April 22, 2025 • 28 mins
Peter Laflin, Chief Data Officer at Morrisons, shares how his team turned customer confusion into a cutting-edge vector search experience—bridging physical retail with AI-powered search. He and John Kutay dive into the practical challenges of implementing LLMs and real-time data pipelines at scale, the importance of starting with actual customer problems, and why the best engineering feels a little lazy (on purpose). A real-world look at what happens when modern search meets supermarket shelves.
What's New In Data is a data thought leadership series hosted by John Kutay who leads data and products at Striim. What's New In Data hosts industry practitioners to discuss latest trends, common patterns for real world data patterns, and analytics success stories.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:05):
Welcome to what's New in Data.
I'm your host, john Coutet.
Today's guest is Peter Laughlin, chief Data Officer at
Morrisons, one of the UK'slargest grocery chains, serving
millions of customers each weekacross nearly 500 stores.
Under Peter's leadership,morrisons has been recognized
for its innovative use of dataand generative artificial
(00:26):
intelligence to transform retailoperations and drive customer
loyalty.
He does it while partneringwith leading technologies like
Google, cloud and Stream.
Let's dive right in.
Speaker 2 (00:38):
Peter, how has real-time data improved your
operations?
Data improved your operations.
Speaker 3 (00:45):
We have a huge amount of data that we generate every
second, every minute, every hourof the day, not least from the
transactions that we have withour customers.
So, whether it's in our store,whether it's online, whether
it's through one of our partnerslike Just Eat or Deliveroo, we
are collecting data points thattell us what people have bought,
(01:06):
where they bought it and,crucially, how much money we
have charged them.
Now it's really importantrunning a business that we can
collate that data really quicklyso that we can take decisions
on that information, and so thefirst horizon really for the
real-time data was to make surethat we had up-to-the-minute
(01:28):
sales information, even slightlylonger, to run batch processes
to collect all that informationfrom the tills, do the
(01:50):
calculations, do all theaggregation and then present
that back into a form that wecan read and look at.
But we have a number ofpipelines in the business that
are collecting thosetransactions together literally
in real time.
So, as we've been speaking inthe last two minutes, I could go
to our data set and see who'sbought something within the last
two minutes.
What was it?
Has there been an increase in aparticular type of product?
(02:12):
Are we seeing particular demandin a certain part of the
country and, crucially, in aworld of technology where we
really care about making sure wehave uptime, being able to use
a stream of data to help sensethe health of the network is
also quite important Because, aswe are processing all this
(02:35):
information, we have aparticular expectation about how
much we'd see every minute, howmuch we might see every five
minutes, ten minutes, so on andso forth.
But being able to stream thatdata in real time means that
actually we can pick up on onissues around us as soon as they
happen, rather than having towait for a few hours to realize
that maybe there's a problemwith a till somewhere or maybe
(02:57):
there's a problem with one ofour pipelines.
So actually the real-time datastarts to mean that we can make
real-time sales decisions withina few hours.
Do we need to think differentlyabout promotion?
Do we need to order more stockbecause the promotion is going
even better than we thought itwas?
Other parts of the country, andcertainly recently in the UK,
(03:18):
we've had a number of big stormevents where we did have to
close a few shops for a fewhours because the weather was so
bad that the government advicewas people had to stay home and
not go out.
In that context, we have thatreal-time information to
understand just how ourcustomers are interacting with
(03:40):
us, to make sure that we canbest service them, to make sure
that we can best service them,and so that real-time data is
super critical in helping keepour finger on the pulse of
what's happening across ourphysical stores and online.
Retail is about detail, butretail is also about what's
happening now and what mighthappen in the next half an hour,
(04:00):
the next hour, the next twohours, and so real-time data is
super critical in that retailenvironment, because the numbers
are so large that you don'twant to get to the end of the
day and feel like you were eventwo or 3% away from where you
wanted to be, because that canbe a material amount of money.
Speaker 2 (04:17):
Being able to be really close to understanding
where we are at any point intime is super important in terms
of how you can run the businessabsolutely, and the way you've
deployed it is remarkable in howmorrison's has optimized that
for your internal operations andhow you're able to respond to
(04:38):
market changes and and eventhings like the weather and
stock and things along thoselines.
And that's what's beenincredible about partnering with
your team, peter, which is justthe obsession with customer
success and making sure thatyour shoppers are having the
best possible experience and theway.
(05:01):
From Stream's perspective,you're using Stream Cloud
multiple services and you havethis vast, complex supply chain
that we're pooling you know dataliterally from different
physical locations and thenmassive volumes and ultimately
applying that into theseapplications.
I think what's made that sosuccessful.
(05:22):
Again, you know, yeah, thestreaming infrastructure is
great.
The you know, the use of, youknow vector databases is really
innovative and it all comes backto the fact that you've applied
it in a way that really helpsthe customers in the business.
And you know I know there's alot of data practitioners and
data engineers listening to this, uh who are working on other
(05:45):
amazing data pipelines.
But just to to reaffirm justhow important it is to make sure
that the pipelines you'rebuilding are are feeding the the
customers and the business andthe people who are making those
operational decisions, andthat's that's really who you
have to focus on, right, so Idid.
You know, morrison's is a verysophisticated operation with you
(06:06):
know, thousands of farmers andsuppliers and growers, and it's
unique in how you're a grocerychain but you also control that
kind of end-to-end supply chainfor the food.
I'd love to hear how you'vebuilt the data infrastructure to
handle all you've built thedata infrastructure to handle
(06:27):
all.
Speaker 3 (06:27):
I think the key is to have a small number of patterns
that you can replicate, butalso recognizing that there's a
huge amount of complexity thatneeds to be simplified.
