Industry sign banking idaho job description template fast
- [Joe] Today we're gonna
talk about graphs and banking. The idea for this really is
to encourage graph making and help people who may
work in the banking industry understand how their data
naturally forms graphs and how graphs can improve their business in a number of areas as well
as Sabine said in her intro, where there's an
opportunity for using graphs with more advanced data science techniques like machine learning and AI. So we'll start first with
an introduction to graphs and Neo4j just to make sure
everyone is on the same page. Then we'll take a look at an overview of the sorts of data
that a bank would have that will then carry forward
into some particular use cases. So we're gonna drill into
four use cases for graphs in banking, first will be fraud and with that we'll see a demo. Then we'll talk about
risk, knowledge graphs and the Customer 360 degree view. Each of those will take a look at a graph put together from the types of data we talked about earlier and will explore how that graph can then be used for each of these particular use cases. And then we'll wrap up
with some Q&A at the end. So to introduce graphs. So everyone is probably familiar
with relational databases. They've been around for ages, it's what many businesses run on. They're really kind of quite structured, columns and rows powered
by the SQL query language. So you select some values,
some columns from a row that matches a particular value
and you get that data back. So, very table oriented,
a lot like spreadsheets. You put data in, you get data out. If you want to join entities together, you can do that with
joins and join tables. There's a known design pattern
for this normalization, but that joining of very
complex sets of relations together can really be
quite slow and cumbersome and difficult to build a query for. On the other side we have graphs and graphs are really
built for connected data, highly connected data, right? So where you have very connected data sets things, entities that are
linked with other things in very complex ways,
graphs can help you leverage the context, the additional information that's held in those relationships. And the query language for that is Cypher. That's our query language from Neo4j and you can see an example of it there where there's someone named Ann who loves someone named Dan. Today we're not really
going to get into Cypher as a language, we're really gonna focus on the graph data model and understanding that representation of data. So the idea of graphs
is as I said, leveraging the power of the
relationships between data. And in Neo4j we always
say relationships between data is as important as
the data itself, right? And we believe the next wave
of competitive advantage in businesses all around the world and across all industries
will be powered by using those connections between data to
identify and build knowledge. And what you see here is a
graphic that spans data silos. So often the relationships
between your data can be difficult to see and
certainly difficult to use because your data lives in all
sorts of different databases across your business. So you have customer data,
you have your product data, you have third party
data like market data, maybe social media data,
credit agency data, joining all this together
in one coherent view to allow you to leverage the
power in those relationships is what Neo4j offers. So you might be thinking
how can graphs help me then? So we're gonna take fraud as an example. In the olden days fraudsters
were maybe less advanced before there was the internet
and all of the technology we have available now, fraud
was maybe more straightforward. And so were the techniques
used to identify, right? So we have a lone fraudster, maybe steals someone's credit card, goes out and starts shopping with it and we have techniques to be able to find that sort of transactional fraud. Fraudsters have grown,
advanced over the years and the techniques they use
are now much more advanced. The technology they use
is a lot more advanced. So the problem is you've
gotta find these guys, right? You have to find a fraud ring. Fraudsters now often work together to try to fly under the radar
to avoid the old techniques that were aimed at old fraud. Instead are using their new techniques, working together and what you need to do to find the patterns in
your data that can show you the behavior of that fraud ring. And that's what Neo4j really does, right? We are the number one
database for connected data. An enterprise-grade graph database and we say we're a native graph database which means we are built
explicitly for working with graphs, for working with connected data. And everything about our
database from the ground up is really engineered to work
with those relationships and that highly connected data. So we allow you to store that data, to access it, to understand
it, to traverse it, to visualize it and to
manipulate it live on the fly. Add data very easily. It's highly performant, it's scalable, it's everything you would expect from an enterprise-grade database. So to introduce the concept
of the graph data model for anyone who's not familiar with it, it's really built of three
very simple components. So first there are nodes,
and that's what we see here. The circles, we have two person nodes and we have one car node. Nodes in the property graph
model tend to be nouns, they're things, they're entities. And the labels that we have
