Speaker:
00:00:00
Hello and welcome to Fibonacci, the Red Olive Data Podcast, all about data
Speaker:
00:00:04
and analytics, where we hear from leading specialists and get their take
Speaker:
00:00:08
on the industry. I'm your host, Nicky Rudd. Today, I'm joined by James
Speaker:
00:00:12
Gardiner, head of Data Technology for Admiral Insurance. James has nearly
Speaker:
00:00:16
25 years of working with data and has been at Admiral for the
Speaker:
00:00:19
past seven years, lately managing its move from on premise to the cloud.
Speaker:
00:00:23
He was a data consultant for 15 years before joining Admiral, working in
Speaker:
00:00:27
the telecoms and finance industries, where he consulted on data solutions
Speaker:
00:00:31
for companies including Lloyds, Barclays, Nokia and WS Atkins. We discuss
Speaker:
00:00:35
data governance, moving to the cloud, and the art of the possible in data
Speaker:
00:00:39
projects. Let's find out more. Let's start off with telling me how you
Speaker:
00:00:50
got into the world of data. What was your journey in?
Speaker:
00:00:53
I finished my degree in digital systems engineering. Way back, oh God, that's
Speaker:
00:01:01
27 years ago I think. My thesis at the time was doing things
Speaker:
00:01:06
with image tracking, tracking objects within images and following them and
Speaker:
00:01:10
things like that. So I started getting into
Speaker:
00:01:14
what was basically a bit of machine learning, so it's what they call
Speaker:
00:01:19
linear regression, and I started getting interested in that. When I came
Speaker:
00:01:24
out of university, I actually landed my first job,
Speaker:
00:01:27
was for a print inspection company. They were doing vision analysis
Speaker:
00:01:33
on print runs. So this could be anything, this could be wallpaper, newspapers,
Speaker:
00:01:37
anything like that. And what they wanted to do was build a system
Speaker:
00:01:41
that identified differences in colour trends while the press was moving,
Speaker:
00:01:46
so they didn't have to stop, saves them a lot of money,
Speaker:
00:01:49
etcetera, etcetera. So I went in there and I started coding in C++
Speaker:
00:01:54
to do some simple machine learning algorithms and things like that, that
Speaker:
00:01:57
they could then apply to the equipment that they sent out that was
Speaker:
00:02:01
bolted to these massive printing presses. Now, I did that for about a
Speaker:
00:02:05
year. And unfortunately it was a very small company, there was only about
Speaker:
00:02:10
12 of us and there was really nowhere for me to go in
Speaker:
00:02:14
there so my options were quite limited. I eventually joined a graduate scheme
Speaker:
00:02:20
for Welsh Water as part of their software division, and they put me
Speaker:
00:02:24
through loads of training with Capgemini and all that, they really invested
Speaker:
00:02:27
in me. I ended up being a junior
Speaker:
00:02:31
DBA SQL Server. So I did that for about two years,
Speaker:
00:02:35
and eventually got out of the junior position, became one of the development
Speaker:
00:02:39
DBAs there and things like that. And then unfortunately I got made redundant,
Speaker:
00:02:44
as did hundreds of other people on this project, you know how it goes,
Speaker:
00:02:47
this typical project just got shut down. And so from there,
Speaker:
00:02:51
I actually went to work for a telecoms company up in Nottingham. I
Speaker:
00:02:56
used to work three weeks at home and one week in the office
Speaker:
00:03:00
there. Basically, I was building and designing databases for them to whole
Speaker:
00:03:05
test data for their mobile devices and things like that, they used to
Speaker:
00:03:07
have big contracts with, at the time, Nokia, Ericsson, and the places that
Speaker:
00:03:12
are testing their handsets, and they always used to do a high level
Speaker:
00:03:16
of reporting on that test data. And I eventually
Speaker:
00:03:20
got into that and I was there for probably another two years
Speaker:
00:03:24
and then I got made redundant again, believe it or not. I got
Speaker:
00:03:28
made redundant again. The bottom dropped out of the mobile industry and
Speaker:
00:03:32
I got made redundant. And so I was looking around then for another
Speaker:
00:03:36
job, and there wasn't many permanent places in South Wales.
Speaker:
00:03:41
So I eventually landed a contract with WS Atkins, where I basically started
Speaker:
00:03:48
looking at the data for the road network, so basically this is the
Speaker:
00:03:53
M4. So in the M4 you have hundreds of sensors underneath the roads that
Speaker:
00:03:58
basically look at traffic flows, look at the speed of the traffic and
Speaker:
00:04:02
things like that, and they actually control the automated signs. So
Speaker:
00:04:06
when you're on the motorway and it says oh, "Slow down to 50,"
Speaker:
00:04:09
and things like that, it's normally because there's something, there's slow
Speaker:
00:04:11
traffic up ahead, like it could be a mile ahead. One of my
Speaker:
00:04:14
jobs was to take that data and actually mine it and start looking
Speaker:
00:04:19
at where all the congestion was on the roads and things like that,
Speaker:
00:04:22
and eventually this would go into road planning. So they would go,
Speaker:
00:04:27
"Right, okay. We need to open up another slip road here,
Speaker:
00:04:30
we need to do another bypass here," and things like that.
Speaker:
00:04:32
So I used to look at all this congestion
Speaker:
00:04:35
and things like that. Again, another two years there as a contractor,
Speaker:
00:04:39
and then I landed another contract at PHS, which then does the washrooms
Speaker:
00:04:45
and things like that. I built a financial data warehouse for them,
Speaker:
00:04:48
it was probably the first proper data warehouse I'd built.
Speaker:
00:04:52
I was there for a couple of years and then I went out
Speaker:
00:04:54
to the finance industry. I had all this data warehousing finance knowledge
Speaker:
00:04:59
and I went to a number of places, it was Lloyds, it was
Speaker:
00:05:02
Barclays, Barclaycard, lots of places like that. I also found myself in
Speaker:
00:05:08
Nokia, I was there for four years with Nokia Music. Again,
Speaker:
00:05:12
as a consultant, building financial data warehouses for them. I was also
Speaker:
00:05:17
with Principality, the building society, I was there for another four years.
Speaker:
00:05:20
Again, building financial data warehouses and things like that, and putting
Speaker:
00:05:24
in teams that could basically deliver these measures that the business wanted.
Speaker:
00:05:29
And then eventually I got a bit sick of contracting and I wanted
Speaker:
00:05:33
to have a bit more power, have a bit more leadership,
Speaker:
00:05:36
responsibility and things like that. So I joined Admiral, and I joined Admiral
Speaker:
00:05:40
about seven and a half years ago now. Since then I've made it
Speaker:
00:05:44
my home. I'm very happy to be there. That's the longest stint.
Speaker:
00:05:49
It is. It is definitely the longest stint. Normally I'd get a bit fed up,
Speaker:
00:05:52
I'd sort of move on. But then again, I was a consultant as
Speaker:
00:05:54
well. So about Admiral, yeah, I've been with Admiral for quite some time
Speaker:
00:05:59
now, and I'm currently the head of Technology for Data for Admiral. And
Speaker:
00:06:03
what that means is that it's all about capabilities. I look after most
Speaker:
00:06:08
of the data people and I construct workable agile teams for data.
Speaker:
00:06:15
As I said, it's all about capabilities, it's all about pushing down that
Speaker:
00:06:19
engineering mentality to the people on the floor to make them think about
Speaker:
00:06:23
data and not just, "Ah just coding away here," that there's a whole
Speaker:
00:06:27
big world and you really gotta understand the bigger picture in these sort
Speaker:
00:06:32
of things. And that's kind of what I do now to this day.
Speaker:
00:06:35
It's got a lot bigger, a lot quicker, 'cause obviously the demand is
Speaker:
00:06:39
just sky high. So now I basically lead and instruct those teams on
Speaker:
00:06:44
how to fulfill Admiral's data requests, and that's where I am now.
Speaker:
00:06:48
And I was gonna say that actually, even in the last
Speaker:
00:06:51
seven years that you've been there, you must have seen some massive changes.
Speaker:
00:06:53
We'll talk a little bit about the big project that's moving from on
Speaker:
00:06:58
premise to cloud, but what things do you think have been drivers over
Speaker:
00:07:02
the last five and a bit years to really push that technology and
Speaker:
00:07:06
that sort of data as a key building block of how the business
Speaker:
00:07:10
move forward? There's two sides to the insurance industry, and every insurance
Speaker:
00:07:15
company, I believe looks at things like that. And first they'll look at
Speaker:
00:07:18
things like value streams and in terms of right pricing, we need to
Speaker:
00:07:23
get the best and fairest prices for our customers. And then the second
Speaker:
00:07:28
is claims. You need to get the best claim service,
Speaker:
00:07:31
but also you have to be aware of things like fraud,
Speaker:
00:07:35
which costs a lot of money, which as the industry has blossomed, so
Speaker:
00:07:40
have bad guys as well, okay? So there's a lot of that out
Speaker:
00:07:44
there as well. But also just essentially giving people the data to make
Speaker:
00:07:50
good decisions, and it's gotta be good quality data. And obviously that's
Speaker:
00:07:54
not always the case, but yeah, I've seen it move from
Speaker:
00:07:59
very static data sets that only get updated once a day to basically
Speaker:
00:08:05
data sets that are essentially getting updated several times a day.
Speaker:
00:08:09
And then you have streaming data and things like that and real time
Speaker:
00:08:13
alerting and things like that. If you can do things like plumber into
Speaker:
00:08:17
your business processes so that your agents on the phone have really good
Speaker:
00:08:23
information about the customer that they are actually talking to,
Speaker:
00:08:27
can actually change or offer new products and things like that while they're
Speaker:
00:08:31
on the phone to them, then I think that's something we've seen in
Speaker:
00:08:35
the last couple of years that's been huge.
Speaker:
00:08:38
Do you think it's been a technology push, or do you think it's
Speaker:
00:08:41
been a leadership push and the sort of a mindset? 'Cause I think
Speaker:
00:08:45
insurance is obviously seen or has been seen as a very traditional setup
Speaker:
00:08:49
and traditional business. Has it been the people that have made the move,
Speaker:
00:08:54
or has it been just the fact that actually there's been advances in how
Speaker:
00:08:57
you can work with the data to give you that reporting and that insight?
Speaker:
00:09:01
I think it's been a bit of everything, to be honest.
Speaker:
00:09:03
I think people have realised what the art of a possible is.
Speaker:
00:09:09
So this is where you get your stream in these embedded processes,
Speaker:
00:09:13
these ML models and things like that that can
Speaker:
00:09:17
help the agent to make decisions on pricing and on a customer.
Speaker:
00:09:21
There's that part of it, and I think leadership have recognised that.
Speaker:
00:09:24
And obviously you can't fail to ignore things like that with the big
Speaker:
00:09:28
players, your Googles and your Amazons and things like that. Obviously there's
Speaker:
00:09:32
an ethics thing that comes into it. There are so many certain things
Speaker:
00:09:35
we're allowed to do. But you've got that side of the coin.
Speaker:
00:09:39
And then you've got the other side of the coin which we now
Speaker:
00:09:42
have these fantastic systems that can handle large amounts of data.