So the way I tend to thinkabout it is that we have a super
highway I think is probablywhat you call them in the US,
(06:48):
isn't it?
Or highway, at least we havethese sort of big multi-lane
roads that all go in the samedirection and broadly they go
quite fast.
Whether they do in practice ornot is probably a different
point.
Let's not get bogged down intraffic science, although
there's a lot of reallyinteresting science in traffic
science but my point is you needa big pipe that is capable of
(07:11):
transferring a lot of dataincredibly quickly, and, and
what you need is lots ofprocesses that are catching the
data.
But, crucially, the data thatyou catch might be in a slightly
different form to whether it'scoming from a manufacturing site
, whether it's coming from asupplier, whether it's coming
from a store, and so one of thebig areas of complexity that you
(07:35):
could find is if we were tobuild individual pipelines for
all of those, you would find youwere doing a lot of bespoke
work and you were doing a lot ofwork to just translate and map
things doing a lot of bespokework, and you were doing a lot
of work to to just translate andmap things.
And the way that we've tried tobuild things is is to recognize
that you need to do and I meanthis in a really nice way, you
need to do the least amount ofwork possible to give you the
(07:56):
most amount of flexibility.
Um across your, your sort ofyour data infrastructure, and
therefore one of the things thatum was always really high up
our list of requirements wasthis idea of how do you catch
the data and know where it needsto go, because actually I could
(08:16):
catch the data, I could spendtime thinking about where it's
going and then I could send iton its way.
But actually one of the thingsI like about where it's going
and then I could send it on itsway, but actually one of the
things I like about where thetechnology is now going is that
we can catch the data and we canworry about where it's going
when it's on its way, becausethat cuts down the time it takes
to do this process.
(08:37):
It improves the opportunity forlower latency, but it also
means I've got less work to dowhen I plug new data sets in,
because if you have this sort ofparadigm that says I've got a
pipe that goes as fast aspossible and it's incredibly
intelligent so it can kind offigure out where things are
going in the pipe and all I needto do is connect the data to
(09:01):
the pipe and then make sure thepipe knows where it needs to
ultimately get to at the otherend.
Now I hope that's not too toomuch of an abstract way of
answering the question, but butreally the the fundamental in
the architecture has been we.
If we built bespoke pipelines,every time we would need an army
of data engineers and we wouldprobably never get through the
(09:22):
work.
So what we've had to do isrecognize that you have to build
something that is flexibleenough but intelligent enough to
mean that our engineering teamcan be a bit lazy.
Um, and they're not lazy.
You know they're all incrediblyhard working and there's always
far too much to do.
But you almost have to set outto make your teams feel like
(09:43):
they are being lazy, because bydoing that you're creating the
right kind of behaviors in thesort of mindset for your
pipelines and, like I say, Imean it's maybe not about
getting into the whys andwherefores of how you can run AI
on the data as it's streaming,but it is about that concept of
ensuring that you can get thingson its way and then eventually
(10:05):
it'll figure out where it needsto go in an intelligent way,
because that makes things veryquick.
Speaker 2 (10:12):
Absolutely, and I love the way you phrase it
because, honestly, the bestengineers set themselves up so
that they can be kind of lazyabout to how Morrison's able to
actually innovate so quickly,given the scale that you're
running at, which is reallyincredible.
And the other thing you saidearlier in the episode was that
(10:52):
retail is detail and you've runMorrison's now, which is an
operation that runs atincredible scale, has, you know,
full supply chain visibility,full shopper visibility, all the
way from.
You know the food beingproduced, so when it's purchased
, and then you know being loyalto the customers and continue to
(11:13):
provide a great experience forthem.
So what's your advice to otherleaders in retail who are
embarking on a similar journeyof modernization and innovation
with AI?
Speaker 3 (11:25):
Start with your customer and be very clear on
why you are proposing whatyou're proposing.
It's actually quite hard to getunderneath the why?
Question because, again, if youlisten to your customers, they
will.
They will give you theirfeedback, they will give you a
very clear steer in terms of howthey feel, what they think is
(11:46):
working well, what they think isnot working so well.
There's an art form, though, inbeing able to interpret that
feedback and build a program ofwork that is able to talk to all
of those um opportunities butdo the right work in the right
way, at the right pace to um toactually deliver against those
(12:08):
those opportunities.
Um, I mean, it's always the uhthe henry ford example that
comes to mind around.
You know, if you ask peoplewhat they want, they wanted a
faster horse back in the uh, theearly, early 20th century.
What they actually needed was acar, and so you do have to make
sure that you're you're notover innovating, you're not over
complicating um.
You know our, our leadershipteam in morrison's, often talks
(12:31):
about complexity and simplicity.
We have a very complex business, but our job is to deliver that
as simply as possible, and sobe really quite clear on the
problem you're trying to solve,because what fascinates me is,
in a room full of technologistsand a room full of incredibly
clever data engineers, dataanalysis experts, data
(12:53):
scientists even with the samequestion, even with the same
observations from our customers,the solutions can be very
wildly different.
And therefore, keep asking whyuntil you get to the point where
you can absolutely crystallizeI'm going to do this, we're
going to do that because it'sgoing to deliver this for our
(13:14):
customers, and I think you'vegot to build your strategy
around that.
It might be that real-time datais super important for you and
your customers.
It might not be.
It depends on your product, theway you set your business up,
your customer value propositionand, equally, the use of ai to
automate various things.
There might be obvious thingsthat you'd like to do because
(13:37):
your competitors have done it,but challenge yourself to say
why do we do that?
Because we want to, or are wecopying?
