I, Fraudbot: Signals and Strategy for Preventing GenAI Fraud

0:03
Right. Hello everybody. Thank you for joining us today for I fraudbot signals
0:08
and strategy for preventing Gen AI fraud. Just like every other webinar
0:13
you’ve attended, we’re going to give it a minute or so for people to join and
0:17
then we will dive right in. Right.
0:54
Thank you everybody for joining. I want to call your attention to the QA section
1:02
of the the tab for the webinar. We would love to get your questions throughout
1:06
the course of our discussion today. So do not hesitate to put them in there.
1:11
But uh without further ado, we’re going to kick off i Fraudbot signals and
1:16
strategy for preventing Gen AI fraud. And we’re going to start out with a few
1:22
introductions from today’s three speakers. So Brian, would you mind
1:27
kicking us off?
1:28
>> Thanks for having me today. Always a big fan of the stiff team. Uh thank you.
1:36
I always like to pay credit when it is due. Uh, I’m Brian Davis, head of prod
1:41
at Dodgeball. I handle a lot of the customer strategy and really helping
1:44
understand what risks are worth solving.
1:49
>> Great, Jimmy.
1:51
>> Awesome. Yeah, thanks Britney. Um, so my name is Jimmy Dunn and I am on the
1:56
product marketing team at Sift. Okay. And I am Britney Allen, senior
2:00
trust and safety architect at SIFT. I’ve been in fraud prevention for about 15
2:05
years focusing on being the voice of the merchant and also education.
2:10
So today we are going to cover three different topics. First is going to be a
2:17
deep dive into some of the data from CF’s Q2 digital trust index which is a
2:22
quarterly report that we put out and in Q2 it focused on consumers and their
2:28
perception of AI. Then we’re going to dive into some of the complicated
2:33
questions that merchants have to work through separating fraud from legitimate
2:38
bot activity. And then lastly, we’re going to focus on leveraging AIdriven
2:43
fraud tools so that you can have some actionable takeaways after the end of
2:47
today. So let’s start with consumers and AI.
2:53
Now, all of the statistics from this section come from a poll that SIFT ran
2:59
which asked adults 18 and older in the US in May 2025 some of the following
3:06
questions. So, let’s start with consumer overconfidence in identifying AI scams.
3:12
Now, of those consumers surveyed, 33% said they were completely or very
3:19
confident they’d be able to identify a scam that was created using AI. And I I
3:23
think we all know what that might look like. We’ve seen those demonstrations
3:27
where uh you’ll go to the is this person real or not website and try to figure
3:32
out how you could tell if a face was AI generated or not, or you’ve seen a deep
3:37
fake video or voice used in a demonstration. But about a third of
3:41
consumers said completely confident or very confident they could identify that
3:47
that was a deep fake. But then of those same consumers asked, 20% said, “Well, I
3:53
was fished successfully in the past year.” And we have a similar connection
3:57
with another set of questions where 33% of consumers said they have been the
4:01
target of an AIdriven scam like a deep fake. And 27% of those who said they had
4:08
been a target were scammed successfully. So you then have to read into the
4:14
questions a little bit more deeply knowing that only a third of consumers
4:19
are completely or very confident that they’ve been able to identify that a
4:23
scam was created using AI. Well, then probably way more than 30% or 33% of
4:30
consumers were targeted by AI because we’ve got 66%
4:36
who don’t feel like they could even identify AI and might not have even
4:40
known AI was used in a scam that targeted them. And then of course we
4:45
know we’ve got people who are confident in identifying it, but still 27% of
4:49
those were scammed successfully. So what we can see here is that some consumers
4:55
are confident a majority aren’t and there are successful fishing and scam
5:02
campaigns that are being run using deep fakes. So Brian, does this sort of line
5:09
up with what you would expect to see from a consumer survey?
5:12
>> Yeah, because people are still learning what can I do? There’s different
5:16
populations who are very aware of the technology and a lot of populations that
5:20
just heard about it on the news. So honestly, I bet you some of the I’ve
5:26
been targeted by AI stands realistically is probably higher uh just because if
5:31
they don’t know it really exists how it can actually be used or they’re
5:36
confident definitely definitely confident that they know how to spot it
5:41
>> but might overlook a few things cuz 27% are still scammed.
5:44
>> Yeah. Bet you it’s still a little bit higher and it’s probably used a little
5:47
bit more so than a lot of people recognize.
5:52
>> Exactly. Jimmy, how about you?
5:56
>> Yeah, the similar thing. I mean, I’ve I I think about the experiences that I’ve
5:59
had and that even just friends have had, right? I had a a good friend whose his
6:04
parents were called on the phone by one of their sons saying that he’d been in a
6:08
car accident using deep fake voice. And the only reason that they caught it was
6:11
because uh uh he said that he was in the wrong location. He said that he was he
6:17
had recently moved states uh and and was actually across the country. So that
6:22
it’s getting better out there. I’m getting some that are saying like, “Hey,
6:24
have you paid your your tolls for, you know, to you on the east coast here,
6:28
right? You got all the easy pass stuff going on and you get all those coming.”
6:32
Like they are getting even to the point where I’m going, “Oh, was that and and
6:36
pretty dialed into this stuff.” So unsurprising that that this is probably
6:41
just the tip of the iceberg.
6:42
>> Yeah. So we’ve got that that base level set then here of consumers having some
6:48
confidence but it not really being an overwhelming majority. And so that would
6:52
then lead us to ask well if they are a bit worried about Genai deep fakes and
6:58
and other tools being used against them for fraud well how are they interacting
7:02
with the tools that are built for consumers? And we’re actually finding
7:07
that they’re interacting quite a bit. So 31% of consumers admit to entering
7:13
personal or sensitive information into a Genai tool. And some examples of what
7:19
that personal or sensitive information would be are things like email
7:22
addresses, phone numbers, home address, uh all the way down to unfortunately
7:27
information that’s not yet public and company trade secrets. Uh now I will say
7:32
when surveys like this go out some people don’t want to tell on themselves
7:37
even if they had done something like that. So you could maybe assume that
7:41
information that’s not yet public or uh information about employer and company
7:45
trade secrets that actually could be a bit higher. People are just hesitant uh
7:49
to admit it because they worry that clicking that button might somehow send
7:52
out a signal somewhere or get them in trouble. But knowing that over half of
7:56
people put their email addresses into a Genai tool, that lines up with the idea
8:01
of a Gentic e-commerce, they would have to be able to turn over some of that
8:06
personal information to allow a Genai agent to make purchases on their behalf,
8:11
for example. Uh, so I will pause here also because I’m not one of those people
8:16
yet. If you’ve heard me speak on this topic before, you won’t be surprised by
8:20
that. I am very, you know, hesitant to use Genai tools, but somebody on our
8:27
call today has put personal sensitive information into a tool. And so, Jimmy,
8:33
can you tell us about how you used a tool?