Speaker:
00:09:48
And if you're a data scientist or a pricing analyst or anything like
Speaker:
00:09:52
that, you're continually looking for other things to basically give that
Speaker:
00:09:56
customer the best price. Or you may be looking at things and actually
Speaker:
00:10:02
go, "Well, this guy is a bad driver here. He may not have
Speaker:
00:10:06
made any claims, but other things have happened. There's only so much we
Speaker:
00:10:09
could do, obviously, but other things," they are a higher risk,
Speaker:
00:10:12
so therefore they deserve a higher premium. And it's about protecting ourselves
Speaker:
00:10:15
as well as our customers and things like that. So it's a bit
Speaker:
00:10:18
of everything, really. And you can't ignore this change in architectures
Speaker:
00:10:22
and in technology as we've gone through time like this. When you're thinking
Speaker:
00:10:27
about all these changes, how do you go about thinking, "This is what
Speaker:
00:10:30
we're gonna do first," and actually the next steps?
Speaker:
00:10:33
First off, the experience will tell you the pitfalls, and it depends what
Speaker:
00:10:36
type of data project you're really starting. Because what you generally
Speaker:
00:10:42
start with is some sort of POC to prove out the technology,
Speaker:
00:10:46
prove out an idea and make sure it's quite ring fenced and your
Speaker:
00:10:51
acceptance criteria are quite well understood and measurable. So I think
Speaker:
00:10:57
it starts with that. And then you have to look at the personas, who's gonna
Speaker:
00:11:02
be accessing this data, whether or not it's a machine that's accessing the
Speaker:
00:11:05
data or a person, an agent and things like that. And you kind
Speaker:
00:11:10
of have to list those out, and you have to go and actually
Speaker:
00:11:13
talk to people. A lot of what my guys do is that they're
Speaker:
00:11:17
very heavily welded in with the business. A lot of business people actually
Speaker:
00:11:22
sit within our delivery teams to help and advise and go,
Speaker:
00:11:26
"That's right, that's not right." "These are the questions we need to be
Speaker:
00:11:29
asking." Exactly. And I think that's really important. That's kind of how
Speaker:
00:11:33
we start. And obviously you gotta get your architecture right. There are
Speaker:
00:11:38
internal processes for that and things like that. It can be a bit
Speaker:
00:11:42
frustrating sometimes and things like that, but that's the way of the world.
Speaker:
00:11:46
But essentially, yeah, I think that's what you start with. You go,
Speaker:
00:11:49
"Right, what's the scope of this? Who are the people accessing this data?
Speaker:
00:11:54
All those personas and things like that. What technology platforms are we
Speaker:
00:11:58
looking at to build this? What capabilities do we need to run some
Speaker:
00:12:01
POCs and things like that?" And also cost is a massive one as
Speaker:
00:12:05
well. Everyone's tightening their belts at the moment, so what is the cheapest
Speaker:
00:12:09
solution but the best solution overall that we can basically grow and we
Speaker:
00:12:14
can maintain properly and support properly? They are all things that need
Speaker:
00:12:19
thinking about. You generally drag in a lot of people into that process,
Speaker:
00:12:24
but I think that's where it starts. And then you start thinking about
Speaker:
00:12:29
also people from a delivery perspective, "Have I got the right capabilities?
Speaker:
00:12:33
Do I need to get some contractors in? What does that do to
Speaker:
00:12:36
the bottom line?" All these sort of type stuff. So
Speaker:
00:12:39
even though I'm quite technical, it does start a level above,
Speaker:
00:12:44
and you really have to get those things right first and a good
Speaker:
00:12:49
outset, and everyone needs to understand why we're doing what we're doing.
Speaker:
00:12:53
I think, yeah, they'll start with the why. I think that's really important
Speaker:
00:12:57
for us. The amount of times you've had a coder or something,
Speaker:
00:13:00
just, "Alright. Okay, so do you understand what you're doing here?"
Speaker:
00:13:04
"Because what I know what, I've gotta get this data from here to here."
Speaker:
00:13:06
Yeah, but why are you doing that?" "Oh, don't really know."
Speaker:
00:13:08
And that doesn't work, and I think... I think it's good to have
Speaker:
00:13:11
those meetings and explain the fundamental aims of what you're trying to
Speaker:
00:13:15
do and in what context, and then everyone can really get on board
Speaker:
00:13:20
and feel like a part of it. Those are the things we really
Speaker:
00:13:24
start with, but then obviously the next phase then is, "Let's do it.
Speaker:
00:13:29
Okay?" And that's when you have to organise teams,
Speaker:
00:13:32
you have to start... We're very agile, so you break everything into stories
Speaker:
00:13:37
and things like that. Everything get sized up.
Speaker:
00:13:39
Again, you're still looking constantly at the capabilities, have we got
Speaker:
00:13:43
enough people from networks to plumb this in? It's not just a data,
Speaker:
00:13:46
I just need data engineers, it's all sorts of disciplines across the board.
Speaker:
00:13:50
And I think that's where you start, basically. You at the moment obviously
Speaker:
00:13:54
have moved from on premise to cloud. I suppose the biggest project in
Speaker:
00:13:59
more recent times, I'm guessing. What sort of challenges has that thrown
Speaker:
00:14:02
up? Have there been things that you've had to consider differently than
Speaker:
00:14:06
perhaps having a non premise sort of set up? We're still halfway through
Speaker:
00:14:10
that at the moment, there's a lot of work, there's always something to
Speaker:
00:14:13
do, but security is the big one, I think.
Speaker:
00:14:17
With our on premise networks, there's a little bit of a safety blanket.
Speaker:
00:14:22
It's not all safe. Do you know what I mean? If someone wants
Speaker:
00:14:24
to get in, they can probably get in. But it's a lot easier
Speaker:
00:14:28
to deploy things knowing that you have things ring fenced and people won't
Speaker:
00:14:32
be able to exploit them. So security is a massive one,
Speaker:
00:14:36
because when you push it up into the cloud, it's so easy to
Speaker:
00:14:39
just make something public by mistake and things like that. So we do
Speaker:
00:14:44
spend a lot of time on security and making sure everything's encrypted in
Speaker:
00:14:49
transit and encrypted at rest and all this.
Speaker:
00:14:53
You do find though, a lot of the cloud services and all this
Speaker:
00:14:56
kind of happens by default, but you still gotta check it out and
Speaker:
00:14:59
make sure it's within your standards, if you like. So security is a
Speaker:
00:15:03
massive one. The other things to consider is governance, which changes drastically
Speaker:
00:15:10
as well when you're going to cloud because it's so easy for things
Speaker:
00:15:14
to get out of control. Because of the power of cloud,
Speaker:
00:15:17
essentially is something that I've led that I haven't learned before,
Speaker:
00:15:21
is that you can push date up into cloud and give everyone access,
Speaker:
00:15:25
but God, it becomes a massive mine field and it turns into the
Speaker:
00:15:29
Wild West if you're not careful. So you really have to basically keep
Speaker:
00:15:33
an eye on that governance, making sure that people have only got access
Speaker:
00:15:36
to the data that they need to do their day to day job.
Speaker:
00:15:39
And I think that's very important. It's always important on premise as well,
Speaker:
00:15:43
but with the power of the cloud and more people can do,
Speaker:
00:15:45
it can really get out of control so quickly.
Speaker:
00:15:49
There are a number of things that I consider, but one thing else
Speaker:
00:15:53
I'd like to say is that if you've got good governance and good
Speaker:
00:15:57
lineage of your current on premise systems, it becomes a lot easier to
Speaker:
00:16:04
go to cloud. Because you understand how the data is getting used.
Speaker:
00:16:08
It goes back to those personas again, who's using it, it's all documented
Speaker:
00:16:13
and things like that. In that case, it's a lot easier to move
Speaker:
00:16:16
and replicate and put those same controls in. So it's important to get
Speaker:
00:16:20
it right wherever you are. So yes, cloud does make a difference,
Speaker:
00:16:25
but you really should be putting these rules in place throughout,
Speaker:
00:16:29
no matter where you are. I think it's interesting you say that because
Speaker:
00:16:33
obviously there's great opportunities with cloud, the speed and
Speaker:
00:16:37
the openness, as you say, but yeah, a lesson learnt from something in
Speaker:
00:16:43
the cloud is probably a harder lesson than if it's actually still just
Speaker:
00:16:47
an internal system, a bit more closed down. Have you had to do
Speaker:
00:16:50
a lot of training or sessions with your team to get them thinking
Speaker:
00:16:55
differently with those points and those challenges? So there's always technical
Speaker:
00:16:59
training, that's a given, but like I said, it's having that engineering
Speaker:
00:17:04
mindset. Admiral's a very young company, we've got a lot of young people
Speaker:
00:17:09
working there, come straight out of university and we can train them up
Speaker:
00:17:13
and things like that. They need to understand the repercussions of things
Speaker:
00:17:17
going wrong. We have processes and things like that to control that,
Speaker:
00:17:21
but we'll generally give that insight into failing systems, a lot of the
Speaker:
00:17:28
time. Because you have to code to have resilience and make things robust.
Speaker:
00:17:33
So quite often we'll have reviews of code go, "Right, okay,
Speaker:
00:17:36
what happens if that thing goes wrong?" And it goes, "Oh, well,
Speaker:
00:17:39
it re tries." "Okay, alright, that's great. It re tries. But does it
Speaker:
00:17:43
clean up the message made when it went wrong?" It's things like that.
Speaker:
00:17:46
So it's having an engineering mindset. A lot of what I try to
Speaker:
00:17:50
put into context is that, look, would you put an aircraft in the
Speaker:
00:17:55
sky if you built it like that? What if something failed on the... Is
Speaker:
00:17:59
there a backup system? How does it get maintained, how does it get
Speaker:
00:18:02
checked? There's warning lights and things like that that you get?
Speaker:
00:18:06
Have you got the same warning lights and things in that in the
Speaker:
00:18:08
code and the data pipelines that you've put into this system?
Speaker:
00:18:12
That's why it's important to monitor, because everyone's brought up with
Speaker:
00:18:16
a CICD mentality, and a lot of CICD doesn't work when you have
Speaker:
00:18:24
a data project going on. And this is why you have data ops
Speaker:
00:18:27
instead of dev ops, because it's not only the code that has to
Speaker:
00:18:31
go through that process, it's the actual data itself. And you can get
Speaker:
00:18:34
bad data, so your code may be the best code written in the
Speaker:
00:18:38
world, but you get some bad data into it and everything can fall
Speaker:
00:18:43
down. And so you have to have these checks, you have to have
Speaker:
00:18:47
this live monitoring and things like that. And that's where I find machine
Speaker:
00:18:51
learning is actually coming into play, because you can actually apply machine
Speaker:
00:18:55
learning to that monitoring, to actually go, "Oh, this premium looks low
Speaker:
00:19:01
today," or on this percentage of measures or things like that.
Speaker:
00:19:05
You can actually get machine learning to actually invoke that alert and
Speaker:
00:19:08
then somebody can react to it and things like that. So although we
Speaker:
00:19:11
look at machine learning as, "Oh yes, gonna help this, gonna make... Better
Speaker:
00:19:15
profile my customers," and things like that, there's so many more applications
Speaker:
00:19:19
throughout the whole data pipeline, if you like, and that data flow,
Speaker:
00:19:24
that I think sometimes people ignore that. It's like turn the skills on
Speaker:
00:19:27
its head and use the machine learning to help make your systems as robust
Speaker:
00:19:32
and get your data as clean and as quality as possible.