And and I think, asking thatreally detailed why question can
be really quite hard and, attimes, can feel quite uh.
It feels like you might slowyourself down, but in a world
(13:59):
where you need to go fast, itactually speeds you up because
at the point that you then startto deliver something.
You're super clear on why andwhen it comes to then building
the requirements and talking tothe vendors and connecting all
the technologies together, youcan be really very focused on
what you need.
Now I think you know, if youlook at the, the way that we
sort of started to work withstream, I think we were very
(14:21):
clear on what we needed and wewere very clear that we needed a
capability that you were ableto provide.
Speaker 2 (14:27):
And I think you know, when you're setting these
things up and working aboutwhere you go in the future and I
, I think, being super focusedon why, I'm super curious on why
you think that that is the mostimportant part, I think yeah,
and that's that's one of thethings that has been really
(14:48):
incredible about partnering withwith morrison's and you know,
uh, you know working with you onyour, on your uh implementation
and rollout, because even whenwe're looking at streaming
real-time pipelines, we benefitfrom knowing the business use
case.
Why?
Because we're going to tune thepipelines accordingly.
And you might say someone mightsay, hey, our use case is just
(15:12):
having some reports that loadnightly.
And we're using stream becausewe want to do low impact change
data capture so we're notpulling the databases and making
that an expensive operationagainst the production databases
and that's fine.
So we'll do the real timechange data capture but we'll
batch and load the data intoyour warehouse, you know, every
24 hours, so you're saving thecost there.
(15:34):
But when customers say you knowwe want to improve the customer
experience by responding toreal time signals, that's when
we really say okay, then let's,let's go full throttle and help
you optimize those pipelines.
And you know it's been a greatpartnership and great teamwork.
You know from from the, the,the folks here at stream that
(15:55):
were working with Morrisons andthen the counterparts at
Morrisons who are reallydedicated to building out this
infrastructure.
But I think it really does allcome back to your vision and
your leadership for what youwanted to accomplish, which made
it easier and more focused forStream and I'm sure Google Cloud
as well to deliver what you'relooking for, because, ultimately
(16:18):
, when people know what theywant, people will rally around
them and help them get there.
So that can apply to all datateams the more the data team
knows about the businessinitiative and what they're
really trying to accomplish andreally who their customers
Because all data teams, whetherthey know it or not, they have a
(16:39):
customer.
It might be an internal customer, but there's someone that's you
know paying with their time anddedication and you know their
precious resources to work withthe data team in one way or
another.
And data teams just have toknow who that customer is and
why they're delivering.
It could be like an endcustomer that's actually buying
(17:01):
a product from the business, orit could be the finance team
right, they could be doingrevenue recognition with that
data or it could be someone inthe sales team is doing lead
routing with it.
But ultimately it all comesback to, peter, like you
mentioned, just reallyunderstanding the business
problems and how you're solvingthem.
So that's what's been reallyfun about working with
(17:24):
Morrison's, and now you'reseeing all these incredible
benefits and the customers arehappy and the financial metrics
all look great, and I thinkthat's certainly no coincidence.
And, peter, I also wanted toask you there's so much changing
in the industry as well.
(17:44):
Just with all the innovationthat's going on, what emerging
trends in data and AI are youmost excited about?
Speaker 3 (17:54):
I'm almost exhausted, to the point where it's hard to
get excited and I'm joking,actually.
I think there's so much goingon, um that it's hard not to be
excited, um, I.
I think that the challenge ishow do you deliver on the
opportunity?
Because as every week goes by,I think the opportunities get
(18:15):
bigger and I think the hype getsbigger.
If you go back 10, 15 years,there'd be new technologies that
would emerge and you'd have towork really hard to go to your
exec and say, on full circle, wenow have board members asking
(18:39):
for ai and asking for newdevelopments, because these
developments are now mainstreamto the point where I don't think
there's anybody who hasn'theard, either directly or
indirectly, of a large languagemodel um or a chat gpt, um, you
know so.
So I think the the really thereally exciting bit, but
actually the big risk, is how doyou realize that the
opportunity?
Because it's clear now theopportunities are massive and
(19:01):
the expectations of ourstakeholders are that we will
get our fair share of thoseopportunities.
But how do you do it safely?
How do you do it in a worldwhere people are learning to
walk?
So you know, we've gone throughvarious paradigms, haven't we?
You know, if you look at the,the mid to late 90s, it was
about how you enable computingin a business.
(19:21):
And then the world of the youknow the, the it sector as we
know it then sort of grew aroundthat, um, you know, data came
probably 10 years later and nowwe have generative ai and sort
of broader ai systems that arecoming through and so we're
going through probably that, youknow, maybe second in my mind,
third revolution you know thesetechnologies are now as
(19:44):
fundamental a change as it wasto start to use a computer in
your day job.
I remember one of the thingsthat inspired me to get into
this world was um.
One of the things that inspiredme, um, to get into this world
was um.
I remember my.
My grandfather was, uh,involved in in putting computers
into schools in the mid 80s andhe was massively inspired by
(20:05):
the bit, but he retired at apoint where he never got to use
a computer at work and I dothink the next generation that
comes through.
So if I look at my children, um, they're probably, you know,
the first generation that willgo into the world of work with
an expectation that they do useAI routinely in what they do,
(20:27):
and so this has been a hugecultural societal expectation of
shift.
It's massively exciting, but itis incredibly scary, because how
do you, how do you, keepcontrol of something which,
inherently, is uncontrollable ifyou want to go after the value?
And when I say it'suncontrollable, people are going
(20:49):
to find their own ways to usethis.