8:35
>> Yeah, I mean, I’ll proudly answer honestly on this survey, I guess. Uh, so
8:40
so yeah, absolutely. I I had my annual physical uh a couple of weeks ago uh and
8:46
got back the results of my my you know the blood test where you get like the
8:51
whole list of results and I went into the uh to the tool that that our medical
8:55
provider has to like look at all the results and there was no context no
8:58
information about like what’s good there were just ranges and dots on ranges and
9:03
I was like well this isn’t helpful at all. So I I just grabbed a copy of it
9:07
and dropped it into chat GPT and said, “Hey, could you help me understand
9:09
this?” And it was super useful of like both like here’s the implications,
9:13
here’s how things are related to each other. So like and I and I think what we
9:17
run into here, there’s a lot of generational stuff both with identifying
9:20
what fraud is, but also with just comfort level with being open and
9:23
sharing things that are online. So my guess too is that this is something that
9:27
is probably higher than is stated and and will continue to be. So people are
9:31
just kind of used to it, especially if there’s actual value on the other side.
9:36
>> Yeah. There’s still a lot of questions though floating around like are are you
9:39
comfortable having the information you entered then being used for training
9:43
purposes for the model?
9:45
>> Yeah. So and that’s a good question because I I I put myself as kind of like
9:49
one foot in each boat. Like I’ll share some things like a blood test like
9:53
what’s that going to do for the world to know like you know how my triglycerides
9:57
are? I don’t know. Um, so, so with that one, yes, but there are certainly other
10:01
things that I would be, uh, a little bit more hesitant to put out there.
10:05
>> Yeah.
10:08
>> So, we break down here another uh, question that was asked, which is
10:11
percentage of consumers, excuse me, percentage of consumers who are
10:18
extremely or very concerned about the security of their personal information
10:22
when using GI tools. That’s 52%. So, you know, maybe that 52% with that weird
10:27
overlap with 55% saying they put their email in. They might have put their
10:31
email in, but they are still concerned. So, we do have a high percentage of
10:35
consumers then that at least express some level of concern. And companies
10:40
have concern over what a consumer or an individual would be willing to share. If
10:45
you’ve got that information that’s not yet public or company trade secret part
10:49
in mind, you could think about the fact that many companies have policies
10:53
against doing just that. And if you entered company trade secrets into a
10:58
genai tool, you could actually be fired per your company’s policy. And then of
11:04
course there’s that idea of data being used for training which could
11:07
potentially result into uh inadvertent disclosure and the possibility of
11:13
fraudsters who were able to access jailbroken models of publicly available
11:20
genai tools who could then maybe you know use them nefariously based on the
11:24
information that’s been entered in and is available then for querying. So that
11:28
that leaves us with a lot of questions up in the air. Uh before I move on,
11:32
Brian, is there anything you’d like to to add here?
11:35
>> Uh no, I got some ideas which I think will kind of blend into a little bit of
11:39
the sneak peek that you sampled uh quickly already.
11:42
>> Um
11:44
>> right, then let’s move on. And you’re right, we already did sample
11:49
a little bit of this because this is a visualization of how consumers have used
11:55
generative AI tools comparing 2024 to 2025. Now, I have my own takeaways from
12:01
this and then a couple of reveals that will pop up at the bottom, but since you
12:05
already said you’ve got something here, Brian, why don’t why don’t you take it
12:07
away for a minute?
12:08
>> I think this just speaks to the advancement and how people are thinking
12:12
about using these tools on the consumer side of like, hey, what can I do? Help
12:16
me with my blood test. Help me understand this. What’s this mean?
12:19
Should I do that? Very, very kind of simple level one. And now you’re seeing
12:24
a lot of competitors in the market. And I’m not even speaking on the
12:28
competitiveness on the jailbroken models, but on what they offer, the
12:33
specialize the tools and now how people are thinking about how to use them.
12:37
Naturally, Britney, I think you probably go with a lot of these questions too of
12:40
like there’s a lot of good that can be with this, but there’s a lot of bad. So,
12:44
I am a tinkerer. I like to kind of see like what can I build? What can I do
12:48
with this? I actually create bedtime stories for my son to have him live
12:53
specific moments. he gets to pick the theme, so he’s involved with that. Uh,
12:57
so that’s like one use case that I like to use so I don’t have to read the same
13:00
book 17 times over. Get a little boring for me, but he gets to be involved with
13:04
that. But on the other side of this, there’s a couple things that stand out
13:08
to me. the improving code, generating code, uh generating ideas, enhancing
13:13
learning. For me, what do those stand out on the risk side of enhancing
13:17
learning, understanding how different businesses can be abused? Why should I
13:22
target insert one company versus the other? Account research, generate ideas.
13:27
What are some things that I can do to communicate to help me with customer
13:32
service? Now you can have social engineering or messaging and writing and
13:36
then improving code and generating code. We think about triangulation fraud uh
13:41
with fake uh websites being built and brought up. The scale and ease to make
13:46
these look legitimate is so quick and easy. Now I don’t know if you’ve played
13:50
around with building any websites with any of these or any apps but it’s pretty
13:54
impressive. Like I’m non-technical. So the ability of all the different ideas
13:59
that I have that I can get at least a working model that looks way better than
14:03
what I could have gotten two years ago. Now think about that for fraud and
14:06
abuse. Think about the scale that they’re able to do. That’s why you’re
14:09
seeing all these messages that the grammar the things that we have taught
14:13
in the past of what to spot, what to look for. I think we’ll talk a little
14:18
bit on like consumer education a little bit, but
14:21
>> like I think about the risks involved with this. So my brain always goes back
14:24
and forth. I am on both sides of this. I think playing for me gives me ideas of
14:30
how could I use this to abuse a company. So then I can think about the
14:34
protections, the risks, the signals. There’s a whole lot to unpack there.