Speaker:
00:19:37
I think it's interesting you say that, 'cause I was just about to
Speaker:
00:19:38
ask you about machine learning and ethics. 'Cause you'd mentioned that.
Speaker:
00:19:42
And I think there is that, yes, you can think about it from
Speaker:
00:19:45
a technology point of view and stuff moving very, very quickly,
Speaker:
00:19:47
but that role of a person in there just checking and putting in
Speaker:
00:19:51
the perimeters of kind of, "This is gonna be safe," or, "This is
Speaker:
00:19:55
ethical," or, "This is actually what we mean by good data,"
Speaker:
00:19:58
and actually the checks and the processes, is still very important.
Speaker:
00:20:01
Are there any particular trends or ways of working that you can really
Speaker:
00:20:05
see how that would jump start or that would really, really push a
Speaker:
00:20:09
different way of thinking? Well, there's a number of new ways of rolling
Speaker:
00:20:14
out data teams and rolling out data architectures and things like that,
Speaker:
00:20:17
one of them being mesh. There's this concept of a mesh architecture and
Speaker:
00:20:23
God, you really gotta be mature for some of this stuff,
Speaker:
00:20:26
and you really have to have those engineering principles embedded.
Speaker:
00:20:31
Because the whole concept of mesh is that you decentralise basically,
Speaker:
00:20:35
okay, and you farm that out to value streams and things like that.
Speaker:
00:20:38
Now, traditionally it's been like a data warehouse team and they've done
Speaker:
00:20:43
all your data and they've ingested it and things like that, but they
Speaker:
00:20:45
become a blocker, they become a blocker because they can only prioritise
Speaker:
00:20:49
so much, and someone over here in, I don't know, household insurance wants
Speaker:
00:20:55
to do some stuff and they can't. So what you try and do
Speaker:
00:20:59
is you try and farm out those capabilities to other value streams,
Speaker:
00:21:03
but that can be difficult unless you've got some blueprints for the central
Speaker:
00:21:08
governance because you want everything to work the same way
Speaker:
00:21:12
so you can maintain it. There's definitely some things that in the delivery
Speaker:
00:21:16
space that are being suggested. Everyone talks about mesh. I went to Big
Speaker:
00:21:20
Data London this year, and everyone was talking about mesh, but no one's
Speaker:
00:21:24
really nailed it. And it's the same thing with machine learning,
Speaker:
00:21:28
you get our pricing analyst thinking, "Right, okay, we've got a good machine
Speaker:
00:21:31
learning model here, we've trained it well. Right. We've got to get it
Speaker:
00:21:35
live." But because it's hooked into the whole quoting process and things
Speaker:
00:21:39
like that, you really gotta be careful that that is correct.
Speaker:
00:21:42
And so it can take a long time sometimes for these things to
Speaker:
00:21:46
get live, and this is where you have things like ML Ops springing
Speaker:
00:21:50
up. Okay? So it's not about the model itself and the machine learning
Speaker:
00:21:55
model itself, it's more about how quickly can you get that value into
Speaker:
00:21:59
the mainstream. And this is what ML Ops is giving us,
Speaker:
00:22:02
and essentially it's a set of governance rules really,
Speaker:
00:22:06
that I think data scientists aren't used to
Speaker:
00:22:09
dealing with that. Data scientists, they're in their own little world,
Speaker:
00:22:11
they're in a bubble and things like that. Now they have to think
Speaker:
00:22:15
in terms of engineering and what matters, "What if this model goes wrong?"
Speaker:
00:22:19
That sort of thing, they have to think about that. So everything has
Speaker:
00:22:22
to be thoroughly tested. And that takes time. So although you've got this
Speaker:
00:22:27
whole, "Oh yeah, ML Ops, AI," and things like that, you have to
Speaker:
00:22:31
determine the risk of certain things, and if you start
Speaker:
00:22:35
relying too much on it without having that whole framework around it that
Speaker:
00:22:40
gives you that robustness, you could be in a lot of trouble.
Speaker:
00:22:43
I was gonna say the kind of reporting and that, like you say,
Speaker:
00:22:46
having different people with different roles and responsibilities across
Speaker:
00:22:50
that sort of data pipeline, if you like, how do you go about
Speaker:
00:22:55
making that reporting and that kind of auditing and the reporting seamless
Speaker:
00:23:00
for everybody at the their worst? 'Cause it's such a big thing,
Speaker:
00:23:03
isn't it? That actually, you could break it down and make it so you're working
Speaker:
00:23:06
in pockets, but that actually would completely blow the overall name of
Speaker:
00:23:10
having this kind of thing. Have you got particularly with Admiral, ways
Speaker:
00:23:14
of working that you think you've nailed that? Or have there been questions
Speaker:
00:23:18
that you've along the way thought, "Actually, if we ask this with these
Speaker:
00:23:22
certain people, we are gonna have a better outcome at the end."?
Speaker:
00:23:25
I think, again, its mindset is a lot of it. So yes,
Speaker:
00:23:30
you can have these different teams and they're all working on different
Speaker:
00:23:32
things, but again, we go back to why.
Speaker:
00:23:35
Okay? Why you're doing this? You're an essential part of the cog...
Speaker:
00:23:39
An essential cog, sorry, not essential part. That is basically gonna deliver
Speaker:
00:23:43
this value at the end, and you are an essential part of it.
Speaker:
00:23:46
I think if people understand that, they generally
Speaker:
00:23:51
start to go out and communicate. And I think that's what it is,
Speaker:
00:23:54
it's all about good solid lines of communication. And obviously that's what
Speaker:
00:23:59
Agile gives you with the stand ups, your retrospectives and things like
Speaker:
00:24:02
that. This is what I'm heavily involved in, is that something is not
Speaker:
00:24:07
working, you change it. That's the the whole point, nothing's etched in
Speaker:
00:24:11
stone. If something's not performing or something went wrong at the end
Speaker:
00:24:15
of the last delivery or whatever, you change it, and communication is a
Speaker:
00:24:19
big factor in that. So when you have teams and you have them
Speaker:
00:24:23
in those silos and they're not communicating, you'll find that they're not
Speaker:
00:24:26
gonna line up at the end of your sprint or whatever your delivery
Speaker:
00:24:30
mechanism is, you have to change it. So
Speaker:
00:24:34
we get people to work together. That's the key. And communications is hard
Speaker:
00:24:39
when we're all working remote and things like that, but
Speaker:
00:24:42
it's still achievable, very much so. But it is about comms.
Speaker:
00:24:47
No one wants to come in and write something bad that doesn't work
Speaker:
00:24:50
at the end of it. But they didn't know, they were just sort
Speaker:
00:24:53
of carrying on. So it's important to get strong leaders to bring people
Speaker:
00:24:58
together and actually go, "This what we're doing. This is how we're doing
Speaker:
00:25:02
it. Everyone clear what they're doing?" "Yes." In the morning and stuff
Speaker:
00:25:06
like that. "If you're not, come and see me, we'll go and speak
Speaker:
00:25:08
to someone else," or whatever like that. That's kind of how it needs
Speaker:
00:25:11
to work. Within the data teams at Admiral, I think we've cracked that.
Speaker:
00:25:16
I think we have. We try to modularise as much as we can,
Speaker:
00:25:20
so somebody doesn't write a code for a specific things, they write it
Speaker:
00:25:25
for many things that can basically be used by another team.
Speaker:
00:25:29
So we try and write these modules and frameworks
Speaker:
00:25:33
for people to work in, rather than, "I'm just writing code." And I
Speaker:
00:25:37
think that's what helps. But there's also a lot of unit test checking
Speaker:
00:25:41
and things like that, there's a lot of automation that you can basically
Speaker:
00:25:45
put in, and we do this at our retros and things are like that. We went,
Speaker:
00:25:49
"Right, that went wrong, didn't it last time? Okay, what we're gonna do
Speaker:
00:25:52
about it?" "Right, we put an automated check in for that."
Speaker:
00:25:54
So as soon as someone deploys something into a test environment,
Speaker:
00:25:58
runs a host of unit test scripts or things are that and goes,
Speaker:
00:26:03
"No, that ain't gonna work with this," so we can build that in.
Speaker:
00:26:07
And we quite often do that. Don't capture everything. 'Cause it is hard.
Speaker:
00:26:11
Data is hard, I'll tell you that now. But if you have that
Speaker:
00:26:14
mindset and that ways of working, and that monitoring mindset and that robustness,
Speaker:
00:26:18
that engineering mindset, you can overcome it. Do you think that the governance
Speaker:
00:26:24
is kind of more stable or more solid because of that understanding of
Speaker:
00:26:28
data engineering mindset? With the governance? Yes, I think with governance
Speaker:
00:26:33
we've been in a better place than Admiral has ever been.
Speaker:
00:26:37
We bake things in, and again, our engineers think about it because they
Speaker:
00:26:43
want the best kudos for their bit of code.
Speaker:
00:26:47
So the only way you're gonna do that is get people actually using
Speaker:
00:26:51
it and using what the output of that code has produced.
Speaker:
00:26:55
Now, if you've got good governance that has put that into a data
Speaker:
00:26:58
catalog 'cause you've coded that up or is automatic or whatever,
Speaker:
00:27:02
then people will find that data and they'll use it. Okay? And this
Speaker:
00:27:06
is another thing about, probably dovetails into the working in silos.
Speaker:
00:27:11
You develop something for a specific thing, I know it has to happen
Speaker:
00:27:14
sometimes, but if you can write it so that it's open ended and
Speaker:
00:27:19
basically other people from outside the intended recipient can see it in
Speaker:
00:27:24
the data catalog, and within reason, they're allowed to see it and all
Speaker:
00:27:28
that. There's security and stuff like that. They can discover through a
Speaker:
00:27:31
data catalog and basically go, "Oh, I didn't realise someone had done that,
Speaker:
00:27:36
calculated that over there. I'll go and use that." And then they attribute
Speaker:
00:27:40
to the data catalog themselves by saying, "I am using this for this
Speaker:
00:27:44
purpose." That kind of mentality then grows, because everyone used to see
Speaker:
00:27:50
data governance is like a blocker, but essentially there's ways and means
Speaker:
00:27:55
of doing it where you can bake it into that whole delivery process
Speaker:
00:28:00
and actually make it an enabler and not a blocker. I think it
Speaker:
00:28:04
took a lot of companies a long time to understand that,
Speaker:
00:28:08
because I think they just thought of it is a regulatory thing and,
Speaker:
00:28:12
"Oh we've got to do this, we're ticking boxes here and there,"
Speaker:
00:28:14
and stuff like that, but I think you do it properly and it
Speaker:
00:28:18
is so valuable and it can aid operational efficiency. People have gotten
Speaker:
00:28:24
better decision making because the quality of the data has increased.
Speaker:
00:28:28
I used to go into meetings and some guy used to go,
Speaker:
00:28:31
"Well, this is what we sold this product for yesterday," and another guy
Speaker:
00:28:34
would go, "Well that don't match my figures." And you're like,
Speaker:
00:28:36
"Well, who's right?" And now with this data governance you've got a gold
Speaker:
00:28:40
standard, and now you got the guy going, "I got these figures."