So even if we put variousregulatory frameworks in place
in the business to say, pleasedo this, but not this, I'm sure
people will find ways around it,whether accidentally or
otherwise, and thereforeactually the governance of this
is the thing that I'm probablymost concerned about.
But if we get the governanceright, I think it is so exciting
(21:11):
that things that took days,weeks, months to do will be done
in seconds.
And what does that do for ourunderstanding as a, as a society
, as a group of humans lookingfor ways to continue to live
sustainably and effectively on aplanet?
Um, and therefore I, I, you know, I think how big do you go here
(21:34):
?
But um, ultimately, I thinkthis is just another stepping
stone into increasing ourcollective intelligence in such
a way that we can then start totackle all those questions that
we still haven't solved, such aswhere do numbers come from and
how do we tackle global warmingand how do we cure cancer
(21:54):
forever, and actually there's somany really valuable and
exciting things that we could dowith the technology.
Um, but actually right now wejust have to figure out how to
use it effectively in a businessto to mean that we don't break
anything.
Uh, because if there's a bigbang where something goes wrong
with generative ai, it mightstop that.
(22:15):
That development of technology,uh, in a way that probably
would be would be bad forsociety but probably better for
everybody in the short term.
But I'm rambling now slightly.
Speaker 2 (22:27):
I mean, yeah, those are all you know.
Yeah, like you said, the nextgeneration is going to use AI.
That's going to be theirexpectation.
That's how they're going towork with everything.
They're just going to chat withsome robot, you know, some
robot-like thing that justunderstands what they're trying
to say, can derive the semanticmeaning from anything.
(22:49):
And you know, I think you putthat, you phrased that very
eloquently, peter, and you knowthat's the world we have to plan
for now and it is sort of arace.
And because you know companiesare trying to outpace their
competitors into who can rollout the best AI enabled service.
Because you know, chat GPTalready has tens of millions of
(23:09):
users, right, so people arealready just chatting with
things and, uh, you know,getting getting the responses
they want, uh, so it's such anexciting time.
And that, tying this back towhat you've already delivered,
right, ai driven search, right,it really sounds like you know,
you, you've kind of understood,so, the essence of the value of
(23:29):
ai, which is kind of this, this,this, this buzziness, this
little, you know, ambiguity ofthe way people describe things
and knowing what they mean andgiving them what they want based
on that.
So it's it's, it's really coolto see what you've already
delivered there, with the powerof getting the real-time
inventory data doing the vectorembeddings, surfacing that as an
(23:54):
AI-driven customer experience.
I just think that's so exciting.
I mean, that's really whatinterests me.
You can build all the AIinfrastructure you want, but
really what's most impressive ishow teams like yours have
actually rolled it out for thecustomer.
Speaker 3 (24:11):
You need to be left-brained and right-brained
at the same time, because we'removing now to a world where the
technology gives you a platformto be creative.
And actually the exciting thingis how do you use this
commoditized capability?
Because it'll all become it'llall effectively become
(24:32):
commoditized.
Everybody's ai out of the boxwill probably be about the same
as everybody else's, because,even if it's not, everyone will
catch up.
So let's just assume you knowthe base ai becomes kind of you
know broadly a constant factor.
How do you then build that intoother processes in such that you
are creating something new andvaluable and that is a creative
(24:55):
process.
And therefore we um, we'reshifting um from being a
knowledge-based tech communityto being a curiosity-based tech
community, and actually theshift is going to be knowledge
can be replaced or found in alarge language model.
(25:16):
The insight and the spark tothink of something new, to apply
it to, will become the valuableresource in the future.
But it's all underpinned by thedata and the pipelines.
So I mean it's fascinatingwhere it's going to go.
I would love to have thisconversation again in five years
, john, to see where we'veactually got to um.
(25:38):
But you know, super exciting,bit scary um, but the journey is
what's fun uh, yeah, it's, itis so much fun.
Speaker 2 (25:47):
I mean it is.
You know, it does remind mewhen, you know, uh, mobile apps
on the iphone and the app storekind of blew up and there's just
this, this kind of blank slateof innovation that we know is
going to happen.
And you know, there was sort ofa similar cadence with data,
(26:07):
but now with AI, it's just soclear that there's going to be
just a new generation ofapplications and use cases and
ways this can be applied.
Well, you know, peter, you know, at Stream, we've been so
delighted to work with you onthis five-year journey with
Morrison's.
I want to ask you about this alittle selfishly, but what is it
about Stream and our peoplethat really made the difference
(26:28):
for you?
Speaker 3 (26:31):
You worked with us really collaboratively and you
co-created the solution with us.
So I talk a lot about thedifference between collaboration
and co-creation with us.
So talk a lot about thedifference between collaboration
and co-creation.
Um, co-creation is the good oneand what I'm looking for, um,
which is where we kind of alljump in.
We're one team, we're clear onthe goal and we work together.
I think, um, we often joked inin the project meetings that it
(26:54):
was difficult to tell um thestream team from the the
morrison's team, because it youjust worked with us really
really well.
Everybody just knew what wewere trying to do.
It was hard yards, it wasdifficult stuff.
I mean, I think it it can bequite challenging to sit back
from this and go.
We did a, we did a really bigthing, um, and we did it really
(27:15):
well, and the reason we did itreally well is because you guys
support us brilliantly and youknow, I think, think the real
difference was you havetechnology, but you have some
fantastic people and workingwith you was easy, and so there
were lots of real positivesthere, because it's really
important that you can actuallyembed the opportunity, but you
(27:36):
can only do that with the people, so your people are great.
So yeah, thank you.
Speaker 2 (27:43):
Yeah, thank you to you as well, Peter.
It's always great catching upwith you.
I hope to see you very soonnext time I'm in the UK or vice
versa.