14:38
>> There definitely is. And you for me, I agree with you where generate code that
14:43
massive increase was one of the the things that caught my eye. So, in less
14:48
than a month, I’m going to be starting a master’s program in cyber security. And
14:52
signing up to do it through Georgia Tech meant that I had to agree to all of
14:57
Georgia Tech’s policies about how I would conduct myself as a student. And
15:01
one of the first things I had to agree to was not using Genai tools to generate
15:06
any part of my application. And I’m therefore assuming there’s going to be
15:11
some sort of prohibitions on using Genai tools when we come to sections where we
15:15
have to code or you know take other steps. Of course, you know, are they
15:19
really going to be able to know if I’m using it or not? That’s sort of up for
15:21
debate based on the uh sophistication of these tools, but we know that quite a
15:28
few people are using Genai to generate code because we’ve even got a new phrase
15:32
for it, vibe coding. But fraudsters have decided that it’s it’s kind of fun to
15:38
have also a play on words for them. And they have decided to call vibe coding,
15:43
which is when you don’t have that technical background, but you’re asking
15:48
a generative AI tool to do something. You’re telling it what the end product
15:52
is that you want and asking it to build the code behind that. Well, fraudsters
15:57
are starting to refer to it as vice coding. I don’t know if that’s going to
16:01
stick around or not, but that is the idea of using generally a jailbroken uh
16:06
generative AI tool to then create malware, create uh I say a fake fishing
16:12
site, whatever they want to do that a legitimate tool would refuse to help
16:16
with. So, I’ve got a couple of examples down here of conversations from Dread,
16:22
which is a site that is basically the Reddit of the dark web, where fraudsters
16:27
are discussing what they might do with their AI agents to accomplish just this.
16:31
So, in this first post here, you see someone is complaining about being busy
16:35
in the summer, but they really want to place sports betting bets. So, they’re
16:40
saying here five to 10 large bets between now and the start of the month.
16:43
They still want to look into how to have good OPSSEAC or to not get caught and
16:48
then also just security in general, but they’re going to feed info into their
16:52
offline AI agent. So, one that they do not have connected to a major provider’s
16:57
servers and then get recommendations. So, in this case, they don’t care to
17:01
study maybe who’s going to be h I’m about to reveal myself as not being a
17:05
football fan, but like who’s the best new quarterback question mark of the uh
17:11
new season, but they’re going to get their GI agent to figure that out for
17:15
them. And then to also place bets on their behalf, so they don’t even have to
17:18
go to those particular platforms. The second example I’ve got is actually
17:24
somebody replying to another post. There was a post where someone said, “How can
17:30
I use Gen AI or AI agents in order to better fish credentials out of victims?
17:36
They were specifically looking to get credit card numbers.” A person replied
17:40
saying, “How much coding do you know? You know, it won’t be that hard just to
17:44
use AI to do it, but you know, if it’s complicated, you could just pay a dev.”
17:48
And so they’re acknowledging that there are limitations to this vice coding or
17:52
vibe coding uh you know let’s say strategy but we know that won’t be going
17:57
anywhere because historically we have seen script kitties or fraudsters who
18:03
just find code available online and then reuse it for their own purpose. This is
18:07
basically the same thing as being a script kitty just without finding
18:10
someone’s human pre-written code asking the genai to do it. So that’s that’s my
18:16
big take on this information here as a fraud and risk indicator.
18:25
Right. Before we go on then into separating out these bots from good or
18:31
bad, is there is there anyone else who wants to make a comment?
18:38
All right. So let’s go right into it. Separating fraud from the bots. So let’s
18:44
start with going back to that Q2 digital trust index from SIFT fraudster identity
18:51
signals. What have we found within our own data that seems to be indicative of
18:57
fraudulent activity? We highlight this because when you are a fraud fighter,
19:03
especially one that’s been working for 15, 20 years, any length of time, you
19:09
get this mindset of what fraud looks like to you. And as that changes because
19:15
fraudsters change their methods, you might still be stuck in relying on old
19:20
indicators or, you know, you may be hesitant to change your mind. And I
19:25
think some of these are, you know, really interesting to discuss because
19:29
they are different from prior conceptions that we may have had. So,
19:33
for example, 36% more payment methods are used by fraudsters than by
19:39
legitimate consumers. That one makes sense to me. Fraudsters using stolen
19:44
credit cards, other stolen payment methods. Of course, they would need to
19:48
cycle through those until they find a card that works. So, that makes sense.
19:53
But on the flip side, fraudsters use 20% fewer IP addresses than legitimate
20:00
users. And and I think a lot of fraud fighters can easily get stuck in that
20:05
mindset of well the more changes to an account the more variances then that
20:11
makes it riskier. Whereas in this instance, we’re pretty much pointing to
20:17
fraudsters who maybe have a stationary device where they aren’t often changing
20:22
their IP address, standing out in comparison to legitimate consumers,
20:28
walking around with a phone, walking around with a laptop or a tablet,
20:32
anything else portable and connecting from a lot of different points. Then
20:37
we’ve also got the call out here for peak hours for fraudsters, which again
20:41
sort of fits the the mindset of what I would have as a fraud fighter of them
20:45
attempting to hit your site during off hours. So I want to defer at this point
20:50
to Brian and say, you know, with with your experience as a merchant,
20:54
conversations you have as a merchant, what’s your feedback on these signals
20:58
here? And, you know, how do you think that actually applies to a day-to-day,
21:04
you know, life of a fraud fighter? Do you think they get hung up on some
21:07
historical patterns or you know what’s your take?
21:10
>> Absolutely. And it’s easy to like I know these patterns but they come and go.
21:15
They might disappear for a couple months but they’ll be back. They always come
21:19
back. Maybe a little bit of a different twist of a flavor and you have these
21:23
general trends and people know when you work. So they you think about
21:28
seasonality. Usually peak season for physical goods is around the holidays.
21:33
for gambling. You brought up gambling. It’s usually around the Super Bowl. So,
21:37
it’s when when is there a lot of pressure where I can kind of either
21:42
blend in with normal activity. Either I know companies are not staffed enough,
21:47
they’re on vacations, or I can take advantage of what I know about the
21:53
target that I’m trying to get. And then thinking about what am I really trying
21:56
to gain access there. Um the fewer IP addresses is one of those things I think
22:01
a lot of people kind of think about like well if there’s a variance there there’s
22:05
this that it’s an anomaly where you’re looking for anomalous behavior. Now,
22:10
when we’re pointing to non-anomalous behavior, that starts to make things a
22:14
little bit trickier of like what you’re thinking about and the way that
22:19
everything we’re talking about here are signals. Signals and data points
22:23
ultimately say what do we trust? What do we not trust? And I think that’s a big
22:27
piece of Gen AI moving forward and AI agents is there’s a lot of unknown and
22:33
what are the signals? Where do I get the signals? How should I consume these
22:38
signals? are there going to be variances to the signals that we know today. So I
22:43
think there’s a lot of curiosity into uncovering a lot of these signals but a
22:47
lot of hesitance of is it going to change everything I know is what got me
22:51
to this point to make me the expert to people think I am is that going to all
22:55
fall through so there’s like a little bit of like in some of the conversation
22:58
I having like from people that are incredibly smart and built awesome
23:02
careers little bit of imposter syndrome going through of like everything that I
23:06
know is what’s it going to look like moving forward. So it all goes back to
23:09
the signals and kind of what the unknown is because we’re all still kind of
23:13
figuring this out. Merchant side, vendor side, even AI companies are all still
23:18
trying to figure out what will it look like not even two years from now, three
23:21
months from now, six months from now.