Speaker:
00:28:43
"Oh, well I got these figures and they're different, but mine
Speaker:
00:28:46
have a tick on them, okay? They've been checked, so they've been gold
Speaker:
00:28:50
standard, this is actually what happened." The other guy's got no response
Speaker:
00:28:53
against that. And this is the type of things that it can enable,
Speaker:
00:28:58
but I think it takes a while for people to see that because
Speaker:
00:29:01
they do see it still as a blocker. But if you know data an you
Speaker:
00:29:05
know how it's used, you can really see the advantages of having that
Speaker:
00:29:09
strong governance. And I think what you're saying there as well,
Speaker:
00:29:13
is that actually just the quality of the reporting across an organisation
Speaker:
00:29:18
and also out to customers, will be better because you know that you're
Speaker:
00:29:22
using correct data and you're using it in the right way,
Speaker:
00:29:25
and it is standardised. It's not just the reporting, you look at the
Speaker:
00:29:28
AI and the machine learning and the data scientists and things like that,
Speaker:
00:29:31
it used to be they'd have this, "I just want all data. All
Speaker:
00:29:35
the time, as quick as you can, all data, all format, and I
Speaker:
00:29:39
want as much data as you can give me," and things like that. I
Speaker:
00:29:42
think the emphasis is slowly drifting away from that now to actually go,
Speaker:
00:29:47
"No, I want quality data." Because if you're training your models on bad
Speaker:
00:29:52
data, you're gonna get bad decisions, okay? So I think the emphasis has
Speaker:
00:29:57
now moved from, "Give me as much data as you can,"
Speaker:
00:30:02
to, "Just give me clean data." And I think that's an important step
Speaker:
00:30:05
because I don't think I could say that to you two years ago,
Speaker:
00:30:09
but I think now there's a machine learning model for everything,
Speaker:
00:30:12
I don't think anyone's really pushing the boundaries of machine learning.
Speaker:
00:30:16
Yeah, don't get me wrong, at Admiral we have custom ones and things
Speaker:
00:30:19
like that, and probably data scientists if they heard this in Admiral they'd
Speaker:
00:30:22
probably have a few choice words to me,
Speaker:
00:30:25
but to be honest, someone else has done it. Someone else has done
Speaker:
00:30:30
something very similar somewhere. You can pretty much buy a model off the
Speaker:
00:30:34
shelf, okay? And everyone is doing it, it's all like self service.
Speaker:
00:30:37
I can go on to Google's marketplace and I can buy
Speaker:
00:30:41
load of models and get the result I want.
Speaker:
00:30:44
What the difficult thing is the data engineering to get the data quality
Speaker:
00:30:47
up to a level when you can train that model, and I think
Speaker:
00:30:51
that's where the emphasis has changed there as well. So it's not just
Speaker:
00:30:53
the reporting, it's definitely for other applications as well.
Speaker:
00:30:58
So with the teams that you've got there, it sounds like obviously there's
Speaker:
00:31:01
a sort of sharing of knowledge and sharing of code, and you have
Speaker:
00:31:05
obviously done a project and read all of... Been involved with it. What's
Speaker:
00:31:09
it been like working with external consultants? Has that challenged the
Speaker:
00:31:12
thinking? Has it kind of steered the project?
Speaker:
00:31:16
You definitely learn from them. Because at the end of the day we're still
Speaker:
00:31:19
quite immature in some ways, Admiral's cloud journey.
Speaker:
00:31:23
Things take a long time and things like that. And you need that
Speaker:
00:31:25
external support, you need good people who have basically done this before
Speaker:
00:31:30
and know what the pitfalls are. You're still in charge, but you're in
Speaker:
00:31:34
charge of the outcomes, "These are the outcomes I have. Okay?
Speaker:
00:31:37
It needs to work like this." People like Red Olive as well, they come
Speaker:
00:31:40
in, they're classic engineers, they're trained in first principles.
Speaker:
00:31:45
And they will go, "Well James, if you do it like that,
Speaker:
00:31:47
then that could break there." So that's great advice. And so they're thinking
Speaker:
00:31:51
the same way, which is fantastic. You don't have to educate them in
Speaker:
00:31:56
robustness and maintenance and things like that. You just learn a lot from
Speaker:
00:31:59
them. You go, "Oh, okay. Well, we did some similar for the couple
Speaker:
00:32:04
of guys two years ago that did this. Would that work?"
Speaker:
00:32:09
And it's just bringing that experience to it. Because cloud is an experience
Speaker:
00:32:13
and you really have to sort of think differently. And especially when you're
Speaker:
00:32:17
on prem, you design for performance and when you get to cloud,
Speaker:
00:32:21
you're actually designing for cost. Because it's so easy to just throw another
Speaker:
00:32:26
100 processes at it and a shed load of memory, but there's a
Speaker:
00:32:31
cost implicated with that. So you still gotta be super efficient,
Speaker:
00:32:35
because otherwise those costs can run out of control and things like that.
Speaker:
00:32:39
And it's good that our external consultants and things like that, they've
Speaker:
00:32:41
all fed into that. They've gone, "Well, you can do it this way,
Speaker:
00:32:44
it's robust, it's really good, but it will cost you this and you
Speaker:
00:32:47
have to figure that out," and that can be very difficult.
Speaker:
00:32:50
But having people that have done it and things are like that,
Speaker:
00:32:53
that can really guide you essentially on that design from a cost perspective.
Speaker:
00:32:59
And also from a robustness perspective. It's a balance, isn't it?
Speaker:
00:33:02
So yeah, they've been really good. Well, though we've been talking data,
Speaker:
00:33:05
we've actually been talking about communications and people throughout this.
Speaker:
00:33:09
For people who are now coming into the data industry, what
Speaker:
00:33:14
skills would you recommend that they have? And what kind of things would
Speaker:
00:33:17
you be looking for for somebody joining industry now?
Speaker:
00:33:20
To be honest, we take on a lot of grads, a lot of
Speaker:
00:33:24
young people that may want to get into this sort of thing.
Speaker:
00:33:27
The key skill or two key skills I'll put down from a technical
Speaker:
00:33:31
perspective, SQL and Python. Because pretty much everything runs off those
Speaker:
00:33:37
two things. Or is interfaced with them in some means.
Speaker:
00:33:42
But what we really look for is... You can teach those skills.
Speaker:
00:33:46
If you've got a bit of a mathematics or science background or things
Speaker:
00:33:51
like, even if you haven't. But you know how an Excel spreadsheet works,
Speaker:
00:33:55
you know how to look through it, you know to organise data and
Speaker:
00:33:58
things like that, those are the key sort of technical skills we look
Speaker:
00:34:02
for. But we look for behaviours, and we do a lot of competency
Speaker:
00:34:07
based type interviews where we look at, "Give us an example of a
Speaker:
00:34:12
time when something went wrong?" And that may not be in data,
Speaker:
00:34:15
or the business could be something completely external, but we look at the
Speaker:
00:34:18
mindset about how people handle it. So there's a little bit of psychology
Speaker:
00:34:23
in there and things like that, but what we're looking for is people
Speaker:
00:34:26
who are conscientious, who care about what they're actually doing, and that's
Speaker:
00:34:31
important. So all the technical stuff, you can learn, you could go on
Speaker:
00:34:35
100 YouTubes there, right? And you can know concepts better than us.
Speaker:
00:34:39
But when it actually comes to implementing it and your day to day
Speaker:
00:34:43
job, I think your psychology and how you're made up has a lot
Speaker:
00:34:47
to do with it. So we do look for a lot of people
Speaker:
00:34:50
who are very driven, as I said conscientious, really care about what they're
Speaker:
00:34:54
doing, and you can make a really good start to mould them into
Speaker:
00:34:59
a classic data engineer. So that's what we look for.
Speaker:
00:35:03
My last question for you, what is the most exciting thing that you're
Speaker:
00:35:07
most passionate about for the data, pushing forward into the future?
Speaker:
00:35:12
Data automation. Because I know we talked about machine learning and AI
Speaker:
00:35:17
and things like that, and those things are great and very impressive and
Speaker:
00:35:21
things like that, but I think unless you're one of the top players,
Speaker:
00:35:26
they can be very difficult to implement on scale.
Speaker:
00:35:29
But what I'm interested in at the moment is a lot of metadata
Speaker:
00:35:34
sort of data driven pipelines, whereby, rather than deploying code you're
Speaker:
00:35:39
making more configuration changes and things like that. So it's the automation
Speaker:
00:35:43
of it all from a delivery point of view,
Speaker:
00:35:46
I like the idea of. Some people have had a go,
Speaker:
00:35:49
WhereScape being one of them, they're really good, but there needs to be
Speaker:
00:35:53
a next level. So when we're talking about... We've spoken about things like
Speaker:
00:35:57
governance, data cataloging and things like that, that type of thing needs
Speaker:
00:36:01
to happen automatically. So we talk about automation in terms of automated
Speaker:
00:36:06
testing or automated deployments and things like that, but what about actually
Speaker:
00:36:10
automating the actual development? Because a couple of years ago, a lot
Speaker:
00:36:14
of people just hand coding things, and there's a lot of now low
Speaker:
00:36:19
code solutions where you still need to have the technical ability and know
Speaker:
00:36:23
what's going on underneath, but you can template a lot of it.
Speaker:
00:36:26
So I can automatically template that it adds, I'm bringing new data in,
Speaker:
00:36:30
it adds to this data catalog. Or I publish a report, it automatically
Speaker:
00:36:34
gets added to this data catalog and things like that. So that's what I'm
Speaker:
00:36:37
really interested in, because everyone goes after the really exciting stuff
Speaker:
00:36:42
and things like that, but nobody thinks about the implementation so much.
Speaker:
00:36:46
Because I live in that space and all my problems have been in
Speaker:
00:36:50
that space, that's what I kind of get excited about now,
Speaker:
00:36:52
can I have a team of three people that can deliver something in
Speaker:
00:36:58
say a month, rather than a team of 10 deliver it in six
Speaker:
00:37:03
months? And that's kind of where I'm going at the moment,
Speaker:
00:37:07
because we're up to our eyeballs at work and it never stops.
Speaker:
00:37:11
And you can only grow so much before it becomes hard to manage.
Speaker:
00:37:15
But if we can deliver quicker and automate more and maintain our quality
Speaker:
00:37:21
standards and our governance standards and things like that, that's what
Speaker:
00:37:24
excites me. Now, that may be a bit of a boring answer,
Speaker:
00:37:27
but that's the honest truth. A fascinating take from James on the importance
Speaker:
00:37:32
of people, teamwork and communications in data projects. Join us for the
Speaker:
00:37:36
next episode of Fibonacci, the Red Olive Data Podcast, where we'll be joined
Speaker:
00:37:40
by another data expert sharing their thoughts on latest trends in AI and
Speaker:
00:37:43
big data, along with some great hints and tips.
Speaker:
00:37:46
Make sure you subscribe to Fibonacci, the Red Olive Data Podcast from wherever
Speaker:
00:37:50
you get your podcast, to make sure you don't miss it.