If you visit us here in SiliconValley, you're always welcome
to hang out at the office here.
Peter Laughlin, Chief DataOfficer at Morrison's.
Thank you so much for joiningtoday's episode of what's New in
(28:03):
Data.
Speaker 3 (28:05):
Thanks, it's been a pleasure.
Thank you, Bye-bye.
Welcome to what's New in Data.
I'm your host, john Coutet.
Today's guest is Peter Laughlin, chief Data Officer at
Morrisons, one of the UK'slargest grocery chains, serving
millions of customers each weekacross nearly 500 stores.
Under Peter's leadership,morrisons has been recognized
for its innovative use of dataand generative artificial
(00:26):
intelligence to transform retailoperations and drive customer
loyalty.
He does it while partneringwith leading technologies like
Google, cloud and Stream.
Let's dive right in.
Speaker 2 (00:38):
Peter, how has real-time data improved your
operations?
Data improved your operations.
Speaker 3 (00:45):
We have a huge amount of data that we generate every
second, every minute, every hourof the day, not least from the
transactions that we have withour customers.
So, whether it's in our store,whether it's online, whether
it's through one of our partnerslike Just Eat or Deliveroo, we
are collecting data points thattell us what people have bought,
(01:06):
where they bought it and,crucially, how much money we
have charged them.
Now it's really importantrunning a business that we can
collate that data really quicklyso that we can take decisions
on that information, and so thefirst horizon really for the
real-time data was to make surethat we had up-to-the-minute
(01:28):
sales information, even slightlylonger, to run batch processes
to collect all that informationfrom the tills, do the
(01:50):
calculations, do all theaggregation and then present
that back into a form that wecan read and look at.
But we have a number ofpipelines in the business that
are collecting thosetransactions together literally
in real time.
So, as we've been speaking inthe last two minutes, I could go
to our data set and see who'sbought something within the last
two minutes.
What was it?
Has there been an increase in aparticular type of product?
(02:12):
Are we seeing particular demandin a certain part of the
country and, crucially, in aworld of technology where we
really care about making sure wehave uptime, being able to use
a stream of data to help sensethe health of the network is
also quite important Because, aswe are processing all this
(02:35):
information, we have aparticular expectation about how
much we'd see every minute, howmuch we might see every five
minutes, ten minutes, so on andso forth.
But being able to stream thatdata in real time means that
actually we can pick up on onissues around us as soon as they
happen, rather than having towait for a few hours to realize
that maybe there's a problemwith a till somewhere or maybe
(02:57):
there's a problem with one ofour pipelines.
So actually the real-time datastarts to mean that we can make
real-time sales decisions withina few hours.
Do we need to think differentlyabout promotion?
Do we need to order more stockbecause the promotion is going
even better than we thought itwas?
Other parts of the country, andcertainly recently in the UK,
(03:18):
we've had a number of big stormevents where we did have to
close a few shops for a fewhours because the weather was so
bad that the government advicewas people had to stay home and
not go out.
In that context, we have thatreal-time information to
understand just how ourcustomers are interacting with
(03:40):
us, to make sure that we canbest service them, to make sure
that we can best service them,and so that real-time data is
super critical in helping keepour finger on the pulse of
what's happening across ourphysical stores and online.
Retail is about detail, butretail is also about what's
happening now and what mighthappen in the next half an hour,
(04:00):
the next hour, the next twohours, and so real-time data is
super critical in that retailenvironment, because the numbers
are so large that you don'twant to get to the end of the
day and feel like you were eventwo or 3% away from where you
wanted to be, because that canbe a material amount of money.
Speaker 2 (04:17):
Being able to be really close to understanding
where we are at any point intime is super important in terms
of how you can run the businessabsolutely, and the way you've
deployed it is remarkable in howmorrison's has optimized that
for your internal operations andhow you're able to respond to
(04:38):
market changes and and eventhings like the weather and
stock and things along thoselines.
And that's what's beenincredible about partnering with
your team, peter, which is justthe obsession with customer
success and making sure thatyour shoppers are having the
best possible experience and theway.
(05:01):
From Stream's perspective,you're using Stream Cloud
multiple services and you havethis vast, complex supply chain
that we're pooling you know dataliterally from different
physical locations and thenmassive volumes and ultimately
applying that into theseapplications.
I think what's made that sosuccessful.
(05:22):
Again, you know, yeah, thestreaming infrastructure is
great.
The you know, the use of, youknow vector databases is really
innovative and it all comes backto the fact that you've applied
it in a way that really helpsthe customers in the business.
And you know I know there's alot of data practitioners and
data engineers listening to this, uh who are working on other
(05:45):
amazing data pipelines.
But just to to reaffirm justhow important it is to make sure
that the pipelines you'rebuilding are are feeding the the
customers and the business andthe people who are making those
operational decisions, andthat's that's really who you
have to focus on, right, so Idid.
You know, morrison's is a verysophisticated operation with you
(06:06):
know, thousands of farmers andsuppliers and growers, and it's
unique in how you're a grocerychain but you also control that
kind of end-to-end supply chainfor the food.
I'd love to hear how you'vebuilt the data infrastructure to
handle all you've built thedata infrastructure to handle
(06:27):
all.
Speaker 3 (06:27):
I think the key is to have a small number of patterns
that you can replicate, butalso recognizing that there's a
huge amount of complexity thatneeds to be simplified.
So the way I tend to thinkabout it is that we have a super
highway I think is probablywhat you call them in the US,
(06:48):
isn't it?
Or highway, at least we havethese sort of big multi-lane
roads that all go in the samedirection and broadly they go
quite fast.
Whether they do in practice ornot is probably a different
point.