23:23
>> Yeah. And that’s just another important point of not relying on these static
23:26
signals and you know being able to if you’re using uh machine learning to
23:30
fight fraud, to update the models to learn to follow those frauds for
23:34
patterns and to better prevent fraud. Uh how about you Jimmy? What what are any
23:38
takeaways you’ve got from these here?
23:41
>> Uh I actually didn’t have any specific ones um for this slide, so I think I
23:45
think I’m good on this one. We can keep going.
23:47
>> Okay. Well, actually, no. I I like to put you on the point or on the spot. I
23:50
think it’s a little fun. Do you connect to public Wi-Fi networks?
23:54
>> Do I connect Oh, uh uh occasionally. Like the the occasional
23:59
like airport Wi-Fi. I’ll do that on occasion, but for the most part, no.
24:03
Unless there’s a password.
24:05
>> How about you, Brian? Let’s get it all out there.
24:08
>> I’ll say absolutely. For the most part, no. I try to avoid you. But I will say
24:11
there are those situations where I’m like, I just need it for whatever it
24:14
need needs to be to connect to something. So yes, I’ll admit it. I will
24:18
be the first one to admit. There are scenarios. Do I try my best to avoid it?
24:24
Yes. Uh but not perfect. I do. I do sometimes.
24:29
>> Look, I I I wouldn’t shame you without also then contributing my own answer,
24:33
which is also that I do. So, there’s simply too many situations I find myself
24:37
in where connectivity is needed for one reason or another, and my only option is
24:41
a public Wi-Fi network. Uh, most recently, I was at an event where I was,
24:47
I guess, I don’t want to say in the basement, but I was sub, you know, level
24:51
of a building where I still needed to be able to use my phone. And so, I had to
24:57
get on their Wi-Fi network because I had no cell signal down there. So, those
25:00
things still still come up. But there’s an idea then that if you were a very
25:04
security-minded person and maybe you traveled with your own router and you
25:09
had your own connection, maybe you would fall then into the few IP address
25:14
category or fewer than a majority of consumers and you might look like fraud
25:18
when really you’re just a privacy-minded person. So I call all of that out then
25:22
to say we still can’t rely on, you know, one signal, two signals to indicate
25:28
fraud or not. It’s a glimpse over the whole user journey.
25:32
But thank you for admitting that. And I do want to also throw out the anecdote
25:35
that one time I was at JFK here in New York and there were two networks that I
25:41
could see that were related to airports. One said something like JFK free Wi-Fi
25:46
and the other said LAX free Wi-Fi. And I knew I wasn’t in LAX, but if that
25:54
person had been smart enough to switch it over to say JFK, I might not have
25:57
known which JFK to pick from. So of course there is still that risk.
26:02
All right. So, let’s talk about assistive technology. Now, this is
26:08
important to bring up because a significant amount of your customer base
26:13
may use assist of technology and assist of technology is going to look very much
26:18
like bot activity or automated activity on your platform. So, just to throw out
26:23
some stats of of how big this is, 20% of people worldwide use assist of
26:28
technology. uh to give you example what that could be could be speech to text.
26:32
It could be eyetracking technology in lie of using a mouse to click on a site
26:37
and that encompasses 13 million people in the UK and 65 million people in the
26:43
US. So using tools like speech detects and eyetracking technology is a benefit.
26:50
It’s something that we want to encourage and support to make the internet more
26:54
accessible. But if all of a sudden merchants who let’s say haven’t been as
27:00
concerned about automated activity or were only concerned about automated
27:04
activity that they already knew how to recognize as fraud and are now dealing
27:09
with incoming agentic AI activity. there’s a chance that they could then,
27:14
you know, make a mistake and potentially discriminate against individuals using
27:19
technology like this because now these types of agents are, you know, just for
27:24
convenience such as with AI powered browsers. Uh, and I know Jimmy, when we
27:29
were chatting earlier, you actually had something uh that was just, I think,
27:33
released this week or discussed this week from OpenAI.
27:37
>> Yeah. So, this is a big deal. Um so chat GPT is still for most people it is AI
27:43
and it is like the consumerfacing uh like they had they’re up to 800
27:48
million active weekly users. So when chat GPT does something it means that it
27:54
has kind of officially launched if that makes sense. And this week, ChatGpt
27:58
launched agent mode, which is where, right, you can use chat GPT to actually
28:03
go and take actions on the web for you that can go and do visit sites and do
28:09
things on your behalf. Uh, and it’s really interesting because if you read
28:12
through the release, there’s a section called novel capabilities, novel risks,
28:17
right? We’re talking about like the the Spider-Man scenario, right? With great
28:20
power comes great responsibility. When we talk about AI, with new capabilities
28:24
comes new risks. And this is maybe the most upfront I’ve seen them be on a
28:29
release about the types of risks that are coming here. We’re taking a new
28:32
step. We’re talking about like if you’re giving payment data, for example, to an
28:36
agentic AI to go and make purchases for you, that data is now potentially
28:40
compromised as it goes out and moves in the world. And they do call out
28:43
specifically prompt injection, which if you’re not familiar with that, that is
28:48
as your AI agent is going out and reading websites. Uh fraudsters could
28:53
potentially put instructions for the AI on a website that the the chatbot would
28:59
go and read and that would be taken in as a new instruction. So it’s it’s
29:03
taking over it’s adversarial manipulation taking over the the bot as
29:07
it is out um completing tasks for you. So getting into the world of a bot being
29:12
able to actually do real world things now opens up a new category of the the
29:18
type of shenanigans that can occur.
29:22
>> Shenanigans is a good word. Brian, anything you’d like to add to this?
29:27
>> I don’t know how I talk shenanigans. I think it’s a great way to kind of
29:29
explain it. It’s really goes into a lot of like the methodology of like when I
29:35
think about researching into it of when these are used on your platforms, how do
29:40
you know? I think that’s a great place to start kind of today of like thinking
29:43
about how do you know and what signals are telling you? And a lot of it will
29:48
kind of come from metadata and there are logs that you’re able to kind of revisit
29:51
and there’s the visual side of things and then there’s the data side of
29:55
things. So that’s really where I’m starting today
29:59
as I kind of have a lot of conversations of helping people understand how do you
30:04
know and where will I find this information. You can’t really build a
30:07
strategy and execute on it if you don’t know where it is and what kind of
30:11
signals you have today to start. Now there will be lots to learn from. You
30:16
got to start somewhere.
