Latest / Leaders in Lending / Cashing in on AI: Streamlining Bank Operations

Transcript

Cashing in on AI: Streamlining Bank Operations · Leaders in Lending

0:02
You are listening to Leaders in Lending from Upstart, a podcast dedicated to helping
0:07
consumer lenders grow their programs and improve their product offerings. Each week, here
0:13
decision makers in the finance industry offer insights into the future of the lending industry,
0:18
best practices around digital transformation, and more. Let's get into the show.
0:26
Welcome everyone to another episode of Leaders in Lending. I'm met snow here
0:30
with Will Robinson from and Capture. Will Welcome to the show. Hey,
0:33
Matt, thanks for having me on. Absolutely Maybe let's get started by a
0:37
little introduction, talk about what you do and Capture what they do, and
0:41
how you got to the role of CEO there today. Yeah, it's kind
0:45
of a wild journey. But in Capture we are a software company in the
0:49
intelligent document processing space, so that means we have artificial intelligence that can read
0:56
documents and understand what type of document is looking at and then can go find
1:00
important data that in those documents and pull it out automatically. We work in
1:07
a bunch of different use cases across a bank, mostly around the kind of
1:12
loan origination and processing space for mortgage, commercial, consumer, indirect auto.
1:22
We do stuff in the branch as well, but essentially any workflow where there's
1:26
a lot of incoming paperwork and there's important data in those documents, we can
1:30
come in and really automate the manual data entry type tasks and the manual stare
1:36
and compare. That is a prevalent in the back office of many financial institutions.
1:42
Gotcha, Yeah, it makes sense. And I know, as we've
1:44
talked a lot of similarities that we've had. You've been on both sides now
1:49
in a bank and supporting banks on the service side. What's your experience been
1:53
like on both sides of that you compare and contrast, like, what are
1:57
the similarities and differences you've seen? Yeah, uh, you know, it's
2:00
there's a lot of paperwork in banks. And it's funny. We have a
2:04
famous story around our business, uh where we had a newly appointed chief digital
2:13
officer of a bank, of a big bank, and we had a meeting
2:15
with him and he confidently said, you know, my mandate is within the
2:20
next twenty four months, we're going to have no more paper around the bank.
2:24
This thing is going to be completely digital. That conversation was twelve years
2:29
ago, and needless to stay, there's still plenty of paper flying around and
2:35
that chief Digital Officer no longer is in his role. It did not last
2:38
very long, so uh, I think it's it's a it's a it's an
2:42
interesting dynamic in our in our industry where banks are trying to become more digital,
2:49
more automated, more efficient, but yet you've got legacy processes. You
2:53
have a lot of regulations that drive collecting, uh collecting documents from your borrowers
2:58
and your customers. Uh. And there's a lot of compliance uh that that
3:02
still revolves around having a physical or even a digital copy at someone's driver's license
3:07
or a bank statement or a pay staff And so how do banks kind of
3:13
figure out ways to uh move into the next next, the next century,
3:17
uh, but also uh satisfy these these demands on them. So it's a
3:22
pretty unique. It's it's a hard thing relative to other industries. I think
3:24
it's hard for banks to try to figure out how to how to navigate all
3:28
this. Yeah, a lot lots we could dig into from there, for
3:31
sure, especially like I said, the requirements around gathering uh documentation. I
3:36
know we see that could be different to the state down to the county level.
3:39
Like how do you guys think about again, you think banking and regulation
3:45
and government like being two very legacy driven, slow to change, Uh,
3:51
areas like how do you manage across all that different nuance and process. Yeah,
3:58
that's a good thing about about AI these days and as it's evolved.