Speaker:
00:37:53
That's all for today. Thanks for listening. I've been your host,
00:00:00
Hello and welcome to Fibonacci, the Red Olive Data Podcast, all about data
Speaker:
00:00:04
and analytics, where we hear from leading specialists and get their take
Speaker:
00:00:08
on the industry. I'm your host, Nicky Rudd. Today, I'm joined by James
Speaker:
00:00:12
Gardiner, head of Data Technology for Admiral Insurance. James has nearly
Speaker:
00:00:16
25 years of working with data and has been at Admiral for the
Speaker:
00:00:19
past seven years, lately managing its move from on premise to the cloud.
Speaker:
00:00:23
He was a data consultant for 15 years before joining Admiral, working in
Speaker:
00:00:27
the telecoms and finance industries, where he consulted on data solutions
Speaker:
00:00:31
for companies including Lloyds, Barclays, Nokia and WS Atkins. We discuss
Speaker:
00:00:35
data governance, moving to the cloud, and the art of the possible in data
Speaker:
00:00:39
projects. Let's find out more. Let's start off with telling me how you
Speaker:
00:00:50
got into the world of data. What was your journey in?
Speaker:
00:00:53
I finished my degree in digital systems engineering. Way back, oh God, that's
Speaker:
00:01:01
27 years ago I think. My thesis at the time was doing things
Speaker:
00:01:06
with image tracking, tracking objects within images and following them and
Speaker:
00:01:10
things like that. So I started getting into
Speaker:
00:01:14
what was basically a bit of machine learning, so it's what they call
Speaker:
00:01:19
linear regression, and I started getting interested in that. When I came
Speaker:
00:01:24
out of university, I actually landed my first job,
Speaker:
00:01:27
was for a print inspection company. They were doing vision analysis
Speaker:
00:01:33
on print runs. So this could be anything, this could be wallpaper, newspapers,
Speaker:
00:01:37
anything like that. And what they wanted to do was build a system
Speaker:
00:01:41
that identified differences in colour trends while the press was moving,
Speaker:
00:01:46
so they didn't have to stop, saves them a lot of money,
Speaker:
00:01:49
etcetera, etcetera. So I went in there and I started coding in C++
Speaker:
00:01:54
to do some simple machine learning algorithms and things like that, that
Speaker:
00:01:57
they could then apply to the equipment that they sent out that was
Speaker:
00:02:01
bolted to these massive printing presses. Now, I did that for about a
Speaker:
00:02:05
year. And unfortunately it was a very small company, there was only about
Speaker:
00:02:10
12 of us and there was really nowhere for me to go in
Speaker:
00:02:14
there so my options were quite limited. I eventually joined a graduate scheme
Speaker:
00:02:20
for Welsh Water as part of their software division, and they put me
Speaker:
00:02:24
through loads of training with Capgemini and all that, they really invested
Speaker:
00:02:27
in me. I ended up being a junior
Speaker:
00:02:31
DBA SQL Server. So I did that for about two years,
Speaker:
00:02:35
and eventually got out of the junior position, became one of the development
Speaker:
00:02:39
DBAs there and things like that. And then unfortunately I got made redundant,
Speaker:
00:02:44
as did hundreds of other people on this project, you know how it goes,
Speaker:
00:02:47
this typical project just got shut down. And so from there,
Speaker:
00:02:51
I actually went to work for a telecoms company up in Nottingham. I
Speaker:
00:02:56
used to work three weeks at home and one week in the office
Speaker:
00:03:00
there. Basically, I was building and designing databases for them to whole
Speaker:
00:03:05
test data for their mobile devices and things like that, they used to
Speaker:
00:03:07
have big contracts with, at the time, Nokia, Ericsson, and the places that
Speaker:
00:03:12
are testing their handsets, and they always used to do a high level
Speaker:
00:03:16
of reporting on that test data. And I eventually
Speaker:
00:03:20
got into that and I was there for probably another two years
Speaker:
00:03:24
and then I got made redundant again, believe it or not. I got
Speaker:
00:03:28
made redundant again. The bottom dropped out of the mobile industry and
Speaker:
00:03:32
I got made redundant. And so I was looking around then for another
Speaker:
00:03:36
job, and there wasn't many permanent places in South Wales.
Speaker:
00:03:41
So I eventually landed a contract with WS Atkins, where I basically started
Speaker:
00:03:48
looking at the data for the road network, so basically this is the
Speaker:
00:03:53
M4. So in the M4 you have hundreds of sensors underneath the roads that
Speaker:
00:03:58
basically look at traffic flows, look at the speed of the traffic and
Speaker:
00:04:02
things like that, and they actually control the automated signs. So
Speaker:
00:04:06
when you're on the motorway and it says oh, "Slow down to 50,"
Speaker:
00:04:09
and things like that, it's normally because there's something, there's slow
Speaker:
00:04:11
traffic up ahead, like it could be a mile ahead. One of my
Speaker:
00:04:14
jobs was to take that data and actually mine it and start looking
Speaker:
00:04:19
at where all the congestion was on the roads and things like that,
Speaker:
00:04:22
and eventually this would go into road planning. So they would go,
Speaker:
00:04:27
"Right, okay. We need to open up another slip road here,
Speaker:
00:04:30
we need to do another bypass here," and things like that.
Speaker:
00:04:32
So I used to look at all this congestion
Speaker:
00:04:35
and things like that. Again, another two years there as a contractor,
Speaker:
00:04:39
and then I landed another contract at PHS, which then does the washrooms
Speaker:
00:04:45
and things like that. I built a financial data warehouse for them,
Speaker:
00:04:48
it was probably the first proper data warehouse I'd built.
Speaker:
00:04:52
I was there for a couple of years and then I went out
Speaker:
00:04:54
to the finance industry. I had all this data warehousing finance knowledge
Speaker:
00:04:59
and I went to a number of places, it was Lloyds, it was
Speaker:
00:05:02
Barclays, Barclaycard, lots of places like that. I also found myself in
Speaker:
00:05:08
Nokia, I was there for four years with Nokia Music. Again,
Speaker:
00:05:12
as a consultant, building financial data warehouses for them. I was also
Speaker:
00:05:17
with Principality, the building society, I was there for another four years.
Speaker:
00:05:20
Again, building financial data warehouses and things like that, and putting
Speaker:
00:05:24
in teams that could basically deliver these measures that the business wanted.
Speaker:
00:05:29
And then eventually I got a bit sick of contracting and I wanted
Speaker:
00:05:33
to have a bit more power, have a bit more leadership,
Speaker:
00:05:36
responsibility and things like that. So I joined Admiral, and I joined Admiral
Speaker:
00:05:40
about seven and a half years ago now. Since then I've made it
Speaker:
00:05:44
my home. I'm very happy to be there. That's the longest stint.
Speaker:
00:05:49
It is. It is definitely the longest stint. Normally I'd get a bit fed up,
Speaker:
00:05:52
I'd sort of move on. But then again, I was a consultant as
Speaker:
00:05:54
well. So about Admiral, yeah, I've been with Admiral for quite some time
Speaker:
00:05:59
now, and I'm currently the head of Technology for Data for Admiral. And
Speaker:
00:06:03
what that means is that it's all about capabilities. I look after most
Speaker:
00:06:08
of the data people and I construct workable agile teams for data.
Speaker:
00:06:15
As I said, it's all about capabilities, it's all about pushing down that
Speaker:
00:06:19
engineering mentality to the people on the floor to make them think about
Speaker:
00:06:23
data and not just, "Ah just coding away here," that there's a whole
Speaker:
00:06:27
big world and you really gotta understand the bigger picture in these sort
Speaker:
00:06:32
of things. And that's kind of what I do now to this day.
Speaker:
00:06:35
It's got a lot bigger, a lot quicker, 'cause obviously the demand is
Speaker:
00:06:39
just sky high. So now I basically lead and instruct those teams on
Speaker:
00:06:44
how to fulfill Admiral's data requests, and that's where I am now.
Speaker:
00:06:48
And I was gonna say that actually, even in the last
Speaker:
00:06:51
seven years that you've been there, you must have seen some massive changes.
Speaker:
00:06:53
We'll talk a little bit about the big project that's moving from on
Speaker:
00:06:58
premise to cloud, but what things do you think have been drivers over
Speaker:
00:07:02
the last five and a bit years to really push that technology and
Speaker:
00:07:06
that sort of data as a key building block of how the business
Speaker:
00:07:10
move forward? There's two sides to the insurance industry, and every insurance
Speaker:
00:07:15
company, I believe looks at things like that. And first they'll look at
Speaker:
00:07:18
things like value streams and in terms of right pricing, we need to
Speaker:
00:07:23
get the best and fairest prices for our customers. And then the second
Speaker:
00:07:28
is claims. You need to get the best claim service,
Speaker:
00:07:31
but also you have to be aware of things like fraud,
Speaker:
00:07:35
which costs a lot of money, which as the industry has blossomed, so
Speaker:
00:07:40
have bad guys as well, okay? So there's a lot of that out
Speaker:
00:07:44
there as well. But also just essentially giving people the data to make
Speaker:
00:07:50
good decisions, and it's gotta be good quality data. And obviously that's
Speaker:
00:07:54
not always the case, but yeah, I've seen it move from
Speaker:
00:07:59
very static data sets that only get updated once a day to basically
Speaker:
00:08:05
data sets that are essentially getting updated several times a day.
Speaker:
00:08:09
And then you have streaming data and things like that and real time
Speaker:
00:08:13
alerting and things like that. If you can do things like plumber into
Speaker:
00:08:17
your business processes so that your agents on the phone have really good
Speaker:
00:08:23
information about the customer that they are actually talking to,
Speaker:
00:08:27
can actually change or offer new products and things like that while they're
Speaker:
00:08:31
on the phone to them, then I think that's something we've seen in
Speaker:
00:08:35
the last couple of years that's been huge.
Speaker:
00:08:38
Do you think it's been a technology push, or do you think it's
Speaker:
00:08:41
been a leadership push and the sort of a mindset? 'Cause I think
Speaker:
00:08:45
insurance is obviously seen or has been seen as a very traditional setup
Speaker:
00:08:49
and traditional business. Has it been the people that have made the move,
Speaker:
00:08:54
or has it been just the fact that actually there's been advances in how
Speaker:
00:08:57
you can work with the data to give you that reporting and that insight?
Speaker:
00:09:01
I think it's been a bit of everything, to be honest.
Speaker:
00:09:03
I think people have realised what the art of a possible is.
Speaker:
00:09:09
So this is where you get your stream in these embedded processes,
Speaker:
00:09:13
these ML models and things like that that can
Speaker:
00:09:17
help the agent to make decisions on pricing and on a customer.
Speaker:
00:09:21
There's that part of it, and I think leadership have recognised that.
Speaker:
00:09:24
And obviously you can't fail to ignore things like that with the big
Speaker:
00:09:28
players, your Googles and your Amazons and things like that. Obviously there's
Speaker:
00:09:32
an ethics thing that comes into it. There are so many certain things
Speaker:
00:09:35
we're allowed to do. But you've got that side of the coin.
Speaker:
00:09:39
And then you've got the other side of the coin which we now
Speaker:
00:09:42
have these fantastic systems that can handle large amounts of data.