on them, person and car are just a way of doing
that, of labeling them so that they're easy to
query and group together, but it doesn't necessarily
imply any kind of schema. So like in the relational
model where your schema is fairly rigid, every row
has the same columns in it and if you don't have
a value for that column in a particular row you
have to fill it with no. Graph databases are more
flexible so one person node doesn't have to look exactly
like another person node. Between them we have relationships. So we can see here, this
person loves this person here, this person loves this person here. Relationships are always
saved in the database with a direction, you can see
that with the arrows here. And each relationship
can only have one label. So you couldn't have loves and lives with in the same relationship,
that's why we have two. The directionality is sometimes important so for example loves is important. So we have this person here, this person here whose name is Dan loves this other person named Ann. Luckily for him Ann loves him in return, but if she didn't that would
relationship would go away. So it's important to know the direction of that loves relationship. Whereas lives with, it's kind of implied that it's bi-directional
so if Dan lives with Ann, then of course Ann lives with Dan, it would be logically impossible
for it not to be that way so we can ignore the
direction at query time and therefore we just
need the one relationship. So we have nodes,
relationships and the last part of the property graph is the properties. So that's what these are up here. Nodes can have properties
as well as relationships. And properties are just key value pairs. So here we see we have
a property called Name and the value in that is Dan. And we know Dan was born on May 29th 1970 and we have his Twitter
handle which is @dan. We know this person has
a name property of Ann. She was born December fifth, 1975. We don't have her Twitter
handle and that's okay. Like I said, there's no schema
enforced out of the box, although you have the
choice to apply a schema to your database. If we get Ann's Twitter
handle down the road, it's very easy to add it. If instead we want to
add her Instagram account or her Facebook account
we could do that too. And we can see as well there's a property on this relationship,
so Dan drives the Volvo that Ann owns and he has driven
it since January 10th, 2011. So those few components
give us all the tools we need to be able to
build a really complex and expressive graph data model
which we'll get to in a bit. We're really gonna focus today
on nodes and relationships. A little less than
properties although as we go through each of the
graphs we're gonna look at we can talk about the sorts of properties that would be there. So now we're gonna take a look,
just a high level overview of the sorts of data
that a bank might have before we then start putting that data into different graphs. So first, we'll look
at organizational data. And this is data that you as a bank hold about yourself really,
it's about your business. So you probably have lots of documentation and that can live all over the place. Maybe you have SharePoint,
maybe you have internet that you might have
documentation at various levels, you've got global documentation,
regional documentation. Then there's processes and the
same would be true of that. You probably have global processes as well as more regional processes. You'll have employee data,
information about the people that work in your company as
well as the organizational hierarchy, so where do those people fit into your organization, who reports to whom, what
projects do people work on. Then you'll have KPIs and reports, obviously this is really important for running your business, understanding how those
KPIs are calculated, how you report those internally as well as reporting them to regulators and governing bodies. And then you have systems and databases. So this is where your data actually lives or the applications that you actually use to run your business. Next we have data about your customers. So, you'll have personal data about them, maybe know where they
live, their phone number, when they were born, so on and so forth. You'll have some documents from them, so this might be
documentation they provided when they opened their accounts or when they applied for a mortgage. You might have some information about their relationships. So you can see here there's
already a little graph. Maybe you know their spouse,
maybe you know details about their children,
any information you have about other people they
might be related to. So far we've been kinda really
talking about personal data. This is kind of retail banking. But of course you could
have commercial banking data as well, right, so you might
be working with companies. You'll have information
about those companies, their corporate data. And we'll explore both scenarios. I think most of it we're
gonna focus on retail banking because I always think that's
kinda the most relatable. Most people have a bank
account and then work with, interact with a bank and so
they understand those processes. But we will look at a particular scenario for corporate banking as well. And then you might have
information about assets. Especially if you're looking
at corporations or businesses, but also individuals, right? If they put up security for
loans or something like that. Next we have information
data about your products and your services. So again you have documentation