4:01
Uh, there is a lot of variance, and there's these you know,
4:05
legacy regulations. Uh, every every local jurisdiction has a different way of doing
4:11
things, Every bank has a different way of doing things, and a lot
4:13
of the way, a lot of the policies at the bank are kind of
4:15
entrenched, and uh, trying to completely overhaul that is is probably kinney,
4:23
It's probably too hard for most banks and lenders. And so instead it's how
4:27
do you use new technology? How can you use AI to really augment your
4:31
existing workflows? You're not going to try to change the whole paradigm of the
4:35
bank or the regulatory landscape under which you you operate, but can you use
4:40
AI to automate some of the most kind of mundane, boring, routine tasks
4:45
that that you've had to do in the past, And so, uh,
4:47
it's it's interesting I hear you know, there's obviously a lot of talk around
4:50
right now about AI This is something that we've been doing for years and years,
4:56
so we're not we're not new to it, but we chuckle because,
5:00
uh, finding like really practical use cases is still a challenge for a lot
5:04
of a lot of banks as well as a lot of technology partners that want
5:09
to sell AI stuff into into a bank. You have to show up with
5:13
a really proven, uh kind of uh really ROI effective use case, and
5:18
that's that's hard for a lot of folks to do. Yeah, I want
5:21
to dig into some of those use cases because I don't know if you've experienced
5:25
this as well, and kind of getting the sense that maybe you have that
5:29
when when banks talk about, you know, becoming digital or making their process
5:32
digital, they're just taking that same like tired, cumbersome offline process and putting
5:39
it on the internet. So they're not you know, thinking about how can
5:42
we use these new technologies tools with AI to make it easier for the consumer
5:46
or even internal staff, and it's just digitizing, uh you know, the
5:50
same process. How do you think about are those things different or similar?
5:55
What's what's yours? It's really important at that, you know, and we
6:00
this a lot when we start out with the bank and we start talking about
6:02
implementing our own product and capture to help streamline, you know, a specific
6:08
workflow. We have to get really specific. What is it that we're actually
6:11
doing at the bank, What are we actually replacing? What steps are you
6:14
guys doing today that either you're not going to have to do anymore or you
6:17
may have to do them different. Uh. That change management process is probably
6:21
one of the biggest risk points for a bank that's trying to adopt AI,
6:29
UH, because if it does, you know, for it to work,
6:31
it requires you to maybe think about changing your process and you can't just necessarily
6:35
always layer it on the existing way you do things. A lot of times
6:39
you should try to start with that, but uh, it will require maybe
6:42
a different way of thinking about a specific process or people's roles and responsibilities.
6:46
And uh, you know, we've seen this unfortunately on our side, where
6:51
you know, we'll come in, we'll build a really good solution that that
6:55
can save that bank a lot of time, a lot of money. Uh,
6:58
But then there's so much resistance to changing a process to accommodate this this
7:02
this AI that like they don't get much value out of it. And that's
7:05
where we have to have these hard conversations on it's not just the technology that's
7:11
going to show up and solve all your problems. You have to be willing
7:14
to change. You have to be willing to adapt, and you know,
7:16
we're going to try to make that as easy as possible. But it's it's
7:19
very much a mindset as well as you go into a you know, kind
7:23
of an AI specific project. Yeah, that's really true about the change management
7:29
of all of this, because, like you said, you want to get
7:30
really specific on the use cases and but not make it a complete overall of
7:34
a system, because then you kind of get paralysis analysis. So you don't
7:39
know where to start if you think you have to rip and replace an entire
7:42
process or system. So I'm sure you know, understanding the entire landscape of
7:46
a process and then digging in on those use cases is probably pretty critical to
7:49
getting folks excited about starting on that journey and then keeping going. Well,
7:55
and it's it's you know, it's funny too. I guess this is human
7:58
nature, but people, you know, people get very comfortable with the way
8:03
they do things. And we had a situation recently where we had built a
8:07
solution to help a bank spread all the tax returns that they gather as part
8:15
of their small business lending process. So you know, a small business lender,
8:20
a banker at the bank is going to work with a barrower and they're
8:22
going to be sending in tax return documents instead of the bank. Now,
8:26
instead of these credit analysts or these processors having to manually go through the tax
8:28
returns and enter all the income information, we do that automatically. We can
8:33
read through the documents, extract the data, we put it in the system.