Speaker:
00:09:48
And if you're a data scientist or a pricing analyst or anything like
Speaker:
00:09:52
that, you're continually looking for other things to basically give that
Speaker:
00:09:56
customer the best price. Or you may be looking at things and actually
Speaker:
00:10:02
go, "Well, this guy is a bad driver here. He may not have
Speaker:
00:10:06
made any claims, but other things have happened. There's only so much we
Speaker:
00:10:09
could do, obviously, but other things," they are a higher risk,
Speaker:
00:10:12
so therefore they deserve a higher premium. And it's about protecting ourselves
Speaker:
00:10:15
as well as our customers and things like that. So it's a bit
Speaker:
00:10:18
of everything, really. And you can't ignore this change in architectures
Speaker:
00:10:22
and in technology as we've gone through time like this. When you're thinking
Speaker:
00:10:27
about all these changes, how do you go about thinking, "This is what
Speaker:
00:10:30
we're gonna do first," and actually the next steps?
Speaker:
00:10:33
First off, the experience will tell you the pitfalls, and it depends what
Speaker:
00:10:36
type of data project you're really starting. Because what you generally
Speaker:
00:10:42
start with is some sort of POC to prove out the technology,
Speaker:
00:10:46
prove out an idea and make sure it's quite ring fenced and your
Speaker:
00:10:51
acceptance criteria are quite well understood and measurable. So I think
Speaker:
00:10:57
it starts with that. And then you have to look at the personas, who's gonna
Speaker:
00:11:02
be accessing this data, whether or not it's a machine that's accessing the
Speaker:
00:11:05
data or a person, an agent and things like that. And you kind
Speaker:
00:11:10
of have to list those out, and you have to go and actually
Speaker:
00:11:13
talk to people. A lot of what my guys do is that they're
Speaker:
00:11:17
very heavily welded in with the business. A lot of business people actually
Speaker:
00:11:22
sit within our delivery teams to help and advise and go,
Speaker:
00:11:26
"That's right, that's not right." "These are the questions we need to be
Speaker:
00:11:29
asking." Exactly. And I think that's really important. That's kind of how
Speaker:
00:11:33
we start. And obviously you gotta get your architecture right. There are
Speaker:
00:11:38
internal processes for that and things like that. It can be a bit
Speaker:
00:11:42
frustrating sometimes and things like that, but that's the way of the world.
Speaker:
00:11:46
But essentially, yeah, I think that's what you start with. You go,
Speaker:
00:11:49
"Right, what's the scope of this? Who are the people accessing this data?
Speaker:
00:11:54
All those personas and things like that. What technology platforms are we
Speaker:
00:11:58
looking at to build this? What capabilities do we need to run some
Speaker:
00:12:01
POCs and things like that?" And also cost is a massive one as
Speaker:
00:12:05
well. Everyone's tightening their belts at the moment, so what is the cheapest
Speaker:
00:12:09
solution but the best solution overall that we can basically grow and we
Speaker:
00:12:14
can maintain properly and support properly? They are all things that need
Speaker:
00:12:19
thinking about. You generally drag in a lot of people into that process,
Speaker:
00:12:24
but I think that's where it starts. And then you start thinking about
Speaker:
00:12:29
also people from a delivery perspective, "Have I got the right capabilities?
Speaker:
00:12:33
Do I need to get some contractors in? What does that do to
Speaker:
00:12:36
the bottom line?" All these sort of type stuff. So
Speaker:
00:12:39
even though I'm quite technical, it does start a level above,
Speaker:
00:12:44
and you really have to get those things right first and a good
Speaker:
00:12:49
outset, and everyone needs to understand why we're doing what we're doing.
Speaker:
00:12:53
I think, yeah, they'll start with the why. I think that's really important
Speaker:
00:12:57
for us. The amount of times you've had a coder or something,
Speaker:
00:13:00
just, "Alright. Okay, so do you understand what you're doing here?"
Speaker:
00:13:04
"Because what I know what, I've gotta get this data from here to here."
Speaker:
00:13:06
Yeah, but why are you doing that?" "Oh, don't really know."
Speaker:
00:13:08
And that doesn't work, and I think... I think it's good to have
Speaker:
00:13:11
those meetings and explain the fundamental aims of what you're trying to
Speaker:
00:13:15
do and in what context, and then everyone can really get on board
Speaker:
00:13:20
and feel like a part of it. Those are the things we really
Speaker:
00:13:24
start with, but then obviously the next phase then is, "Let's do it.
Speaker:
00:13:29
Okay?" And that's when you have to organise teams,
Speaker:
00:13:32
you have to start... We're very agile, so you break everything into stories
Speaker:
00:13:37
and things like that. Everything get sized up.
Speaker:
00:13:39
Again, you're still looking constantly at the capabilities, have we got
Speaker:
00:13:43
enough people from networks to plumb this in? It's not just a data,
Speaker:
00:13:46
I just need data engineers, it's all sorts of disciplines across the board.
Speaker:
00:13:50
And I think that's where you start, basically. You at the moment obviously
Speaker:
00:13:54
have moved from on premise to cloud. I suppose the biggest project in
Speaker:
00:13:59
more recent times, I'm guessing. What sort of challenges has that thrown
Speaker:
00:14:02
up? Have there been things that you've had to consider differently than
Speaker:
00:14:06
perhaps having a non premise sort of set up? We're still halfway through
Speaker:
00:14:10
that at the moment, there's a lot of work, there's always something to
Speaker:
00:14:13
do, but security is the big one, I think.
Speaker:
00:14:17
With our on premise networks, there's a little bit of a safety blanket.
Speaker:
00:14:22
It's not all safe. Do you know what I mean? If someone wants
Speaker:
00:14:24
to get in, they can probably get in. But it's a lot easier
Speaker:
00:14:28
to deploy things knowing that you have things ring fenced and people won't
Speaker:
00:14:32
be able to exploit them. So security is a massive one,
Speaker:
00:14:36
because when you push it up into the cloud, it's so easy to
Speaker:
00:14:39
just make something public by mistake and things like that. So we do
Speaker:
00:14:44
spend a lot of time on security and making sure everything's encrypted in
Speaker:
00:14:49
transit and encrypted at rest and all this.
Speaker:
00:14:53
You do find though, a lot of the cloud services and all this
Speaker:
00:14:56
kind of happens by default, but you still gotta check it out and
Speaker:
00:14:59
make sure it's within your standards, if you like. So security is a
Speaker:
00:15:03
massive one. The other things to consider is governance, which changes drastically
Speaker:
00:15:10
as well when you're going to cloud because it's so easy for things
Speaker:
00:15:14
to get out of control. Because of the power of cloud,
Speaker:
00:15:17
essentially is something that I've led that I haven't learned before,
Speaker:
00:15:21
is that you can push date up into cloud and give everyone access,
Speaker:
00:15:25
but God, it becomes a massive mine field and it turns into the
Speaker:
00:15:29
Wild West if you're not careful. So you really have to basically keep
Speaker:
00:15:33
an eye on that governance, making sure that people have only got access
Speaker:
00:15:36
to the data that they need to do their day to day job.
Speaker:
00:15:39
And I think that's very important. It's always important on premise as well,
Speaker:
00:15:43
but with the power of the cloud and more people can do,
Speaker:
00:15:45
it can really get out of control so quickly.
Speaker:
00:15:49
There are a number of things that I consider, but one thing else
Speaker:
00:15:53
I'd like to say is that if you've got good governance and good
Speaker:
00:15:57
lineage of your current on premise systems, it becomes a lot easier to
Speaker:
00:16:04
go to cloud. Because you understand how the data is getting used.
Speaker:
00:16:08
It goes back to those personas again, who's using it, it's all documented
Speaker:
00:16:13
and things like that. In that case, it's a lot easier to move
Speaker:
00:16:16
and replicate and put those same controls in. So it's important to get
Speaker:
00:16:20
it right wherever you are. So yes, cloud does make a difference,
Speaker:
00:16:25
but you really should be putting these rules in place throughout,
Speaker:
00:16:29
no matter where you are. I think it's interesting you say that because
Speaker:
00:16:33
obviously there's great opportunities with cloud, the speed and
Speaker:
00:16:37
the openness, as you say, but yeah, a lesson learnt from something in
Speaker:
00:16:43
the cloud is probably a harder lesson than if it's actually still just
Speaker:
00:16:47
an internal system, a bit more closed down. Have you had to do
Speaker:
00:16:50
a lot of training or sessions with your team to get them thinking
Speaker:
00:16:55
differently with those points and those challenges? So there's always technical
Speaker:
00:16:59
training, that's a given, but like I said, it's having that engineering
Speaker:
00:17:04
mindset. Admiral's a very young company, we've got a lot of young people
Speaker:
00:17:09
working there, come straight out of university and we can train them up
Speaker:
00:17:13
and things like that. They need to understand the repercussions of things
Speaker:
00:17:17
going wrong. We have processes and things like that to control that,
Speaker:
00:17:21
but we'll generally give that insight into failing systems, a lot of the
Speaker:
00:17:28
time. Because you have to code to have resilience and make things robust.
Speaker:
00:17:33
So quite often we'll have reviews of code go, "Right, okay,
Speaker:
00:17:36
what happens if that thing goes wrong?" And it goes, "Oh, well,
Speaker:
00:17:39
it re tries." "Okay, alright, that's great. It re tries. But does it
Speaker:
00:17:43
clean up the message made when it went wrong?" It's things like that.
Speaker:
00:17:46
So it's having an engineering mindset. A lot of what I try to
Speaker:
00:17:50
put into context is that, look, would you put an aircraft in the
Speaker:
00:17:55
sky if you built it like that? What if something failed on the... Is
Speaker:
00:17:59
there a backup system? How does it get maintained, how does it get
Speaker:
00:18:02
checked? There's warning lights and things like that that you get?
Speaker:
00:18:06
Have you got the same warning lights and things in that in the
Speaker:
00:18:08
code and the data pipelines that you've put into this system?
Speaker:
00:18:12
That's why it's important to monitor, because everyone's brought up with
Speaker:
00:18:16
a CICD mentality, and a lot of CICD doesn't work when you have
Speaker:
00:18:24
a data project going on. And this is why you have data ops
Speaker:
00:18:27
instead of dev ops, because it's not only the code that has to
Speaker:
00:18:31
go through that process, it's the actual data itself. And you can get
Speaker:
00:18:34
bad data, so your code may be the best code written in the
Speaker:
00:18:38
world, but you get some bad data into it and everything can fall
Speaker:
00:18:43
down. And so you have to have these checks, you have to have
Speaker:
00:18:47
this live monitoring and things like that. And that's where I find machine
Speaker:
00:18:51
learning is actually coming into play, because you can actually apply machine
Speaker:
00:18:55
learning to that monitoring, to actually go, "Oh, this premium looks low
Speaker:
00:19:01
today," or on this percentage of measures or things like that.
Speaker:
00:19:05
You can actually get machine learning to actually invoke that alert and
Speaker:
00:19:08
then somebody can react to it and things like that. So although we
Speaker:
00:19:11
look at machine learning as, "Oh yes, gonna help this, gonna make... Better
Speaker:
00:19:15
profile my customers," and things like that, there's so many more applications
Speaker:
00:19:19
throughout the whole data pipeline, if you like, and that data flow,
Speaker:
00:19:24
that I think sometimes people ignore that. It's like turn the skills on
Speaker:
00:19:27
its head and use the machine learning to help make your systems as robust
Speaker:
00:19:32
and get your data as clean and as quality as possible.