30:18
>> Exactly. Uh, and I think that leads us really well into our next slide of
30:22
talking about blah versus the blah bot. So, my initials are bla or blah. Thank
30:29
you, mom and dad. And so, I as an example for comparing human activity to
30:36
an agent activity, we’re going to talk about blah, the human, and the blah bot,
30:41
the agent. So, here’s four different potential scenarios that you as a
30:45
merchant could see on your platform. If you’re looking at a, you know, user
30:51
journey from start to finish in number one, I have never bought from your
30:56
platform before, but I’ve asked an agent, an AI agent to do it for me. And
31:00
so, BLBOT makes the first purchase. Now, I’m a legitimate consumer. We’re not
31:04
talking about fraud here. It’s using all of my legitimate information that I’m
31:08
providing to it. In number two, I’m already familiar with your platform as a
31:14
human and I have made at least one prior purchase, done some browsing activity,
31:18
saved things to favorites, whatnot, but then it’s followed by a purchase from
31:24
blah bot. Maybe I was then, you know, just comfortable enough to use a an AI
31:29
agent because I already like your service and I say just repeat my last
31:33
order. So, you get that activity. On the flip side, in scenario three, have you
31:38
thought about if the blah bot is the one that makes the first purchase and then I
31:43
liked the item that I got so much or I liked the service so much, I then when I
31:48
had time as a human came back to your platform and I walk through and make the
31:53
second purchase or four, which is when me the human experience with multiple
32:00
different AI agents because I want to see which one I like the best and Maybe
32:05
I am making four or five small purchases through your platform. Each one using a
32:11
different bot. So, it is all automated activity coming from the virtual machine
32:17
the bots’s running through or whatever signals you’re able to scoop up from
32:20
this activity, but it’s multiple different bots. Are you able to say that
32:27
looking at the signals from these four scenarios, you would be able to tell
32:31
that there’s no fraud here? that all of these four scenarios represent
32:37
legitimate customer activity. Uh I will click to reveal another quote from
32:42
Dread. This is a very dramatic fraudster here saying humans are corruptible and
32:46
then saying as AI agents take over there will be fewer signals or traces as he
32:52
says to point to for fraud but there’s also going to be you know fewer human
32:57
signals if the adoption of AI agents like this increases. So, I I have a
33:03
couple of takeaways, but I think I’ll I’ll pause here and ask Brian, what do
33:09
you think about these four scenarios? I know there could be even more, but I’ve
33:13
got a lot here, I think, for merchants to think on and try to discern whether
33:19
their system would be able to correctly identify all of this activity as
33:24
legitimate.
33:25
>> I want to call out this quote that it feels like it’s the start of a trailer
33:29
to a movie. goes to a black screen that pops up
33:33
>> cinematic music and then the it takes off. After that,
33:38
>> we’re living in the movie now today. But I think about it, there’s some news of
33:43
recently. It starts at the policy level. There’s the major online retailers
33:47
who’ve decided today we’re not taking anything we determine that are agentic
33:53
AI agents coming to our site to shop. That’s a policy that you’re deciding as
33:58
a company where we fall within all this. I’ll say my personal biases are it’s a
34:04
hybrid approach. It’s not a this or that and it shouldn’t be today because we
34:07
don’t do that with humans because a lot of humans are good. There’s some
34:11
malicious actors who are doing it manual and then there’s the gray area that we
34:16
as the experts live in through our entirety of a career. Same thing happens
34:20
with bots. Uh not to forget just bots that we’ve known today. There’s some
34:25
good bots, there’s some bots we don’t really know, and there’s some malicious
34:28
bots. The same thing goes with AI agents. So each one now, as the fraud
34:34
and risk experts that we are, we need to have a human strategy, a bot strategy,
34:39
and an AI agent strategy. There can be a lot of overlap, but it all starts with
34:43
the policy of where do we stand with what we want to allow or not allow on
34:50
our website. So for me, it starts at the policy level. you you’ll start to see
34:54
probably over the next couple months other companies making some of those
34:58
headlines of we’re not allowing it at all and some companies saying we love it
35:02
we’re going to find a strategy to drive them to our site. There’s going to be a
35:06
whole lot of that coming over the next couple months starts with policy where
35:09
you as a merchant want to land.
35:12
>> Well before we move on into some of those recommendations and to talk about
35:16
you know what is in the works to assist merchants I wanted to ask a question of
35:21
our audience. Now, I could have asked this up top, but there’s also
35:24
potentially people in attendance today who kind of weren’t quite sure what uh
35:29
AI could actually be doing from a a fraudster angle. But I would love if we
35:35
could get some human clicks on this poll question, which is, have you seen a
35:41
spike in AI based payment fraud? So, thinking through everything we’ve talked
35:45
through today, uh before we get to those
35:48
recommendations, is this something you’ve seen? Yes. No. And it’s okay to
35:53
click I’m not sure if you’re not sure because maybe uh you’re not able to yet
35:58
discern some of this activity. I I think that was really useful what you said a
36:03
couple slides back, Jimmy, about the idea of even the legitimate agents that
36:09
are initiated by a consumer being taken over through I think you said it was
36:15
like code injection or otherwise something malicious in that process and
36:20
then used by a fraudster just like man-in-the-middle attacks collecting
36:24
information work today. So, you know, not trying to influence anybody,
36:30
but I think we’ve got Okay, we’ve got well over half of people. So, I can end
36:37
the poll and share the results. I hope I didn’t influence too many people to
36:42
click the I’m not sure. I don’t think I did because I was watching it go up
36:45
early as before I even began to speak to that section. But we’ve got 53%
36:51
saying I’m not sure, 37% saying yes, and 11% saying no. So, with that, uh, Jimmy,
37:00
what do you what do you think about the responses to this poll? It’s a it’s a
37:05
it’s an interesting mix and and I think that that what’s going to be happening
37:08
as we talked about upfront too with AI is that this this amount of I’m not sure
37:14
makes all of the other answers a little bit probably directionally accurate but
37:18
also a little bit difficult um to to to gauge. And the term, by the way, is
37:23
prompt injection, right? Like prompts are the are the codes for an AI. It’s
37:27
what you tell an AI to do. And prompt injection is taking your prompts as like
37:32
a little virus and sticking it into somebody else’s AI um bot that is that
37:37
is taking action. So interesting to see that that either people aren’t sure or
37:43
it is growing. Just a few that are are positive that nothing is happening.
37:52
So take that away and we will move then into leveraging AIdriven fraud tools. So
38:00
the the origin of the traffic, you know, establishing your known uh Jimmy, I’m
38:05
going to hand this over to you to kick us off.