8:37
And we were talking to one lady and about her team, and I
8:41
said, hey, how's it working. She goes, Oh, your software
8:43
work's great, but nobody's using it. And I was like, well why.
8:46
She said, well, they don't like skipping this step. They really
8:50
like manually entering all the data because they say it quote unquote helps them understand
8:54
the loan better and get to know the loan and get to know the borrower
8:58
and they can kind of think as they go along. And I kind of
9:00
threw my hands up and laughed. I said, I don't know if I
9:03
can do anything about that, and she laughed too. She said, no,
9:05
it's it's hard to get people to think about changing their behavior, especially
9:09
if it's becoming grained in the way that they work, in the way that
9:13
they make decisions. And so we talked about a couple of kind of practical
9:16
ways to maybe let these people trust the system to do it correctly and they
9:22
can go back and review later. But that's that's a hard thing, you
9:24
know. And I would you know, if you're if you are kind of
9:28
an exec for a lender and you're thinking about, you know, implementing new
9:33
technology, whether it's AI or anything else, I think it's good to honestly
9:37
ask yourself, do I have a team that is willing to change, embrace
9:41
change, try things differently, or can I go find maybe a few folks
9:45
that can be leaders in this area and can kind of be the advocates for
9:48
the change, Because if you if you're not willing to change or really push
9:50
that and push your team to change, whatever you implement, regardless of technology,
9:56
it probably is not going to work very well. Yeah, that's true,
10:00
And I think that's a really good story, really specific example, but
10:03
maybe highlights some things that we could dig in more broadly around AI. As
10:07
you said, like being able to trust us any system really is probably key
10:11
to adoption and AI for better or worse. I think it brings with it
10:18
a reputation of being a quote unquote black box or something harder to understand.
10:22
So maybe we could start a little bit just like getting more specific around the
10:26
termin as you've used it a couple of times, like when you think about,
10:30
you know, what a capture does and how you guys use AI,
10:33
like what does that specifically mean, or what are some of the approaches that
10:35
you're taking to make make your bags easier. It's funny, Matt, we've
10:41
been using AI and specifically what we use as supervised machine learning. I can
10:46
explain what that is in a second, but we've been using this for like
10:50
the last four to five years, and you know, up until about a
10:54
year and a half ago, we didn't even say the words AI or say
10:56
machine learning because it would just scare everybody. UH people have these preconceived ideas
11:03
of what AI means, what it is when it isn't what it does at
11:05
the bank, and we just felt there was so much barrier, there was
11:09
so much kind of pre existing UH biases against it, that we really referred
11:15
ourselves more to kind of an automation technology or automation platform, and it wasn't
11:20
until chat GPT about eighteen months ago kind of entered the public lexicon and everybody
11:24
kind of got comfortable talking about AI and realized, hey, for better or
11:28
for worse, I'm gonna have to figure out a way to use some flavor
11:31
of AI going forward. That's really allowed us to speak I think a lot
11:37
more candidly about it. And so, uh, you know, I won't
11:39
get in all the weeds and all the types of the AI that are out
11:43
there, but you know, there's there are several different uh strategies and techniques
11:50
of using artificial intelligence. That's a really really broad term. Uh. You
11:52
know what we do around supervised machine learning is we are it's almost like we
11:58
have you know this this for that almost behave someone like like a fresh college
12:01
grad that's maybe in their very first job, and they're smart, they're willing
12:05
to learn, but you really have to teach them what the job is and
12:09
what they're expected to do. And so in our use case, you know,
12:11
we have trained our machine learning models to you know, recognize certain types
12:18
of documents. So we'll feed it samples of documents, and we'll feed it
12:20
a bunch of different driver's licenses and say hey, these are driver's licenses.