Speaker:
00:19:37
I think it's interesting you say that, 'cause I was just about to
Speaker:
00:19:38
ask you about machine learning and ethics. 'Cause you'd mentioned that.
Speaker:
00:19:42
And I think there is that, yes, you can think about it from
Speaker:
00:19:45
a technology point of view and stuff moving very, very quickly,
Speaker:
00:19:47
but that role of a person in there just checking and putting in
Speaker:
00:19:51
the perimeters of kind of, "This is gonna be safe," or, "This is
Speaker:
00:19:55
ethical," or, "This is actually what we mean by good data,"
Speaker:
00:19:58
and actually the checks and the processes, is still very important.
Speaker:
00:20:01
Are there any particular trends or ways of working that you can really
Speaker:
00:20:05
see how that would jump start or that would really, really push a
Speaker:
00:20:09
different way of thinking? Well, there's a number of new ways of rolling
Speaker:
00:20:14
out data teams and rolling out data architectures and things like that,
Speaker:
00:20:17
one of them being mesh. There's this concept of a mesh architecture and
Speaker:
00:20:23
God, you really gotta be mature for some of this stuff,
Speaker:
00:20:26
and you really have to have those engineering principles embedded.
Speaker:
00:20:31
Because the whole concept of mesh is that you decentralise basically,
Speaker:
00:20:35
okay, and you farm that out to value streams and things like that.
Speaker:
00:20:38
Now, traditionally it's been like a data warehouse team and they've done
Speaker:
00:20:43
all your data and they've ingested it and things like that, but they
Speaker:
00:20:45
become a blocker, they become a blocker because they can only prioritise
Speaker:
00:20:49
so much, and someone over here in, I don't know, household insurance wants
Speaker:
00:20:55
to do some stuff and they can't. So what you try and do
Speaker:
00:20:59
is you try and farm out those capabilities to other value streams,
Speaker:
00:21:03
but that can be difficult unless you've got some blueprints for the central
Speaker:
00:21:08
governance because you want everything to work the same way
Speaker:
00:21:12
so you can maintain it. There's definitely some things that in the delivery
Speaker:
00:21:16
space that are being suggested. Everyone talks about mesh. I went to Big
Speaker:
00:21:20
Data London this year, and everyone was talking about mesh, but no one's
Speaker:
00:21:24
really nailed it. And it's the same thing with machine learning,
Speaker:
00:21:28
you get our pricing analyst thinking, "Right, okay, we've got a good machine
Speaker:
00:21:31
learning model here, we've trained it well. Right. We've got to get it
Speaker:
00:21:35
live." But because it's hooked into the whole quoting process and things
Speaker:
00:21:39
like that, you really gotta be careful that that is correct.
Speaker:
00:21:42
And so it can take a long time sometimes for these things to
Speaker:
00:21:46
get live, and this is where you have things like ML Ops springing
Speaker:
00:21:50
up. Okay? So it's not about the model itself and the machine learning
Speaker:
00:21:55
model itself, it's more about how quickly can you get that value into
Speaker:
00:21:59
the mainstream. And this is what ML Ops is giving us,
Speaker:
00:22:02
and essentially it's a set of governance rules really,
Speaker:
00:22:06
that I think data scientists aren't used to
Speaker:
00:22:09
dealing with that. Data scientists, they're in their own little world,
Speaker:
00:22:11
they're in a bubble and things like that. Now they have to think
Speaker:
00:22:15
in terms of engineering and what matters, "What if this model goes wrong?"
Speaker:
00:22:19
That sort of thing, they have to think about that. So everything has
Speaker:
00:22:22
to be thoroughly tested. And that takes time. So although you've got this
Speaker:
00:22:27
whole, "Oh yeah, ML Ops, AI," and things like that, you have to
Speaker:
00:22:31
determine the risk of certain things, and if you start
Speaker:
00:22:35
relying too much on it without having that whole framework around it that
Speaker:
00:22:40
gives you that robustness, you could be in a lot of trouble.
Speaker:
00:22:43
I was gonna say the kind of reporting and that, like you say,
Speaker:
00:22:46
having different people with different roles and responsibilities across
Speaker:
00:22:50
that sort of data pipeline, if you like, how do you go about
Speaker:
00:22:55
making that reporting and that kind of auditing and the reporting seamless
Speaker:
00:23:00
for everybody at the their worst? 'Cause it's such a big thing,
Speaker:
00:23:03
isn't it? That actually, you could break it down and make it so you're working
Speaker:
00:23:06
in pockets, but that actually would completely blow the overall name of
Speaker:
00:23:10
having this kind of thing. Have you got particularly with Admiral, ways
Speaker:
00:23:14
of working that you think you've nailed that? Or have there been questions
Speaker:
00:23:18
that you've along the way thought, "Actually, if we ask this with these
Speaker:
00:23:22
certain people, we are gonna have a better outcome at the end."?
Speaker:
00:23:25
I think, again, its mindset is a lot of it. So yes,
Speaker:
00:23:30
you can have these different teams and they're all working on different
Speaker:
00:23:32
things, but again, we go back to why.
Speaker:
00:23:35
Okay? Why you're doing this? You're an essential part of the cog...
Speaker:
00:23:39
An essential cog, sorry, not essential part. That is basically gonna deliver
Speaker:
00:23:43
this value at the end, and you are an essential part of it.
Speaker:
00:23:46
I think if people understand that, they generally
Speaker:
00:23:51
start to go out and communicate. And I think that's what it is,
Speaker:
00:23:54
it's all about good solid lines of communication. And obviously that's what
Speaker:
00:23:59
Agile gives you with the stand ups, your retrospectives and things like
Speaker:
00:24:02
that. This is what I'm heavily involved in, is that something is not
Speaker:
00:24:07
working, you change it. That's the the whole point, nothing's etched in
Speaker:
00:24:11
stone. If something's not performing or something went wrong at the end
Speaker:
00:24:15
of the last delivery or whatever, you change it, and communication is a
Speaker:
00:24:19
big factor in that. So when you have teams and you have them
Speaker:
00:24:23
in those silos and they're not communicating, you'll find that they're not
Speaker:
00:24:26
gonna line up at the end of your sprint or whatever your delivery
Speaker:
00:24:30
mechanism is, you have to change it. So
Speaker:
00:24:34
we get people to work together. That's the key. And communications is hard
Speaker:
00:24:39
when we're all working remote and things like that, but
Speaker:
00:24:42
it's still achievable, very much so. But it is about comms.
Speaker:
00:24:47
No one wants to come in and write something bad that doesn't work
Speaker:
00:24:50
at the end of it. But they didn't know, they were just sort
Speaker:
00:24:53
of carrying on. So it's important to get strong leaders to bring people
Speaker:
00:24:58
together and actually go, "This what we're doing. This is how we're doing
Speaker:
00:25:02
it. Everyone clear what they're doing?" "Yes." In the morning and stuff
Speaker:
00:25:06
like that. "If you're not, come and see me, we'll go and speak
Speaker:
00:25:08
to someone else," or whatever like that. That's kind of how it needs
Speaker:
00:25:11
to work. Within the data teams at Admiral, I think we've cracked that.
Speaker:
00:25:16
I think we have. We try to modularise as much as we can,
Speaker:
00:25:20
so somebody doesn't write a code for a specific things, they write it
Speaker:
00:25:25
for many things that can basically be used by another team.
Speaker:
00:25:29
So we try and write these modules and frameworks
Speaker:
00:25:33
for people to work in, rather than, "I'm just writing code." And I
Speaker:
00:25:37
think that's what helps. But there's also a lot of unit test checking
Speaker:
00:25:41
and things like that, there's a lot of automation that you can basically
Speaker:
00:25:45
put in, and we do this at our retros and things are like that. We went,
Speaker:
00:25:49
"Right, that went wrong, didn't it last time? Okay, what we're gonna do
Speaker:
00:25:52
about it?" "Right, we put an automated check in for that."
Speaker:
00:25:54
So as soon as someone deploys something into a test environment,
Speaker:
00:25:58
runs a host of unit test scripts or things are that and goes,
Speaker:
00:26:03
"No, that ain't gonna work with this," so we can build that in.
Speaker:
00:26:07
And we quite often do that. Don't capture everything. 'Cause it is hard.
Speaker:
00:26:11
Data is hard, I'll tell you that now. But if you have that
Speaker:
00:26:14
mindset and that ways of working, and that monitoring mindset and that robustness,
Speaker:
00:26:18
that engineering mindset, you can overcome it. Do you think that the governance
Speaker:
00:26:24
is kind of more stable or more solid because of that understanding of
Speaker:
00:26:28
data engineering mindset? With the governance? Yes, I think with governance
Speaker:
00:26:33
we've been in a better place than Admiral has ever been.
Speaker:
00:26:37
We bake things in, and again, our engineers think about it because they
Speaker:
00:26:43
want the best kudos for their bit of code.
Speaker:
00:26:47
So the only way you're gonna do that is get people actually using
Speaker:
00:26:51
it and using what the output of that code has produced.
Speaker:
00:26:55
Now, if you've got good governance that has put that into a data
Speaker:
00:26:58
catalog 'cause you've coded that up or is automatic or whatever,
Speaker:
00:27:02
then people will find that data and they'll use it. Okay? And this
Speaker:
00:27:06
is another thing about, probably dovetails into the working in silos.
Speaker:
00:27:11
You develop something for a specific thing, I know it has to happen
Speaker:
00:27:14
sometimes, but if you can write it so that it's open ended and
Speaker:
00:27:19
basically other people from outside the intended recipient can see it in
Speaker:
00:27:24
the data catalog, and within reason, they're allowed to see it and all
Speaker:
00:27:28
that. There's security and stuff like that. They can discover through a
Speaker:
00:27:31
data catalog and basically go, "Oh, I didn't realise someone had done that,
Speaker:
00:27:36
calculated that over there. I'll go and use that." And then they attribute
Speaker:
00:27:40
to the data catalog themselves by saying, "I am using this for this
Speaker:
00:27:44
purpose." That kind of mentality then grows, because everyone used to see
Speaker:
00:27:50
data governance is like a blocker, but essentially there's ways and means
Speaker:
00:27:55
of doing it where you can bake it into that whole delivery process
Speaker:
00:28:00
and actually make it an enabler and not a blocker. I think it
Speaker:
00:28:04
took a lot of companies a long time to understand that,
Speaker:
00:28:08
because I think they just thought of it is a regulatory thing and,
Speaker:
00:28:12
"Oh we've got to do this, we're ticking boxes here and there,"
Speaker:
00:28:14
and stuff like that, but I think you do it properly and it
Speaker:
00:28:18
is so valuable and it can aid operational efficiency. People have gotten
Speaker:
00:28:24
better decision making because the quality of the data has increased.
Speaker:
00:28:28
I used to go into meetings and some guy used to go,
Speaker:
00:28:31
"Well, this is what we sold this product for yesterday," and another guy
Speaker:
00:28:34
would go, "Well that don't match my figures." And you're like,
Speaker:
00:28:36
"Well, who's right?" And now with this data governance you've got a gold
Speaker:
00:28:40
standard, and now you got the guy going, "I got these figures."