38:08
>> Yeah, absolutely. So, so I want to start with some some basics first when we’re
38:13
talking about how do we combat this new world of of AIdriven fraud that is uh
38:19
evolving really really quickly. Um, and when you when you start, you want to
38:23
start with the fundamentals. You want to make sure that your building blocks are
38:26
in place. And when I think about that, I think about global data and think about
38:33
making sure that you’re getting data from a vast cohort. Um, so that you are
38:38
seeing changes in uh fraud behavior from across industries, from across
38:43
geographies, and that it’s something that that the model is seeing and that
38:47
is adapting to really quickly. Um, so that’s why at SIFT we have our global
38:50
data network which has a trillion events annually. um we have 220 million plus
38:55
signals that are being added into that model uh by professionals such as
39:00
yourselves to make sure um that we’re staying uh ahead of what’s happening
39:04
there. The other piece which is something that that we work on
39:08
continuously is making sure that our that the modeling that we’re using what
39:11
we’re doing with that data and how we’re slicing it is as effective as possible.
39:16
Um so this uh we have this thing called ensemble tuning or having an ensemble
39:21
set of models that we used for our customers. Um so we have the global
39:25
model that we just talked about on the left side. In the yellow we talk about
39:28
custom models which is building a model just based on your data. And this is a
39:32
little bit of like a a Goldilocks scenario. Global model lots of data. You
39:36
see lots of things not very tightly tuned to your type of fraud. custom
39:40
model hyper tuned to your fraud because it’s just your data but you don’t get
39:44
that visibility from other places and something that we added uh about a
39:48
little over a year ago um that we call threat clusters which is basically
39:52
cohort modeling based on your vertical um so we’ve created a series of threat
39:58
clusters uh in travel and e-commerce and fintech and other verticals so that
40:03
you’re getting that kind of perfect view of uh I’m getting a lot of data that’s
40:08
not just me so that I don’t have to react to fraudsters for the first time
40:12
every time and I’m getting uh data or information that is tightly tuned to the
40:17
type of fraud that is common in my industry. Now there is no silver bullet.
40:21
You can’t use any one of these alone. So we actually use all three of them in
40:24
balance to then create the most accurate uh scoring that we can. So, so having
40:29
that foundation of data and the data that comes in um and the data that comes
40:34
not just from your own sources is key to being able to do things like identity
40:40
trust where you understand who is coming to you and what their intention is when
40:45
they come to um your sites. Now I’m going to give a few uh one very specific
40:51
example of how we are leveraging AI a tool that we are creating and then uh
40:56
talk a little bit about some uh some other things that that uh you can think
41:00
about in the space of of where those tools should be.
41:05
So this is a tool that we are actually excited about. We’re launching it uh
41:09
right now. We’re starting a a limited roll out for GA and this is for our ATO
41:15
um customers, right? Account takeovers are some of the trickiest fraud to
41:20
identify because there’s a lot of gray space. So when we look at one of the
41:24
major pain points um for atto it’s that atto investigation can take anywhere
41:29
from 5 to 30 minutes because you’re sorting through sessions, you’re sorting
41:33
through users, you’re trying to draw connections, you end up with like the
41:36
like beautiful mind board with yarn going to a bunch of pegs on a string and
41:40
then finally you get the whole picture and you’re like yeah I’ve got them
41:43
right. It’s it’s one of the most difficult things even for really
41:46
experienced uh fraud analysts and this is a perfect use case for Genai for
41:52
taking a lot of potentially related data and looking for connections. So, we’ve
41:58
built what we call activity IQ, which is an active Genai um bot that is looking
42:04
at through the data related to these atto investigations. And when you go in
42:08
and look, it will pop up um a summary for you of key data points that could be
42:13
related to the investigation as well as other related sessions or individuals
42:18
and uh uh aberrances of of things that are unusual or outside the norm that you
42:24
might want to look at. So it takes what would be a five to 30 30 minute session
42:28
of of real manual work and it really cuts it down to to giving you the key
42:34
insights and allowing you like somebody’s already done all the all the
42:37
string and the and the dots and you just get it back up and look at it and go
42:40
okay what’s my conclusion here maybe I want to dig into this part or that part.
42:44
So a great example of using Genai for doing intelligent work that’s going to
42:50
make you better at your job and faster. And
42:57
then uh I’ll tee you up, Brian, on this next one here where we just talk a
43:02
little bit about some of the the the variety of ways that you can use AI. And
43:08
AI, I know for some folks can can mean both Gen AI, that’s what most people are
43:13
thinking of now when we talk about AI. It’s also includes broader machine
43:17
learning, which Sift has been doing for the last 14 years. Um, but there’s a lot
43:21
of different things in here that you can dive into. Uh, like like in intelligent
43:27
rule builders where you can speak in normal language and you can have an
43:31
agent, a bot, an AI help you um come up with, okay, well, here’s how you should
43:37
tune a workflow. Here’s the the suggested package of things that you
43:42
should should put into your score to your workflow to help get to that
43:45
outcome that that that is desired. Investigation efficiency. We just talked
43:48
about that with ATO. Score explanability is not just getting back numbers and
43:53
terms, but also getting back real natural language around um some of those
43:58
kinds of things that that or or around the the data that you’re seeing so that
44:02
you can draw um so that you can draw good conclusions. And and as we flow
44:07
down, we get kind of more into that monitoring, pattern recognition, anomaly
44:10
detection, the stuff that uh that bots can be good at at a broader scale that
44:14
would take human eyes and and hands and minds a little bit more time to kind of
44:19
pull together. Brian, why don’t you jump in here? Yeah, I think there’s an
44:23
interesting So, I’m going to actually talk about a use case of one time when I
44:26
was trying to actually buy the ATO SIFT product inhouse about seven years ago
44:31
and it was really hard for me specifically, not because I didn’t get
44:35
the value, I got the value and that’s where I think a lot of the AI can
44:38
actually help here. I was a fifth customer about seven years ago. I wanted
44:43
to expand to ATOS’s ATO was a huge issue for us. The problem for me was ATOS’s
44:48
were happening. We were recurring subscription products. So like what are
44:52
they doing on our platform? For me it was really hard to ultimately tie enough
44:56
of a business case to my exec team to say this is a no-brainer. Instead it was
45:01
like well you already have the transaction model and I was drowning in
45:06
a backlog knowing I had a solution that would work wonderfully. And I think a
45:10
lot of these AI tools of what you broke down here is an area to that would have
45:14
helped me tremendously. you know, from the explanability to understand the
45:18
impact, the anomaly detection, all of this not only helps you fight fraud, but
45:23
if you’re leading the program helps you show the value, the scoreboard, the wins
45:27
of everything that you’re doing and making it easy. So, I don’t have to put
45:30
it in a jury ticket to say why I need a data scientist to help me pull the
45:34
reporting on something. I know exactly what I want to explain to get what I
45:37
need. And then it’s just always lagging. And then when you’re battling actually
45:42
under pressure from fraud attack, you’re just lagged. And then that’s how we
45:47
always end up in being reactive of just we’re waiting for the lagging indicators
45:52
ultimately to figure out what broke and then fix it. These are just so many
45:56
different levels and I think this can sometimes even be overwhelming of okay
46:01
Jimmy this is a lot. These are all wonderful ideas. The way that I kind of
46:05
start to kind of like put the boundaries around people of like you got to start
46:08
somewhere. So, is it about analyzing the impact of your rules to help you create
46:13
better rules? Is it in the investigation summaries of the ATO? Is it score
46:18
explanability of really understanding why did Brian get caught up because he
46:22
was on public Wi-Fi because he signed into LAX public bit instead of at and he
46:29
was at JFK. Something doesn’t really make sense here. Uh so there’s lots of
46:33
different areas of opportunities where I help people is I kind of help them
46:37
understand the risk surface area and how their program is running and we look for
46:41
different areas for this to start and saying what can I what are the goals
46:47
like really why are you thinking about AI like when people are say hey Britney
46:51
what’s your AI strategy that’s fully loaded well I’m a little hesitant I
46:55
don’t really want to be the first one to adapt I need to see some use cases so
46:58
there are some areas that you can use AI into your strategy that might be a
47:04
little bit easier to build some confidence personally for you and your
47:08
own reputation internally and then build that into something more comprehensive.