about those products and services, processes related to them. What is the process that you go through when someone applies for a current account or a mortgage? You have the details of
specific products and services as well as the hierarchy. So you might have an
extensive set of services and products that you've
offered over the years, you might have like a
closed book of services you don't offer anymore,
but people still have them. So being able to relate all those together and understand the details
of each of those products and services is quite important. And then you have the
information and knowledge that you use when you're
pricing individual products and services. Next there's events data. So I kinda grouped
together a few things here, there's money movements, transactions, there's web and app activity
that you might be tracking when someone logs into
their banking website or uses an app. And then there's call center or customer contact information. And lastly you might be
using third party data. So you maybe track social media mentions of your bank, of your organization. You certainly interact
with credit rating agencies both for retail banking as well
as for corporate customers. And then there's market data again, which is more in the
commercial banking space. You need to watch the market movements and understand how that
will impact your business. So certainly not an exhaustive list, but this will definitely
give us enough to work with where we can understand
how you can build graphs from all these different bits of data to then power individual use cases, right? And again, it's important to remember, this data could live in a
number of different systems across your organization. So if that's true, if you
have very siloed data sources and right now it's
probably quite difficult to join together all this data
and to get that holistic view that you need for these
individual use cases. So hopefully you'll start
to see how the graph can really change the way that
you can run your business. Okay, so first, we're
going to look at fraud. So this is a graph that
I built to illustrate the sorts of data you might
use for detecting fraud or for investigating fraud. I built this using a tool,
you can see there's a link on the bottom here and it will be included on the resources page at the end. But we have a tool called arrows which some of my colleagues have built, and it's a really great
tool for kind of dragging and dropping and visually
building a data model in a browser, so it's
quite convenient today. And here we have the color coding matching the different classifications of data that we talked about before. As you can see on the right hand side. We don't have any organizational data in this particular fraud example, so this is really looking
at the potential for credit card fraud or money transfer fraud
or something like that. You could of course, have a separate graph that would include organizational data if you were looking for
fraud inside of your company. If you were looking for your own employees who might be committing fraud. So here we have in blue,
some customer data, what you know about your customer. So here we have customer. That customer has some documents
that they've submitted, so we have a copy council tax statement, and their passport from when
they set up their account. We know their home address,
and we know their phone number. So when we look at our fraud demo this information is quite useful during, this is the information
you gathered during KYC, Know Your Customer, as
you're setting up an account. And you can use this to then detect fraud. One of the ways that fraud rings operate is by signing up for multiple accounts. And they do this using bits
of synthesized identities, bits or real identities,
so that when they sign up for an account it looks real, but actually maybe some of
that information is stolen or it's being reused. It's really difficult to fake
one complete false identity that will pass through the KYC checks. But using synthesized
IDs or bits of real IDs fraudsters have a better
chance of success. Then in green we have some information about the products that our customer has. So here, for example,
this customer might have a current account, they
might have a credit card through their bank. That current account has a debit card. And then all the rest
of the data that we have round about here is event data, which is not surprising, right? And in fraud investigations,
fraud detection is usually driven by suspicious events that happen around an account. So starting here at the bottom, we can see this credit
card was used at an ATM to get a cash advance, to take cash out from the credit card. So that may or may not
be suspicious behavior for this particular customer. We're not showing properties here, but you might want to know
where this ATM was located. Is this customer typically in that area? Is this ATM, has it already
been flagged as suspicious? Has other fraud been
associated with that ATM? These are the sorts of things
you would want to look at. Here we have some purchases
from two different retailers using that credit card. Again, you would want
to look for a pattern of suspicious behavior. Often what people do, fraudsters do, when they get a hold of
the credit card information or debit card information,
they might make a few small transactions to
verify that the card works before going all in and
buying something really big, really going through the