12:24
And then we'll feed a bunch of samples of bank statements and say hey,
12:26
these are bank statements, and all the while the system is looking at these
12:30
documents and looking at maybe some of the similarities and some of the differences between
12:33
the document types, so that you know, one day I can show up
12:37
with kind of a mystery document, give it to the system, and they
12:39
can look at it and say, Okay, is this a bank statement,
12:41
is this a driver's license, or is this something else? And you know,
12:46
the more that you could train them and teach them and give them hints,
12:50
give them rules, give them reminders, or reinforce kind of positive decision
12:54
making the better it is at kind of I would say, guessing the right
13:00
the right outcome in terms of what documents looking at or the data it's looking
13:03
for. So that's a really kind of simplified version of how our our machine
13:07
learning works, our AI. That's actually very different than what chat GPT is
13:11
as you as you guys have heard around the generative AI. That's a that's
13:16
a different type of AI than instead of kind of guessing an answer uh to
13:20
a specific question based on previous training, this AI is more about creating new
13:24
content and trying to figure out what's the next word that I should say in
13:28
the Senate so that the sentence is complete and makes sense. Relative the question
13:31
that was asked, so use cases there for generative AI are are are a
13:37
bit different than kind of the AI we use for reading documents. And I
13:41
think it's important for you know, someone who's looking to bring on AI to
13:45
understand, Hey, tell me how your AI works. Tell me how much
13:48
of a black box it is, because you know, in our case,
13:52
we have kind of intentionally designed ra AI to where it's less of a black
13:54
box. In some ways. You would say, well, maybe it's not
13:58
as effective. I would argue, well, we at least know how it's
14:01
making the decisions it's making, and so every time it makes it a decision,
14:05
we can point to kind of the way it went about that versus it
14:09
being some just magical thing that comes up with an answer and we don't know
14:13
how it gets there. So that's a big you know, that's a big
14:16
part of it. And then the other thing is, especially for banks and
14:18
lenders credit unions, how are you know, where are you using AI in
14:22
your organization? It's one thing if you're using AI like ours, which is
14:26
just really kind of automating these Monday tasks, you know, doing replacing manual
14:31
data entry. It's different if you're trying to use AI to make a credit
14:35
decision, or you're using AI to have a conversation with a customer or provide
14:41
kind of a more native kind of conversational onboarding experience for a customer. Those
14:46
all have kind of their own unique risks and challenges. And again for lenders
14:50
specifically, there's a lot of regulation that would say, hey, if you're
14:52
going to put AI in to make a decision on the loan, you better
14:54
have a lot of confidence that you understand how that system is making that decision
15:00
so that you can prove it's not discriminating against certain types of borrowers. Hey
15:05
there, former host Jeff Kelviner here to let you know about an exciting opportunity
15:09
to strengthen your understanding of artificial intelligence and financial services. If you want to
15:15
better understand the fundamentals of how AI models work and how they will impact the
15:18
financial services industry, your in Luck Upstart is launched an online course and certification
15:22
designed to do just that. This course will give you the knowledge you need
15:26
to lead the conversation about how your organization can best leverage and manage this revolutionary
15:33
technology. If that sounds interesting, check out the course as it's currently free
15:37
for a limited time. You can learn more at upstart dot com slash AI
15:41
Certification. Again, that's upstart dot Com slash AI Certification. Thanks and now
15:46
back to the show. Yeah, totally. I think every application, we
15:50
keep coming back to the idea of use cases, but defining those use cases,
15:54
understanding the risk and trade offs are definitely important. I think when you're
15:58
embarking on this kind of a journey or deciding if an AI application is right
16:03
for a certain use case and really understanding uh, the benefits you get there
16:07
And you didn't you mentioned compliance too. I think sometimes you know, folks
16:11
maybe view that as a more adversarial relationship, or different levels of government will
16:18
come out with their perspective or recommendations on use of AI. But we talked
16:23
before about you know, how how can you use AI more of a proactive
16:27
way or being more compliant or helping uh, you know, stay within regulation,
16:32
regulation and guidelines. How have you guys thought about that and in ways
16:36
that you can actually use this technology to help you maybe show up better in
16:41
those cases. Yeah, it's it's funny. There has been this growth in
16:48
UH regulations for banks and lenders and how you know, UH, there's more
16:52
and more data that has to be reported to the government to make sure that
16:55
you're not discriminating in your lending practices. There's recently been a revamp of you
17:02
know, the CRA rules, which have been around for fifty years, but
17:04
those have expanded. There's this upcoming you know Dodd frankedin seventy one rule.