47:12
So when there are these wonderful tools like this that then kind of like I think
47:15
about the signals you talk about thread clusters. So all these different signals
47:20
is a lot for let alone one person and a team to monitor. There’s just some
47:25
different ways to get higher output per employee. It’s not necessarily always
47:29
about cutting the team, but getting higher output, better leverage, and
47:34
really using that human intelligence where human intelligence should be spent
47:38
and then complementing that with artificial intelligence. So, I think
47:41
about it like this hybrid approach. What you have up on the screen right here are
47:46
incredible examples of where to start, but you got to start somewhere.
47:53
>> So, that then will take us through to these three takeaways. And we have had a
47:58
couple of questions come through and we’ll take the time to answer them on uh
48:02
this slide. But first of all, what do we want you to take away from this webinar?
48:07
It’s got these three points. Consumers are wary of sharing sensitive data with
48:11
AI agents but are becoming more comfortable doing so. So only expect to
48:16
see an increase in that. Number two, you separating good bots from bad will be
48:21
vital to maintain customer trust. You don’t want to discriminate against a
48:26
population using assist of technology or against somebody using an AI agent to
48:31
pick out the uh birthday present for their niece. And then third, be vocal
48:37
with your fraud providers about developing AI tools and giving feedback
48:42
on them. If you want to see these tools available to you as a fraud fighter and
48:47
not just the fraudsters, speak up. Say what those use cases are that you want
48:52
help with. And when those tools are developed and become available, give
48:56
your feedback, help hone them, I I know I’ve been in the fraud fighting world
49:01
for over 15 years, and there’s very few times in that history when there have
49:07
been big shifts in the tools that are available and technology changing. This
49:12
is one of them. And so, it’s a great opportunity to speak up. Now, with that,
49:18
turning to some of the questions. One of them is a clarification question. So I
49:23
think we can start off with that. And we had a person ask what is the difference
49:28
between AI and a bot. So those are both automated systems. AI being a subset of
49:36
machine learning. But I I think what we were talking about through this webinar
49:40
was mostly we use the term bot and we use the term AI agent. So even though AI
49:46
agents and bots are both automated systems, uh when I was trying to make a
49:51
differentiation there, I was talking about bots as meaning sort of the old
49:56
the classic use of a script or code, let’s say via an open bullet config, for
50:02
example, to then take actions on a platform, but not really be intuitive
50:07
about it. So fraudsters would have a config that might be able to access
50:13
somebody’s email and then search their email account for specific names of
50:18
let’s say banks because this fraudster specializes in cash out fraud could then
50:23
see this individual how many bank accounts or or financial accounts or
50:28
whatever they’re looking into investment platforms this person had and then be
50:31
able to target them. Whereas, if they were using something like um an AI agent
50:38
that was able to be plugged into an email account, they wouldn’t have to
50:43
specify the names of those banks of those investment platforms, but they
50:47
could instead just say something as general as, you know, find all my emails
50:52
with financial and crypto, etc. companies that I have accounts with and
50:56
then you know summarize you know recent activity summarize balance summarize
51:00
whatever via that tool and that’s more intuitive and can give the fraudsters
51:06
better access to that data and of course those AI agents are going to be used
51:12
commonly now by legitimate consumers whereas maybe that what I described as
51:16
bot activity earlier wouldn’t be but we could definitely see a time when it’s
51:21
really common for people to use an AI agent to go through their email address.
51:26
So, they’ve connected it to say, “Hey, what subscriptions do I have that I
51:29
haven’t canled and other useful actions.” So, that was what I attempted
51:33
to do. I do apologize if I wasn’t clear with those terminology, but that’s kind
51:37
of how I was trying to separate out sort of the old more restrictive way of using
51:41
a configure code and then now what’s available to fraudsters. So, with that,
51:47
let’s dive into some of our more general questions. Uh to start with, is there
51:52
any way to help consumers get better at identifying deep fakes and other Gen AI
51:59
content related to scams? I got a lot of opinions on that, but I’m going to see
52:03
who unmutes first for letting Brian or Jimmy speak.
52:07
>> Jimmy,
52:08
>> I’ll jump I’ll jump in here. Um, so, so yes, I think that and one of the the
52:14
other I think it was in a recent survey that we saw, I don’t know if it was this
52:17
one where where it talked about um there’s like a social network that’s
52:22
happening like the Tik Tok of fraud, right? Like where people are learning
52:26
from each other how to commit fraud. Uh, one of the ways I think that that and
52:32
this is maybe something that’s a little bit less influencable by us, but people
52:35
starting to share what fraud they’ve seen, sharing what the types of attacks
52:40
that they’ve seen, the types of things that are working. And I know that I’ve
52:42
gotten that from like friends. I’ve sent things to my parents because I’m worried
52:46
about them even though they’re fairly savvy just to be like, “Hey, this is
52:49
something that I saw. It’s not actually you’re not actually late on paying your
52:53
toll bills, you know, that kind of thing.” um as well as getting examples
52:58
from like my bank um that says like, “Hey, here’s some of the types of fraud
53:02
that we’ve seen in the last six months.” I know I’ve started to get those types
53:05
of emails more and more and more often, which I find to be helpful, but I also
53:09
pay attention to that stuff.