balance of that credit card. So you might be looking for
patterns of transactions that would indicate
potentially fraudulent behavior or again you might look at these retailers and understand if they've
recently been associated with fraud on other customer accounts. Same goes here with the debit card. It was used to make two
purchases at a retailer. The same things apply. Cash withdrawals from a different ATM. So again you'd want to know
the location of that ATM. This is probably less suspicious, you've got a money transfer
into the current account from that person's employer. Then we have a money transfer here that was made during a web session by this customer from
a particular IP address and the money was transferred
to a foreign account. And again, that may or
may not be suspicious depending on the typical
patterns of behavior for that particular customer. And again whether there's
anything about that IP address or that foreign account that ties it to other incidences of fraud. So, hopefully you got a
sense of how the connections between these pieces of
data can actually be used to do some really advanced
investigations around fraud, looking for patterns of
behavior that are unusual or if you identified some fraud before, seeing if this matches that same pattern. So, to think through
how we might use this. You might be thinking can I
find patterns in the graph indicating fraud in a particular account? Yes, I think we've talked
about how that can be done. You can look at one account,
look at all of the activity and the patterns and the data around that. If you find something that's suspicious, maybe withdrawals in short time frames from multiple ATMs, purchases
from different IP addresses as well using that same account, going to suspicious ATM machines that have been linked to fraud before. Looking for a pattern
in a particular account is definitely a use case here. Then the next question is, if you find a pattern
of fraud in the graph for one account, can you then look for that pattern on other accounts? And the answer is yes. Once you identify a pattern that says when I see these things happening, then you can apply that pattern to your entire graph, all your customers, all your transactions. And see where else that pattern appears. So then you can find more
incidences of fraud potentially that you hadn't known about before once you identify that pattern. Next, you might think,
okay, I've detected fraud, but can I use the same
approach to prevent fraud? And the answer is yes. So once you've identified a pattern that is typically fraudulent, right? Or that is at least worth investigating, you can then set up alerts. So as you see that
pattern starting to form, you can raise an alert and
you can prevent fraud, right? So it's kind of shifting from
detection after the event into prevention, being more proactive, stopping fraud before it happens. Again by identifying patterns of behavior and being able to spot those
patterns as they develop. And then lastly, could you use this graph alongside some more advanced
data science techniques, things like machine learning, to find new patterns of fraud
you didn't even know about? And the answer is yes. This is a really exciting area for us and for graph databases in general. The ability to link the
graph with machine learning really is more powerful
than the sum of its parts. So the graph can be
used to feed information into a machine learning pipeline so you can identify features
and their relationships in the graph that you can
then feed into the machine learning pipeline, use
that machine learning to find patterns, to create
a model that can then be applied back to the graph. So you can use that to
identify new patterns of fraud, apply that into the graph,
look for those patterns where they may exist. Again, try to prevent those
patterns from happening by spotting them as they form. And then that becomes a loop. So as more data gets added to the graph, you can see how successful
that machine learning model is. You can then feed the graph
back into your machine learning to refine it and continue that cycle. Continuously looking for new patterns, improving those patterns using that loop, that feedback loop between the graph and your machine learning pipeline. Okay, so now we'll take a quick look at a fraud demo in Neo4j
just so you get a sense for how you might
actually use the database. So this is a tool called Bloom. This is a more business user friendly way of interacting with Neo4j
where you don't need to know the query language. It's all powered by this search bar. And I've set up a few queries
for us to take a look at. So first we're going to
look for some fraud rings. And I can show you the query for this. It's something I've written in Cypher, as I said before our query language. I typed in find fraud rings,
which is a search phrase and what it does is it
kicks off this Cypher here, which basically looks for rings of, individual's rings of account holders in our database that share contact information amongst themselves. So we can take a look at a few of these and see why exactly they've shown up in our query results. So first we have two
account holders, no three. We have Carolyn, we have
Simone and we have Jay. Jay and Simone share a phone number. Not uncommon, maybe they live together. What is uncommon is that they also share a social security number. So in the U.S. social
security number is a unique identifier, it shouldn't be possible for two people to have the
