
35:37
absolute pleasure to catch up with you
35:40
um as we come to a close are there any
35:43
last thoughts that you'd like to
35:45
share uh well as said it's been great to
35:47
great to catch up with you it's been how
35:49
many years since we since we had a good
35:51
chat um but I think I think it's a
35:54
really as I said earlier a really
35:56
interesting time to be in the
35:58
uh in the digital media and advertising
35:59
industry at the moment I think a few
36:02
years from now we'll we'll we'll look
36:04
back and we'll be amazed at the journey
36:06
that that you know that the that the
36:08
industry has been able to reinvent
36:09
itself kind of yet again um and a lot of
36:12
you know a lot of what's old is new
36:14
again you know a lot of the stuff that
36:15
you and I worked on 10 years ago is kind
36:17
of back in Vogue You Know You quality
36:20
content is all of the sou um King of the
36:23
Hill again and I'm and I'm as I said
36:25
earlier I'm delighted about that because
36:26
I think it you know could be good for
36:28
Publishers good for advertisers and
36:30
actually good for uh you know us suckers
36:32
actually having to use the internet on a
36:33
day-to-day basis excellent thank you so
36:36
much Ian have a good rest of your day
36:38
yeah thanks
36:42
games