Speaker:
00:28:43
"Oh, well I got these figures and they're different, but mine
Speaker:
00:28:46
have a tick on them, okay? They've been checked, so they've been gold
Speaker:
00:28:50
standard, this is actually what happened." The other guy's got no response
Speaker:
00:28:53
against that. And this is the type of things that it can enable,
Speaker:
00:28:58
but I think it takes a while for people to see that because
Speaker:
00:29:01
they do see it still as a blocker. But if you know data an you
Speaker:
00:29:05
know how it's used, you can really see the advantages of having that
Speaker:
00:29:09
strong governance. And I think what you're saying there as well,
Speaker:
00:29:13
is that actually just the quality of the reporting across an organisation
Speaker:
00:29:18
and also out to customers, will be better because you know that you're
Speaker:
00:29:22
using correct data and you're using it in the right way,
Speaker:
00:29:25
and it is standardised. It's not just the reporting, you look at the
Speaker:
00:29:28
AI and the machine learning and the data scientists and things like that,
Speaker:
00:29:31
it used to be they'd have this, "I just want all data. All
Speaker:
00:29:35
the time, as quick as you can, all data, all format, and I
Speaker:
00:29:39
want as much data as you can give me," and things like that. I
Speaker:
00:29:42
think the emphasis is slowly drifting away from that now to actually go,
Speaker:
00:29:47
"No, I want quality data." Because if you're training your models on bad
Speaker:
00:29:52
data, you're gonna get bad decisions, okay? So I think the emphasis has
Speaker:
00:29:57
now moved from, "Give me as much data as you can,"
Speaker:
00:30:02
to, "Just give me clean data." And I think that's an important step
Speaker:
00:30:05
because I don't think I could say that to you two years ago,
Speaker:
00:30:09
but I think now there's a machine learning model for everything,
Speaker:
00:30:12
I don't think anyone's really pushing the boundaries of machine learning.
Speaker:
00:30:16
Yeah, don't get me wrong, at Admiral we have custom ones and things
Speaker:
00:30:19
like that, and probably data scientists if they heard this in Admiral they'd
Speaker:
00:30:22
probably have a few choice words to me,
Speaker:
00:30:25
but to be honest, someone else has done it. Someone else has done
Speaker:
00:30:30
something very similar somewhere. You can pretty much buy a model off the
Speaker:
00:30:34
shelf, okay? And everyone is doing it, it's all like self service.
Speaker:
00:30:37
I can go on to Google's marketplace and I can buy
Speaker:
00:30:41
load of models and get the result I want.
Speaker:
00:30:44
What the difficult thing is the data engineering to get the data quality
Speaker:
00:30:47
up to a level when you can train that model, and I think
Speaker:
00:30:51
that's where the emphasis has changed there as well. So it's not just
Speaker:
00:30:53
the reporting, it's definitely for other applications as well.
Speaker:
00:30:58
So with the teams that you've got there, it sounds like obviously there's
Speaker:
00:31:01
a sort of sharing of knowledge and sharing of code, and you have
Speaker:
00:31:05
obviously done a project and read all of... Been involved with it. What's
Speaker:
00:31:09
it been like working with external consultants? Has that challenged the
Speaker:
00:31:12
thinking? Has it kind of steered the project?
Speaker:
00:31:16
You definitely learn from them. Because at the end of the day we're still
Speaker:
00:31:19
quite immature in some ways, Admiral's cloud journey.
Speaker:
00:31:23
Things take a long time and things like that. And you need that
Speaker:
00:31:25
external support, you need good people who have basically done this before
Speaker:
00:31:30
and know what the pitfalls are. You're still in charge, but you're in
Speaker:
00:31:34
charge of the outcomes, "These are the outcomes I have. Okay?
Speaker:
00:31:37
It needs to work like this." People like Red Olive as well, they come
Speaker:
00:31:40
in, they're classic engineers, they're trained in first principles.
Speaker:
00:31:45
And they will go, "Well James, if you do it like that,
Speaker:
00:31:47
then that could break there." So that's great advice. And so they're thinking
Speaker:
00:31:51
the same way, which is fantastic. You don't have to educate them in
Speaker:
00:31:56
robustness and maintenance and things like that. You just learn a lot from
Speaker:
00:31:59
them. You go, "Oh, okay. Well, we did some similar for the couple
Speaker:
00:32:04
of guys two years ago that did this. Would that work?"
Speaker:
00:32:09
And it's just bringing that experience to it. Because cloud is an experience
Speaker:
00:32:13
and you really have to sort of think differently. And especially when you're
Speaker:
00:32:17
on prem, you design for performance and when you get to cloud,
Speaker:
00:32:21
you're actually designing for cost. Because it's so easy to just throw another
Speaker:
00:32:26
100 processes at it and a shed load of memory, but there's a
Speaker:
00:32:31
cost implicated with that. So you still gotta be super efficient,
Speaker:
00:32:35
because otherwise those costs can run out of control and things like that.
Speaker:
00:32:39
And it's good that our external consultants and things like that, they've
Speaker:
00:32:41
all fed into that. They've gone, "Well, you can do it this way,
Speaker:
00:32:44
it's robust, it's really good, but it will cost you this and you
Speaker:
00:32:47
have to figure that out," and that can be very difficult.
Speaker:
00:32:50
But having people that have done it and things are like that,
Speaker:
00:32:53
that can really guide you essentially on that design from a cost perspective.
Speaker:
00:32:59
And also from a robustness perspective. It's a balance, isn't it?
Speaker:
00:33:02
So yeah, they've been really good. Well, though we've been talking data,
Speaker:
00:33:05
we've actually been talking about communications and people throughout this.
Speaker:
00:33:09
For people who are now coming into the data industry, what
Speaker:
00:33:14
skills would you recommend that they have? And what kind of things would
Speaker:
00:33:17
you be looking for for somebody joining industry now?
Speaker:
00:33:20
To be honest, we take on a lot of grads, a lot of
Speaker:
00:33:24
young people that may want to get into this sort of thing.
Speaker:
00:33:27
The key skill or two key skills I'll put down from a technical
Speaker:
00:33:31
perspective, SQL and Python. Because pretty much everything runs off those
Speaker:
00:33:37
two things. Or is interfaced with them in some means.
Speaker:
00:33:42
But what we really look for is... You can teach those skills.
Speaker:
00:33:46
If you've got a bit of a mathematics or science background or things
Speaker:
00:33:51
like, even if you haven't. But you know how an Excel spreadsheet works,
Speaker:
00:33:55
you know how to look through it, you know to organise data and
Speaker:
00:33:58
things like that, those are the key sort of technical skills we look
Speaker:
00:34:02
for. But we look for behaviours, and we do a lot of competency
Speaker:
00:34:07
based type interviews where we look at, "Give us an example of a
Speaker:
00:34:12
time when something went wrong?" And that may not be in data,
Speaker:
00:34:15
or the business could be something completely external, but we look at the
Speaker:
00:34:18
mindset about how people handle it. So there's a little bit of psychology
Speaker:
00:34:23
in there and things like that, but what we're looking for is people
Speaker:
00:34:26
who are conscientious, who care about what they're actually doing, and that's
Speaker:
00:34:31
important. So all the technical stuff, you can learn, you could go on
Speaker:
00:34:35
100 YouTubes there, right? And you can know concepts better than us.
Speaker:
00:34:39
But when it actually comes to implementing it and your day to day
Speaker:
00:34:43
job, I think your psychology and how you're made up has a lot
Speaker:
00:34:47
to do with it. So we do look for a lot of people
Speaker:
00:34:50
who are very driven, as I said conscientious, really care about what they're
Speaker:
00:34:54
doing, and you can make a really good start to mould them into
Speaker:
00:34:59
a classic data engineer. So that's what we look for.
Speaker:
00:35:03
My last question for you, what is the most exciting thing that you're
Speaker:
00:35:07
most passionate about for the data, pushing forward into the future?
Speaker:
00:35:12
Data automation. Because I know we talked about machine learning and AI
Speaker:
00:35:17
and things like that, and those things are great and very impressive and
Speaker:
00:35:21
things like that, but I think unless you're one of the top players,
Speaker:
00:35:26
they can be very difficult to implement on scale.
Speaker:
00:35:29
But what I'm interested in at the moment is a lot of metadata
Speaker:
00:35:34
sort of data driven pipelines, whereby, rather than deploying code you're
Speaker:
00:35:39
making more configuration changes and things like that. So it's the automation
Speaker:
00:35:43
of it all from a delivery point of view,
Speaker:
00:35:46
I like the idea of. Some people have had a go,
Speaker:
00:35:49
WhereScape being one of them, they're really good, but there needs to be
Speaker:
00:35:53
a next level. So when we're talking about... We've spoken about things like
Speaker:
00:35:57
governance, data cataloging and things like that, that type of thing needs
Speaker:
00:36:01
to happen automatically. So we talk about automation in terms of automated
Speaker:
00:36:06
testing or automated deployments and things like that, but what about actually
Speaker:
00:36:10
automating the actual development? Because a couple of years ago, a lot
Speaker:
00:36:14
of people just hand coding things, and there's a lot of now low
Speaker:
00:36:19
code solutions where you still need to have the technical ability and know
Speaker:
00:36:23
what's going on underneath, but you can template a lot of it.
Speaker:
00:36:26
So I can automatically template that it adds, I'm bringing new data in,
Speaker:
00:36:30
it adds to this data catalog. Or I publish a report, it automatically
Speaker:
00:36:34
gets added to this data catalog and things like that. So that's what I'm
Speaker:
00:36:37
really interested in, because everyone goes after the really exciting stuff
Speaker:
00:36:42
and things like that, but nobody thinks about the implementation so much.
Speaker:
00:36:46
Because I live in that space and all my problems have been in
Speaker:
00:36:50
that space, that's what I kind of get excited about now,
Speaker:
00:36:52
can I have a team of three people that can deliver something in
Speaker:
00:36:58
say a month, rather than a team of 10 deliver it in six
Speaker:
00:37:03
months? And that's kind of where I'm going at the moment,
Speaker:
00:37:07
because we're up to our eyeballs at work and it never stops.
Speaker:
00:37:11
And you can only grow so much before it becomes hard to manage.
Speaker:
00:37:15
But if we can deliver quicker and automate more and maintain our quality
Speaker:
00:37:21
standards and our governance standards and things like that, that's what
Speaker:
00:37:24
excites me. Now, that may be a bit of a boring answer,
Speaker:
00:37:27
but that's the honest truth. A fascinating take from James on the importance
Speaker:
00:37:32
of people, teamwork and communications in data projects. Join us for the
Speaker:
00:37:36
next episode of Fibonacci, the Red Olive Data Podcast, where we'll be joined
Speaker:
00:37:40
by another data expert sharing their thoughts on latest trends in AI and
Speaker:
00:37:43
big data, along with some great hints and tips.
Speaker:
00:37:46
Make sure you subscribe to Fibonacci, the Red Olive Data Podcast from wherever
Speaker:
00:37:50
you get your podcast, to make sure you don't miss it.
Speaker:
00:37:53
That's all for today. Thanks for listening. I've been your host,