same social security number and so therefore, straight
away, that's a red flag. Carolyn shares an ID with Simone. So both Carolyn and Simone have submitted the same password details. Whereas Carolyn has her
own social security number and we can see Carolyn and Jay and Simone all have the same address together. So this is an example of
what I was explaining before, synthetic identities
shared bits of identities being used together to set up accounts. Take a look maybe at one more example. So here we have two individuals, Kathy and Chang. They share an address. Again, not very unusual, what
is unusual is that they have the same driver's license details. Their ID is the same, so
again, another red flag. All right, two people can't have the same driver's license detail. Bruce here shares a phone
number with Chang and Kathy. So even though he doesn't have
the shared driver's license or indeed any kind of major red flags, he is implicated in this
ring because he shares a phone number with Kathy and Chang. So even though he doesn't share passports or driver's license or
whatever, we would still think he's probably part of this
ring because he's tied to the other two by this
relationship of the phone number. Just to show you a little bit
more about how Bloom works, we can drill into some of these entities. So if I double click on Chang, I can see the properties
associated with him, so these are his details, his
birth date, his full name, he has a unique identifier in our system. I can see all of the relationships that are associated with him. So they, I don't think
they're all out now. And we can see the neighbors. So these are the nodes that sit at the end of those relationships and we can see some of the details for those. If I wanted to I could then
reveal any that aren't out, so his bank account details weren't out, I've now brought those out on the display. Okay, we'll take a look then
at another type of fraud. So this is more kind of
eCommerce oriented fraud. This is using transactions
that look suspicious to identify potential fraud. So we'll start by saying
find eCommerce fraud and what this does is, it's
pretty straightforward, it's just looking for events that happen within a certain time of each other. You know where, so here
we have a particular card, a bank card that was
used in two purchases, this one at 09:55:13 and this one here at 09:55:09 so within four seconds of each other from different IPs which
are located in different U.S. states, so potentially suspicious, maybe you need to know a little bit more. It could be a parent and their child who are using the same credit card, the same bank card. So we have some additional details here. We can see the accounts that were used to make the purchases. We can see they were both delivered to this same address which is in Maryland. Even though one of the
purchases was in Idaho, one was in Rhode Island. So again, this may or
may not be suspicious, but if you tie this together
with other information you have in the graph it
might build a bigger pattern of fraudulent behavior. We can also see eCommerce fraud
with subsequent purchases. So we identified the behavior that we thought was suspicious, which was the two original purchases made from different IPs in
such a short timeframe from each other. Here we have two other purchases that were made subsequently,
one at 9:58 and one at 9:56. If you think the earlier ones are fraud then these two might also
be something that you want to dig into and understand whether or not they could also be fraud. All right, so again we
have a third account, a third delivery address, however this node down here at the bottom was made to the same delivery address. Maybe it's this one
here that's the outlier. So it's quite a visual way to
perform your investigation. Very easy for a non-technical user to be able to use, as
long as you understand the domain and a little
bit about the data model, it can be quite easy to
visually explore the graph and look for patterns without
having to know any code. Okay, so that was our Neo4j Bloom demo. Now we're going to go
through a few more use cases again looking at the
graphs that we can form from our banking data and
understand how the graph can power these use cases. So next we'll look at risk. In this scenario, this is really focused on a business customer
rather than a retail banking customer or personal customer. So again in blue we have
some information that we have about that customer. We know some information
about their business. They own another business
so they have a subsidiary. We know some information
about their primary contact, their phone number, their business address so on and so forth. We also know who owns them
and who ultimate beneficial owner is, right, so
this is quite important to understand how companies
relate to other companies. Particularly if you're looking
into anti-money laundering or indeed in risk if you need to know something that affects that subsidiary. If their business goes down,
or their business shuts down, if they go into bankruptcy
or something like this, it certainly could affect this
particular business customer. We know their risk rating,
which we have assigned them using our internal risk rating process. So here in red we have more
of our organizational data. And we can see that this
risk rating is aggregated up into a KPI which is then
included in the report. That would be given to the board or indeed maybe a regulating body. Here in black we have some external data, some third party data. So we have a credit rating,