53:15
>> Brian, anything you want to add for?
53:17
>> You and I have a lot of similarities. I know we are talking uh about this. I
53:22
think it’s just due to the scope of our role. I’ll let you I’m not going to
53:25
steal it from you. I’ll let you go and if there’s anything that I want to add
53:28
on, I’ll piggy back.
53:30
>> Okay. So, earlier um we we had been chatting. I I did bring up the idea of
53:36
uh consumer education being really important, but it also still just falls
53:40
on us as merchants to be able to protect our consumers. We can’t assume that
53:47
they’re going to be always educated to the point of being able to identify
53:51
these scams. the scams are going to change. They’re going to become more
53:54
sophisticated. You’ve also got newer users online, especially if you allow
53:58
minors on your platform. You can’t assume that they’ve got that experience
54:01
and the maturity and the knowledge to be able to identify these scams. And so it
54:06
also then just falls on us to use some of the tools and strategies that we
54:10
discussed today to be able to identify where someone is taking action
54:16
on a platform that is potentially indicative of them having fallen for a
54:19
scam or somebody’s credentials are being used. Think back to that uh email
54:24
example that I shared of finding all the financial accounts. It it still does
54:28
fall on us for detection. Did I say everything that we had chatted about
54:33
earlier?
54:34
>> Absolutely. And then I just want to tie back to an idea that Jimmy had said
54:38
earlier. Um, you said something that stood out to me, can’t assume. And Jimmy
54:43
said something earlier about the intention of the user in the platform.
54:47
And both of those combined obviously are very hard to execute on. It’s really
54:53
hard to understand intention. You can watch behavior, but intention and
54:57
behaviors are different. So really going on the fact that you just said we can’t
55:03
assume people know and a lot of us fall within the space because we want to help
55:08
and protect people kind of the altruistic nature of working within
55:12
fraud and risk. So always leading with that within your northstar of like
55:16
carrying a little bit more of the conservative protection side of things
55:20
on the merchant bit of like how can I help others naturally we all have that
55:24
northstar when you’re thinking about things like this especially in this next
55:28
era of what Gen AI will bring don’t forget that northstar
55:32
>> yeah and and I personally feel completely confident that I can identify
55:36
AI because I don’t trust anything and nobody is measuring my false positive
55:40
rate as a consumer so I’ll just trust nothing and then get it all. All right,
55:45
with that we have time for one more question and it is about uh fraudster
55:50
identity signals. So think back to that conversation on the IP addresses and the
55:53
the credit cards. Uh won’t they just keep improving their activity to mimic a
55:58
real customer to the point where these trends aren’t helpful? Yeah. Okay. Will
56:02
there be a bot or an an AI agent, whatever term we we want to use, that is
56:09
just so close to a customer’s activity as to not give us useful signals.
56:17
>> I can start with that. I think the blobot was a great exercise to kind of
56:21
start thinking about, yes, they’re going to improve. Yes, they’re going to test
56:26
what gets through through. They’re going to test your protections and controls.
56:31
Yes. To all of that, but with all of that, they leave breadcrumbs of how
56:34
they’re testing. They leave constraints of what they have. They only have so
56:38
many devices. They only have so many of these like cornerstone identity signals
56:43
before they run out. So, you got to think about the constraints that these
56:47
individuals have and these broad operations have. There’s there still
56:51
operate within boundaries and constraints. So that’s one piece of
56:54
where can I attack them on their constraints to get a better
56:57
understanding of how did they get through how did they improve why did
57:01
this happen again and ask those basic questions the other thing is I think
57:05
about like we have a session replay where we’re able to actually watch the
57:09
behavior and the actions of what is happening on platform. So actually
57:15
connecting the data, the logs with the visualization
57:19
uh calling out your football reference uh from the beginning of like going to
57:23
the booth to watch the action replay. That’s what I think about sometimes of
57:27
like if I just can’t really tell and I need to know a little bit. I’m a visual
57:31
learner myself. So it helps me a lot in investigations to be able to start to
57:35
connect the data says one thing. what actually happened on platform might tell
57:40
a little bit of a different story or have just enough context to give me
57:45
additional breadcrumbs that I didn’t know I have. So, um kind of going into
57:49
like always asking those questions. Why us? How did it happen? Where did this
57:55
start? What is their intention? What are they trying to get out of us? And that
57:59
will start to give you leading indicators and lagging indicators to
58:03
help kind of perform. It’s the game we all play. is the game we’re all used to.
58:08
Yes, they’re going to improve. Yes, we’re going to improve as well.
58:12
>> One thing that I’ll just add to that, I mentioned it a little bit earlier um and
58:15
it’s up on my background. Uh this uh concept of identity trust. Um I compare
58:20
it to uh speed dating, right? We’re all speed dating all the time where we have
58:24
somebody sitting in front of us and we have to make a quick decision on whether
58:27
or not we’re going to trust that person, take that payment information, send that
58:30
product or accept the subscription or whatever it is based on the model. We
58:33
got to make quick decisions about trust right away. Well, wouldn’t it be nice if
58:38
before making that decision, you could talk to that person’s exes or the other
58:42
people that they have been in relationships with just to get a feel
58:45
for for for what they’ve done in the past and for what what do they like?
58:48
What’s their what’s their vibe? Um, and that’s what identity trust does. It
58:52
actually collects information related to uh specific identities. Um, and and it
58:58
shares it through the global data network. I can see this person’s
59:01
actually been had transactions accepted and none of them have been marked
59:04
fraudulent by six different vendors over the last two years or over the last six
59:08
years on our network. And those kinds of signals, those behavioral signals over
59:13
time are very expensive and very difficult to fake. They may eventually
59:17
get there, but we’re talking several years down the road. And in the
59:20
meantime, we’ve got some leverage that we can use to win.
59:25
>> Well, with that, uh, one more question came in that says, “Can Britney tell us
59:28
how she identifies fraud?” And I think that was calling me out as how do you
59:32
trust nothing? I don’t have time to talk into it today, but you know what? That
59:35
sounds like a good idea for a little video to put together maybe to share on
59:38
LinkedIn. So stay tuned. I’ll I’ll share some of those tips. But with that, we
59:42
thank you very much for attending our webinar I fraudbot today. If you have
59:46
any further questions, if something springs to mind as soon as you leave the
59:50
session today, do not hesitate to reach out to any of us. We’d be happy to
59:54
answer those questions. Uh, and with that, I hope everybody has a great rest
59:59
of your Thursday.
1:00:01
>> Thanks, everybody.