Let's not get bogged down intraffic science, although
there's a lot of reallyinteresting science in traffic
science but my point is you needa big pipe that is capable of
(07:11):
transferring a lot of dataincredibly quickly, and, and
what you need is lots ofprocesses that are catching the
data.
But, crucially, the data thatyou catch might be in a slightly
different form to whether it'scoming from a manufacturing site
, whether it's coming from asupplier, whether it's coming
from a store, and so one of thebig areas of complexity that you
(07:35):
could find is if we were tobuild individual pipelines for
all of those, you would find youwere doing a lot of bespoke
work and you were doing a lot ofwork to just translate and map
things doing a lot of bespokework, and you were doing a lot
of work to to just translate andmap things.
And the way that we've tried tobuild things is is to recognize
that you need to do and I meanthis in a really nice way, you
need to do the least amount ofwork possible to give you the
(07:56):
most amount of flexibility.
Um across your, your sort ofyour data infrastructure, and
therefore one of the things thatum was always really high up
our list of requirements wasthis idea of how do you catch
the data and know where it needsto go, because actually I could
(08:16):
catch the data, I could spendtime thinking about where it's
going and then I could send iton its way.
But actually one of the thingsI like about where it's going
and then I could send it on itsway, but actually one of the
things I like about where thetechnology is now going is that
we can catch the data and we canworry about where it's going
when it's on its way, becausethat cuts down the time it takes
to do this process.
(08:37):
It improves the opportunity forlower latency, but it also
means I've got less work to dowhen I plug new data sets in,
because if you have this sort ofparadigm that says I've got a
pipe that goes as fast aspossible and it's incredibly
intelligent so it can kind offigure out where things are
going in the pipe and all I needto do is connect the data to
(09:01):
the pipe and then make sure thepipe knows where it needs to
ultimately get to at the otherend.
Now I hope that's not too toomuch of an abstract way of
answering the question, but butreally the the fundamental in
the architecture has been we.
If we built bespoke pipelines,every time we would need an army
of data engineers and we wouldprobably never get through the
(09:22):
work.
So what we've had to do isrecognize that you have to build
something that is flexibleenough but intelligent enough to
mean that our engineering teamcan be a bit lazy.
Um, and they're not lazy.
You know they're all incrediblyhard working and there's always
far too much to do.
But you almost have to set outto make your teams feel like
(09:43):
they are being lazy, because bydoing that you're creating the
right kind of behaviors in thesort of mindset for your
pipelines and, like I say, Imean it's maybe not about
getting into the whys andwherefores of how you can run AI
on the data as it's streaming,but it is about that concept of
ensuring that you can get thingson its way and then eventually
(10:05):
it'll figure out where it needsto go in an intelligent way,
because that makes things veryquick.
Speaker 2 (10:12):
Absolutely, and I love the way you phrase it
because, honestly, the bestengineers set themselves up so
that they can be kind of lazyabout to how Morrison's able to
actually innovate so quickly,given the scale that you're
running at, which is reallyincredible.
And the other thing you saidearlier in the episode was that
(10:52):
retail is detail and you've runMorrison's now, which is an
operation that runs atincredible scale, has, you know,
full supply chain visibility,full shopper visibility, all the
way from.
You know the food beingproduced, so when it's purchased
, and then you know being loyalto the customers and continue to
(11:13):
provide a great experience forthem.
So what's your advice to otherleaders in retail who are
embarking on a similar journeyof modernization and innovation
with AI?
Speaker 3 (11:25):
Start with your customer and be very clear on
why you are proposing whatyou're proposing.
It's actually quite hard to getunderneath the why?
Question because, again, if youlisten to your customers, they
will.
They will give you theirfeedback, they will give you a
very clear steer in terms of howthey feel, what they think is
(11:46):
working well, what they think isnot working so well.
There's an art form, though, inbeing able to interpret that
feedback and build a program ofwork that is able to talk to all
of those um opportunities butdo the right work in the right
way, at the right pace to um toactually deliver against those
(12:08):
those opportunities.
Um, I mean, it's always the uhthe henry ford example that
comes to mind around.
You know, if you ask peoplewhat they want, they wanted a
faster horse back in the uh, theearly, early 20th century.
What they actually needed was acar, and so you do have to make
sure that you're you're notover innovating, you're not over
complicating um.
You know our, our leadershipteam in morrison's, often talks
(12:31):
about complexity and simplicity.
We have a very complex business, but our job is to deliver that
as simply as possible, and sobe really quite clear on the
problem you're trying to solve,because what fascinates me is,
in a room full of technologistsand a room full of incredibly
clever data engineers, dataanalysis experts, data
(12:53):
scientists even with the samequestion, even with the same
observations from our customers,the solutions can be very
wildly different.
And therefore, keep asking whyuntil you get to the point where
you can absolutely crystallizeI'm going to do this, we're
going to do that because it'sgoing to deliver this for our
(13:14):
customers, and I think you'vegot to build your strategy
around that.
It might be that real-time datais super important for you and
your customers.
It might not be.
It depends on your product, theway you set your business up,
your customer value propositionand, equally, the use of ai to
automate various things.
There might be obvious thingsthat you'd like to do because
(13:37):
your competitors have done it,but challenge yourself to say
why do we do that?
Because we want to, or are wecopying?
And and I think, asking thatreally detailed why question can
be really quite hard and, attimes, can feel quite uh.
It feels like you might slowyourself down, but in a world
(13:59):
where you need to go fast, itactually speeds you up because
at the point that you then startto deliver something.
You're super clear on why andwhen it comes to then building
the requirements and talking tothe vendors and connecting all
the technologies together, youcan be really very focused on
what you need.