that business gets externally, maybe from Moody's or something like that. We have information
about their share price which can certainly fluctuate, so we probably have lots
and lots of instances of share price data about them so we can see historically how
their share price has fared. And maybe we have some
press releases about them. Maybe we track information
about this company in the news or that they
release to the news. Or it could be their
quarterly earnings reports or yearly reports, things like that. So that we can understand
how their business is doing and how that might
impact their risk rating, whether or not we need to adjust them. Then we have some product data about them. So they have a business account, they have a business loan. We can see they've made loan payments from that business account
into their business loan. We can also see this foreign account has transferred money
into that business account and also this business
account has transferred money to their owners account. And so a little bit of end
data, some transaction data. So, what sorts of questions can we ask about this graph then? How might we be able to use it? Can we use the patterns in the graph to help improve the risk rating process? I would say yes. The more you have a holistic
view of your customer and all the data associated with them, and the more real time that view is, the more effective you can be
in your risk rating process. The quicker you can react to things that may be changing. So that's the answer to
number two as well, right? Get to more quickly react to changes. If you're tracking external
data, press releases, share prices, you're looking at patterns in the behavior of their
accounts and money movements. All of these things will
help you quickly react to changes that may alter the risk profile for that particular customer. So next, can you use the graph
to understand the lineage of data which feeds your
KPIs and your reporting. And the answer to that of course is yes. When you know where your data lives, you have this map across all your systems of your customer and you
know where the data lives, you can prove the lineage of that data and how it's been fed
into that risk rating and how then all of your
risk ratings are aggregated up to a KPI and to reports. And so this is very important. Can you use the graph for
regulatory compliance? And the answer would be yes. BCBS 239 specifically is all about this, how you report risk,
how you aggregate risk. Being able to demonstrate
that you really understand how you determine your
overall risk profile, where that data comes from, how you know that it's
accurate and how you know how it was calculated and aggregated. Next we'll look at knowledge graphs which is another kind of very
exciting area for graphs. You can see this is, this
picture is very green. A lot of this is about product data. So this particular scenario
is a knowledge graph that's customer face. So here we have our
customer, a retail customer. They have a current account,
they have a mortgage and they're interested in an ISA. Maybe they want more
information about an ISA. So from their current account
there's a lot of information that we can offer them, right? So a knowledge graph is
really a map of information. It's all the information that you have about your services, your
products, their documentation. And building a map of
that so you can understand how it all fits together and find the information you
really need most effectively. So for current accounts in general, you might have general processes, you have a web page that has information about current accounts,
maybe have an FAQ page about current accounts
so this is all kind of generic information
about current accounts. Here, because they have
specific current account, you need to know the
definition of that particular current account, right? What are the things that
make up their current account and is it different from
someone else's current account? Do they have a premium
account or a gold account? Are there specific benefits
that come with that account? Is there a specific fee? Are there any charges that get applied? Is there criteria on that? So all of that product
specific information is probably more
interesting to this customer than some of the other
more general information, unless they're interested in
moving their current account. So we want to know what are processes that are associated with
this particular product and what is the documentation associated with that particular product. So if this customer comes to us and they want to know the information about their current account
and all the details, we can get that
information to them easily. Then we also see this current account is managed by a current accounts team who has specific contact details. So if this customer needs to reach out to someone specifically
that they can talk to about their current account, this is how they can
find that information. It's a very similar
setup for their mortgage. We have a mortgage page, mortgage FAQs, general processes about
mortgages in general and then we have the specific details about their mortgage program. What are the processes
associated with that, the documentation associated
with their specific mortgage. And again that's managed
by a mortgages team with their contact details. If they're interested in ISAs
then the general information is probably where they want to go, if they don't have an ISA yet. So the general ISA processes,