17:10
And I realize I'm saying a lot of acronyms here, Matt, but I'm
17:11
sure your listeners will know what these are. There's just this growing body of
17:15
reporting data to the government. And you know, what we've found for our
17:19
use case is there's actually a really there's a really good solution here. A
17:23
lot of times these compliance teams are having to go review you know, loan
17:30
packages to make sure that the data they've pulled out of their system that they're
17:33
about to report to the regulator is accurate and complete, and so they spend
17:37
a lot of time hunting and pecking through big loan package PDFs, you know,
17:42
double checking various data points against the system. It's very time intensive and
17:48
kind of cumbersome process. Most teams can't even check every loan. They don't
17:52
have the staff or the time to do that. So we'll just sample a
17:55
few loans. And that's a great use case for AI. And again we're
17:59
not you know, it's it's it's not something where we're manipulating data and reporting
18:03
without your you know, it's not this black box scenario. It's simply saying,
18:06
okay, AI, I want you to go through all of my loan
18:11
packages and I want you to pull out the data that I need to report.
18:12
I want you to compare it to my system, and I want you
18:15
to flag any discrepancies. And you know what we found is ninety to ninety
18:19
five percent of the data points that are being reported, they're accurate, they're
18:23
consistent, and you don't need to do anything with them. But maybe it's
18:26
five or ten percent of your data points are they're missing data fields or it's
18:30
a different name. You know, it should say Will Robinson, but instead
18:33
it says Matt Snow and you need to go figure out why that is.
18:37
And so it cuts down a lot of this kind of mundane data scrubbing activity
18:42
that these compliance teams do. So I think that's kind of an interesting thing
18:45
about you know, you hear about compliance and banking and regulations going to restrict
18:51
AI. Ironically, banks are using AI to facilitate their obligations under the regulations.
18:57
So, you know, I think that's kind it's kind of been an
19:00
ironic twist of AI. AI helping banks even though it's going to get cracked
19:06
down on I think more and more. Yeah, I think you're right.
19:08
I think that's a good example too of where, especially with human resources that
19:12
are very expensive and limited, like, can you use a combination of these
19:17
two tools, right the AI that can help you focus and prioritize or find
19:22
higher risk signals. I think we see that a lot in terms of you
19:25
know, verification in underwriting as well, Like you're seeing tens of thousands of
19:30
applications in any day, which ones do you focus on, where do you
19:33
put the human capital against that? And where should you focus on? Types
19:38
of documents or types of fraud? And being able to partner those two together,
19:42
each with their unique skills and either handling volume or you know, novelty
19:47
in the data I think can be a great win. Yeah, you know,
19:52
We get asked this question a lot, and you know, it's like
19:52
every conference you go to, that's a big question, how do I use
19:56
AI my bank? And I kind of chuckle because I had not heard many
20:02
people show up on these panels and give like three or five or seven or
20:07
even you know, ten good answers, Like usually there's kind of one use
20:10
case or two. And I think it's because sometimes you hear about all the
20:15
potential of AI and it's hard to figure out where to start, and there's
20:22
so much hype around what it could do that I think people miss like what
20:29
are some really really practical things that can do today? So we always just
20:32
encourage people, and again this is unique to our business a little bit,
20:33
and the fact that we you know, we process, we process documents.