Now I think you know, if youlook at the, the way that we
sort of started to work withstream, I think we were very
(14:21):
clear on what we needed and wewere very clear that we needed a
capability that you were ableto provide.
Speaker 2 (14:27):
And I think you know, when you're setting these
things up and working aboutwhere you go in the future and I
, I think, being super focusedon why, I'm super curious on why
you think that that is the mostimportant part, I think yeah,
and that's that's one of thethings that has been really
(14:48):
incredible about partnering withwith morrison's and you know,
uh, you know working with you onyour, on your uh implementation
and rollout, because even whenwe're looking at streaming
real-time pipelines, we benefitfrom knowing the business use
case.
Why?
Because we're going to tune thepipelines accordingly.
And you might say someone mightsay, hey, our use case is just
(15:12):
having some reports that loadnightly.
And we're using stream becausewe want to do low impact change
data capture so we're notpulling the databases and making
that an expensive operationagainst the production databases
and that's fine.
So we'll do the real timechange data capture but we'll
batch and load the data intoyour warehouse, you know, every
24 hours, so you're saving thecost there.
(15:34):
But when customers say you knowwe want to improve the customer
experience by responding toreal time signals, that's when
we really say okay, then let's,let's go full throttle and help
you optimize those pipelines.
And you know it's been a greatpartnership and great teamwork.
You know from from the, the,the folks here at stream that
(15:55):
were working with Morrisons andthen the counterparts at
Morrisons who are reallydedicated to building out this
infrastructure.
But I think it really does allcome back to your vision and
your leadership for what youwanted to accomplish, which made
it easier and more focused forStream and I'm sure Google Cloud
as well to deliver what you'relooking for, because, ultimately
(16:18):
, when people know what theywant, people will rally around
them and help them get there.
So that can apply to all datateams the more the data team
knows about the businessinitiative and what they're
really trying to accomplish andreally who their customers
Because all data teams, whetherthey know it or not, they have a
(16:39):
customer.
It might be an internal customer, but there's someone that's you
know paying with their time anddedication and you know their
precious resources to work withthe data team in one way or
another.
And data teams just have toknow who that customer is and
why they're delivering.
It could be like an endcustomer that's actually buying
(17:01):
a product from the business, orit could be the finance team
right, they could be doingrevenue recognition with that
data or it could be someone inthe sales team is doing lead
routing with it.
But ultimately it all comesback to, peter, like you
mentioned, just reallyunderstanding the business
problems and how you're solvingthem.
So that's what's been reallyfun about working with
(17:24):
Morrison's, and now you'reseeing all these incredible
benefits and the customers arehappy and the financial metrics
all look great, and I thinkthat's certainly no coincidence.
And, peter, I also wanted toask you there's so much changing
in the industry as well.
(17:44):
Just with all the innovationthat's going on, what emerging
trends in data and AI are youmost excited about?
Speaker 3 (17:54):
I'm almost exhausted, to the point where it's hard to
get excited and I'm joking,actually.
I think there's so much goingon, um that it's hard not to be
excited, um, I.
I think that the challenge ishow do you deliver on the
opportunity?
Because as every week goes by,I think the opportunities get
(18:15):
bigger and I think the hype getsbigger.
If you go back 10, 15 years,there'd be new technologies that
would emerge and you'd have towork really hard to go to your
exec and say, on full circle, wenow have board members asking
(18:39):
for ai and asking for newdevelopments, because these
developments are now mainstreamto the point where I don't think
there's anybody who hasn'theard, either directly or
indirectly, of a large languagemodel um or a chat gpt, um, you
know so.
So I think the the really thereally exciting bit, but
actually the big risk, is how doyou realize that the
opportunity?
Because it's clear now theopportunities are massive and
(19:01):
the expectations of ourstakeholders are that we will
get our fair share of thoseopportunities.
But how do you do it safely?
How do you do it in a worldwhere people are learning to
walk?
So you know, we've gone throughvarious paradigms, haven't we?
You know, if you look at the,the mid to late 90s, it was
about how you enable computingin a business.
(19:21):
And then the world of the youknow the, the it sector as we
know it then sort of grew aroundthat, um, you know, data came
probably 10 years later and nowwe have generative ai and sort
of broader ai systems that arecoming through and so we're
going through probably that, youknow, maybe second in my mind,
third revolution you know thesetechnologies are now as
(19:44):
fundamental a change as it wasto start to use a computer in
your day job.
I remember one of the thingsthat inspired me to get into
this world was um.
One of the things that inspiredme, um, to get into this world
was um.
I remember my.
My grandfather was, uh,involved in in putting computers
into schools in the mid 80s andhe was massively inspired by
(20:05):
the bit, but he retired at apoint where he never got to use
a computer at work and I dothink the next generation that
comes through.
So if I look at my children, um, they're probably, you know,
the first generation that willgo into the world of work with
an expectation that they do useAI routinely in what they do,
(20:27):
and so this has been a hugecultural societal expectation of
shift.
It's massively exciting, but itis incredibly scary, because how
do you, how do you, keepcontrol of something which,
inherently, is uncontrollable ifyou want to go after the value?
And when I say it'suncontrollable, people are going
(20:49):
to find their own ways to usethis.
So even if we put variousregulatory frameworks in place
in the business to say, pleasedo this, but not this, I'm sure
people will find ways around it,whether accidentally or
otherwise, and thereforeactually the governance of this
is the thing that I'm probablymost concerned about.
But if we get the governanceright, I think it is so exciting
(21:11):
that things that took days,weeks, months to do will be done
in seconds.
And what does that do for ourunderstanding as a, as a society
, as a group of humans lookingfor ways to continue to live
sustainably and effectively on aplanet?