the general ISA web page and the general ISA FAQ will
all be applicable to them along with the ISA team
and their contact details. So, could you use that knowledge graph to then improve your
customer's experience? And the answer would be yes. No customer wants to
struggle to find information so being able to get the information that they're looking for quickly and with accuracy is very important. So you can use this
knowledge graph to power a search engine, to
improve search results. Could you also use it
with a chat bot as well as a search engine? Absolutely. You have lots of use
cases, Ebay in particular is a very interesting
chat bot knowledge graph implementation where they
use a knowledge graph that describes all the
attributes of different products that are for sale on Ebay. And the chat bot uses that knowledge graph to then make recommendations
and have a conversation with someone about the products
that they're interested in, what attributes might be
most interesting to them. So if you're interested
in implementing a chat bot maybe for self service help,
building a knowledge graph to help that chat bot
traverse all the information and find the most specific information that would be applicable, it's
certainly a great way to go and could really again improve
your customer experience by giving them the right information in a very easy format. Could you use a knowledge
graph internally as well? Absolutely. So we only looked at a
customer facing example, but you could do the same thing with all of your internal documentation. You can build a knowledge
graph that describes the hierarchy of your company,
all of the documentation that's available and then
again you could use that to power search results,
to give additional context to be able to recommend documents
that might be applicable to a particular person. Or indeed, if you wanted
an internal chat bot, to help power that chat bot
and give it an understanding of all of your information
and how it's linked together so that it can traverse that graph and provide the right information
during its conversation at the right time. And finally, customer 360 view. So this is the idea that you
want all of the information about your customer in one
easily accessible view, whereas often this can live
across a number of systems. Your customer data might
live in a CRM system, the information about
their accounts might live in another system, event
data like transactions could live in a payment system. So being able to gather all
this information together in one place that you can view it can be really powerful and
can really have an effect in a number of different
places in your business. So this is kind of the
biggest and most complex graph that we've looked at. Again, still in blue we have information about our customer. So here's our core customer here. We can see this customer has a spouse. They're married to someone, they both live at the same address, and they are both signatory to a mortgage. That mortgage was for the
purchase of their home where they live. We can see documentation that they have associated with that mortgage, so here's their signed mortgage agreement. Here's the documentation
that the spouse supplied, a passport and a bank statement. And here we can see the main customer supplied a passport and
accounts tax statement. Can also see the customer's
marketing preferences. So maintaining this sort of information, particularly in the context of GDPR is really important now right? You need to know how your
customer has given you permission to use their
data and how you're able to market to them to
remain compliant with GDPR. In black we have some
credit score information that would have been retrieved during their mortgage application. We can see that in
addition to their mortgage the customer himself or
herself has a current account, and a credit card. Can see each of those products has a specific definition. We can see a few transactions. So we can see the current account has paid into the mortgage twice
as well as an overpayment. And we can see the current account has some debit card activity, a few purchases from a single BTI. So of course this is quite a small view, we have only one credit
card purchase in our view, but normally you would have
lots of historical data. Neo4j does provide the ability to stamp all of your relationships
and nodes with a time. So you can maintain an historical view, you can also take a snapshot of the graph in a particular time range. So certainly there
would be a lot more data that would sit behind this. And then lastly in red we have internal data, organizational data. We can see this customer
has a mortgage advisor who works at a particular branch and that customer uses that branch. And we can see there's been
some customer contact as well. They've made some increase. We have a record of
all the contact points, all the times they've
called into our call center or emailed us or what-have-you. There's even more information
you can add to this. You probably have information
about your customer's phone number, maybe you
have information about their social media interactions with you, if they've ever mentioned
your organization in a hashtag or something like that. The idea is to have a map of you, of all your customer data and then you can use that
to improve your interactions with that customer. So, could we use the customer 360 view to improve the customer experience? Again, absolutely. I personally get very frustrated when I contact an organization, especially one I've
