April 9, 2020
Microsoft Dynamics 365 Fraud Protection: The Journey

Microsoft Dynamics 365 Fraud Protection: The Journey

Hi, welcome to the learning
series for Fraud Protection. We will be inviting industry
experts to talk about best practices that top concerns and
the solutions that they have put in place. Not only that, we will
also be inviting Microsoft engineering teams to talk about
the innovation that is happening in Microsoft around Fraud
Protection. In this episode, we have a very special guest with
us today. His name is Jay Nanduri and he’s the GM and a
Distinguished Engineer with Microsoft. Welcome, Jay. Thanks, Kapil.
Thanks for having me here. So why don’t you tell us a
little bit about yourself, Jay. I’ve been in the software
industry for more than 24 years, and in the last 20 of those
years have spent at Microsoft, and in the last five years have
been working in cloud on AI in the Commerce ecosystem. How do
we understand our customers and their purchase behaviors? What
kind of data assets that we need to drive a better business for
Microsoft, but also create products that the customers
love. Wow, that’s very impressive all
together. So tell us a little bit about Fraud Protection. How
did you get involved with Fraud Protection? Microsoft was going through a
kind of a transformation as you know, you know, Microsoft is a
predominantly an enterprise, you know, engine player and
most of the economics come from there, but we were also getting
a lot of kind of purchases customer purchases from app
stores. So as we were going through the transition, we have
seen that we needed to make the system more efficient. It was a
play where we wanted to increase our top line and also kind of
reduce our OPEX, right. That’s very cool actually
impacting the customer experience as well as the
revenue altogether. What did you guys do? And how did this
actually end up becoming Dynamics 365 for Fraud
Protection? We pretty much had a
predominantly a rule-based system, any fraud system that we
were doing is we were looking at, “Hey, is this a good
geolocation where the purchase is coming from?” Then approve, or
deny. The only way we could do it is look at Microsoft
holistically, so that we can look at a customer’s journey
into Microsoft, across all the product segments. In the same
vein, we can also look at the fraudster’s journey across the
products. And that itself has given us a little bit of a, you
know, uplift because now fraudsters were not able to go
attack business by business because if we stopped them at
one gate, they cannot come in through another gate. Right.
And the second thing what we saw is the rule-based systems could
not scale right? We saw that, hey, we need to actually improve
this technology. Let’s bring all our AI power to this problem. So basically going from reactive to
proactive. Proactive, right. So then we, the data that we brought
from all these stores. Now we let loose all of our ML capabilities
on them. We see that we were almost 5 to 6% of
transactions were getting manually reviewed, and it had
gone again, there is a cost associated. Right, there’s an
operations cost. Operations cost. So now we took our AI assets and ML assets
and we generated deep learning models on top of this data. And
so we could reduce the manual reviews and also increase what
we call as our efficiency in which we can catch the fraud.
Right, okay. So, when we send this transaction over to the
banks, banks do not have the kind of knowledge that we have
on the transaction, right. So our legitimate shoppers were
getting rejected. So at that particular point, we go we
created a KPI called Profit Efficiency. What Profit Efficiency
is, what is the maximum profit that we can achieve by selling a
product- is in the denominator. And in the numerator, we were
actually looking at what is the profits were actually making? We
were between 90 and 92% between the businesses. And then we
looked at it it is just not a decision sized problem. It is
also an operations problem because you need to be very
diligent, right about how to approve or deny transaction,
how do you move forward to the next party. All this innovation can
be given to our customers on Azure as an offering, right? And
that’s exactly what became a Dynamics 365 Fraud Protection.
Why it is a compelling product is because it came from
Microsoft’s own experience of processing billions of
transactions every year, right, which actually is based on OR as
well as ML. So tell me one thing, one last question for
you. This is an amazing journey all together. What is the, what
are the innovation pieces that you would say? Make Dynamics
Fraud Protection, very different from what’s that in the market? Whenever a transaction comes in,
we can tell the probability of fraud, probably very accurately.
We also have an, a connected fraud protection network, that
is- when merchants are coming in. Yeah. And their data actually,
when we look at the fraud lead behavior pattern, we bring into
an anonymized consortium. Right. And the consortium actually gets
better every day. It learns a lot more about fraud patterns,
we have a world class, what we call as a virtual fraud analyst,
kind of an offering, which looks at a merchant’s transaction, and
automatically actually recommends where is the operating point
the model should be performing at. And a business manager from
the merchant side, you can look at it and actually adjust so
that they can achieve the kind of profit efficiencies that
Microsoft has been enjoying. Last, but not the least is
we have a feature called Transaction Acceptance Booster.
As I told you earlier, in an e-commerce situation banks do
not know all the information that the merchants
knows about. And we have found a way in which actually we can
transmit this knowledge to the banks in a very anonymized way
so that the banks can incorporate this data in making
their acceptances better. Coming from the real world experience
of a merchant like Microsoft. And as you know, Microsoft is
one of the top 10 and it gets attacked quite a lot. Yep. So we
learn on an every second basis. Yep. And our models learn
and I’m proud to announce that our time to learn and
actually update a model is less than 45 minutes today. Wow, that
is amazing. So the technology itself is
pretty amazing. And just understanding that Dynamics
365 Fraud, Protection actually improves the customer
experience, decreases fraud and increases acceptance is just
unreal. So thank you so much for being here. And thank you for
the education session that we just had with you. So I hope
this episode was valuable for you as as much as it was
valuable for me. It was really exciting talking to Jay. And
going forward, we will have industry experts who will be
talking about Fraud Protection, including our own engineering
team, who will talk about innovations. Now to learn more.
And to see more of these videos, join our LinkedIn group, where
all of these videos will be available as a learning for
everyone in the industry who is impacted by fraud. Thank you so

Leave a Reply

Your email address will not be published. Required fields are marked *