Um, and therefore I, I, you know, I think how big do you go here
(21:34):
?
But um, ultimately, I thinkthis is just another stepping
stone into increasing ourcollective intelligence in such
a way that we can then start totackle all those questions that
we still haven't solved, such aswhere do numbers come from and
how do we tackle global warmingand how do we cure cancer
(21:54):
forever, and actually there's somany really valuable and
exciting things that we could dowith the technology.
Um, but actually right now wejust have to figure out how to
use it effectively in a businessto to mean that we don't break
anything.
Uh, because if there's a bigbang where something goes wrong
with generative ai, it mightstop that.
(22:15):
That development of technology,uh, in a way that probably
would be would be bad forsociety but probably better for
everybody in the short term.
But I'm rambling now slightly.
Speaker 2 (22:27):
I mean, yeah, those are all you know.
Yeah, like you said, the nextgeneration is going to use AI.
That's going to be theirexpectation.
That's how they're going towork with everything.
They're just going to chat withsome robot, you know, some
robot-like thing that justunderstands what they're trying
to say, can derive the semanticmeaning from anything.
(22:49):
And you know, I think you putthat, you phrased that very
eloquently, peter, and you knowthat's the world we have to plan
for now and it is sort of arace.
And because you know companiesare trying to outpace their
competitors into who can rollout the best AI enabled service.
Because you know, chat GPTalready has tens of millions of
(23:09):
users, right, so people arealready just chatting with
things and, uh, you know,getting getting the responses
they want, uh, so it's such anexciting time.
And that, tying this back towhat you've already delivered,
right, ai driven search, right,it really sounds like you know,
you, you've kind of understood,so, the essence of the value of
(23:29):
ai, which is kind of this, this,this, this buzziness, this
little, you know, ambiguity ofthe way people describe things
and knowing what they mean andgiving them what they want based
on that.
So it's it's, it's really coolto see what you've already
delivered there, with the powerof getting the real-time
inventory data doing the vectorembeddings, surfacing that as an
(23:54):
AI-driven customer experience.
I just think that's so exciting.
I mean, that's really whatinterests me.
You can build all the AIinfrastructure you want, but
really what's most impressive ishow teams like yours have
actually rolled it out for thecustomer.
Speaker 3 (24:11):
You need to be left-brained and right-brained
at the same time, because we'removing now to a world where the
technology gives you a platformto be creative.
And actually the exciting thingis how do you use this
commoditized capability?
Because it'll all become it'llall effectively become
(24:32):
commoditized.
Everybody's ai out of the boxwill probably be about the same
as everybody else's, because,even if it's not, everyone will
catch up.
So let's just assume you knowthe base ai becomes kind of you
know broadly a constant factor.
How do you then build that intoother processes in such that you
are creating something new andvaluable and that is a creative
(24:55):
process.
And therefore we um, we'reshifting um from being a
knowledge-based tech communityto being a curiosity-based tech
community, and actually theshift is going to be knowledge
can be replaced or found in alarge language model.
(25:16):
The insight and the spark tothink of something new, to apply
it to, will become the valuableresource in the future.
But it's all underpinned by thedata and the pipelines.
So I mean it's fascinatingwhere it's going to go.
I would love to have thisconversation again in five years
, john, to see where we'veactually got to um.
(25:38):
But you know, super exciting,bit scary um, but the journey is
what's fun uh, yeah, it's, itis so much fun.
Speaker 2 (25:47):
I mean it is.
You know, it does remind mewhen, you know, uh, mobile apps
on the iphone and the app storekind of blew up and there's just
this, this kind of blank slateof innovation that we know is
going to happen.
And you know, there was sort ofa similar cadence with data,
(26:07):
but now with AI, it's just soclear that there's going to be
just a new generation ofapplications and use cases and
ways this can be applied.
Well, you know, peter, you know, at Stream, we've been so
delighted to work with you onthis five-year journey with
Morrison's.
I want to ask you about this alittle selfishly, but what is it
about Stream and our peoplethat really made the difference
(26:28):
for you?
Speaker 3 (26:31):
You worked with us really collaboratively and you
co-created the solution with us.
So I talk a lot about thedifference between collaboration
and co-creation with us.
So talk a lot about thedifference between collaboration
and co-creation.
Um, co-creation is the good oneand what I'm looking for, um,
which is where we kind of alljump in.
We're one team, we're clear onthe goal and we work together.
I think, um, we often joked inin the project meetings that it
(26:54):
was difficult to tell um thestream team from the the
morrison's team, because it youjust worked with us really
really well.
Everybody just knew what wewere trying to do.
It was hard yards, it wasdifficult stuff.
I mean, I think it it can bequite challenging to sit back
from this and go.
We did a, we did a really bigthing, um, and we did it really
(27:15):
well, and the reason we did itreally well is because you guys
support us brilliantly and youknow, I think, think the real
difference was you havetechnology, but you have some
fantastic people and workingwith you was easy, and so there
were lots of real positivesthere, because it's really
important that you can actuallyembed the opportunity, but you
(27:36):
can only do that with the people, so your people are great.
So yeah, thank you.
Speaker 2 (27:43):
Yeah, thank you to you as well, Peter.
It's always great catching upwith you.
I hope to see you very soonnext time I'm in the UK or vice
versa.
If you visit us here in SiliconValley, you're always welcome
to hang out at the office here.
Peter Laughlin, Chief DataOfficer at Morrison's.
Thank you so much for joiningtoday's episode of what's New in
(28:03):
Data.
Speaker 3 (28:05):
Thanks, it's been a pleasure.
Thank you, Bye-bye.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.