been doing business with for quite a long time and the person at the other end of the phone doesn't really seem to
know anything about me. Customers expect you, as an organization, to have access to all
of their information. And it's not really their problem whether or not they're talking
to the right division or that information's
held in another system. If you want a really smooth and delightful customer experience, you
really need all the information you can get about them in one place. And having this 360 degree view graph can really help you do that. What else could you use that graph for? Could you use it for anti
money laundering, for example? And I'm gonna say, absolutely. Having that 360 degree
view of your customer gives you a full view
of all of their behavior and all the data you have about them which can then be used to spot anomalies. So if you start seeing strange behavior, different accounts,
sending money into them, and they're not a business,
then that could be suspicious. If they start opening
all sorts of new accounts and sending new money all over the place, that could be suspicious. So when you have a full
view of all their data you can more readily spot outliers, different behavior than the norm. Can you use that 360 degree view to detect and prevent church? Yes. So again, when you have
that historical view and that complete view
of customers' behavior you can identify patterns
that tend to indicate churn. And you can identify that potential churn in advance and work to
correct it before it happens. So some obvious things
that could indicate churn, someone starts draining their bank account and maintaining a very low balance. Maybe they're paying off their credit card or their mortgage from
a different account, rather than from their
account they have with you. Maybe there's a series of
calls into the call center, they have a negative score where
they've called to complain. Maybe there's some social media activity where they're being
negative about your brand. Whatever that looks
like, if you can identify those patterns of behavior,
again potentially using more advanced data science techniques like machine learning, using
the data you have in the graph to inform your machine learning and then finding patterns of behavior that can identify churn. Then once you have that pattern or those patterns that indicate churn, you can apply them to the whole graph, find people who may be about to churn and correct that or again, set up alerts. Where as you see a
pattern starting to form you can then be proactive. And then finally, can we
use the Customer 360 view to improve upsell and cross-sell? Absolutely. When you have a view of your customer, and a view of all your customers, you can start segmenting your customers, comparing customers to each other. If you have customers that look or behave very similarly
and a portion of them are buying a new product,
maybe you can recommend that new product to others. Maybe you can identify pattern
that's in your customer's behavior that might indicate
maybe they have a new employer, they've gotten a raise,
maybe they're in the market for a new house, all of
these sorts of things that you might be able
to detect from patterns in their behavior, again
potentially using machine learning. And then proactively use
that pattern detection to approach your customers for upsell and cross-sell opportunities. So there's other use cases
that we could go into. For the sake of brevity today we won't, but these are more general, I think, across a number of different industries. So there's identity and access management. It's sort of like an
internal knowledge graph where you can understand your hierarchy maybe when someone
joins your organization, where they fit in, the
complex relationships between global and
regional reporting lines, what projects they work on and so forth. And then identify therefore what systems and what documentation
they need access to. We have a published use
case from UBS on this. It's what they've done quiet successfully. I think they were able to
reduce the onboarding time for new joiners by weeks. So that when you hire someone they're able to get up and running more quickly. Infrastructure and network management, managing your IT estate,
understanding the connections between all of your servers
and your networking equipment, being able to manage that. Master and meta-data management,
which is very similar to the customer 360 degree view that we talked about before. So being able to understand
where all data sits across all your systems,
how it fits together. And then that of course, can
lead to regulatory compliance. So GDPR is very much
about that understanding your data, where it lives, being able to demonstrate that you're using the correct data
and that if someone wants the data you hold on them to be removed, that you know where that data lives in your various systems. And the power of the graph is really not just at the small scale
that we've been looking at, you know, very small
graphs of a dozen nodes or something like that, right? The power of it is when
your graph looks like this. When you've got to scale to
the size of a very large bank, very large financial organization. Lots of transactions, lots of customers, tons and tons and tons of
relationships between them. That's really where the power of the graph comes in to play, where you
can identify these patterns at that sort of scale
across your entire business.