20:37
But go look through your existing workflows and look at where people are spending a
20:41
lot of time doing repetitive tasks that are that are highly rules based. Right,
20:48
you probably have a rule or a process because in those situations you can
20:51
use technology to automate that. It doesn't even have to be AI. So
20:55
much of what a load origination system is, which doesn't been around forever,
20:57
is really just making sure you're applying your bank policies and bank rules around a
21:03
process and around an underwriting set of decision making. So that and there's no
21:07
AI in that. So I do think it's an It's important you know that
21:12
you first take stock of do we have tight, repetitive processes that can be
21:18
automated. The flip side of that is, don't try to go use technology
21:22
or AI until you've got that part down. We struggled with this a lot
21:26
with some of our customers. We will go they will be so excited about
21:30
using AI. Oh, you know, you're going to automate our loan origination
21:33
process and you're going to help, you know, make sure that we're following
21:37
all these rules and we're collecting all the right documents. And we're like,
21:40
yeah, that's great. Tell us your process, tell us your rules so
21:41
we can make sure the AI is following your rules. And then quickly you
21:45
realize they don't really have clean rules. They don't have consistency on what the
21:48
rules are. We did. This was a couple of years ago, but
21:52
we rolled out a solution based on what the project team at the bank had
21:55
said. These are our rules and this is how we process loans, and
21:59
so we we started testing it with actual users and they revolted. They were
22:03
like, this thing is terrible, and we said, well, why is
22:06
it terrible? And then we uncovered it's because they could not agree internally on
22:08
what even the rules were or the standard processes were, and so they thought
22:14
that, you know, it was AI's fault. And we were like,
22:15
hey, we we can make this do anything you want, but you have
22:18
to have consistent standards at the bank that you can follow. Otherwise this is
22:22
not going to help at all. So I think starting with kind of,
22:25
you know, making sure you have clear process, making sure that if you're
22:29
going to be using data, your own data to help with an AI enabled
22:33
project, making sure that data is clean and well structured is really important.
22:37
Again, this happens a lot with us. We look at we look at
22:40
documents, and so a lot of times we'll go into a bank and say,
22:42
hey, pull us samples of these documents so we can make sure our
22:47
machine learning model can read it. Well. They don't have their documents safe
22:49
down very well, and they don't know where they are, or they have
22:52
a miss misclassified or misindexed, and so we start with a really dirty sample
22:57
set and it just doesn't go well, So, uh, there's a there's
23:00
a lot to do here on the AI front. But I think just taking
23:03
a step back and looking at hey, what are the what are the most
23:06
mundane, routine, repetitive things that my team's doing every day that's taking a
23:10
lot of time and maybe it's not very much fun and how can I go
23:12
automate that? And they'll typically be a good use case there. Yeah,
23:17
that makes sense. So as if you were advising somewhere, you know,
23:19
at a bank or a credit union, thinking about automation use of AI,
23:26
how do you think about the you know, bill versus buy or what are
23:29
the different dimensions? Is it like, like you said, the type of
23:32
data that I have, or maybe for a third party would have access to
23:36
the skills I have natively, you know, and evaluating each of these different
23:40
use cases that might exist. How would you advise someone to think about,
23:44
you know, is this something we should try to tackle on our own or
23:45
start looking for help outside. I think it's really a function of how big
23:51
is your bank, how big is your budget, and what are you trying
23:53
to do. I do think the biggest banks, you know, the JP
23:57
Morgan's of the world, they have the budget, in the in the appetite
24:02
to kind of build as much as they can in house. At the same
24:06
time, we work with some of those those biggest names, and what we
24:10
found is the problem that we're solving for them is really important to the business,
24:15
but from like a from an IT or product perspective inside the bank,
24:19
it's not very it's not very important. So the bank's not going to get
24:22
to it even though they could build it there, it's going to take them
24:26
a year or two to get there, and so the business is like,
24:27
hey, I need this solve now. So you know, I would say,
24:32
if you're like a line of business leader at a lender, I think
24:36
just being honest with yourself and with your team on maybe my bank says they
24:38
want to build it, or they say they can, or they say they're
24:41
working on it, but when's it really going to be ready? Versus if
24:45
I can find a vendor that can solve this, and what's the cost of
24:48
that and what's the timeline up getting that in, I think you'll you'll realize
24:51
even if the long term plan is for the bank to build it themselves,
24:56
I think you can probably get pretty far with the vendor up early on.
24:59
So in terms of by most banks, I mean just cutting to the chase,
25:02
most banks are ill equipped to build technology internally. Just most banks are.
25:08
They're just not set up for that. It's not in their DNA,
25:10
it's not their culture. They don't they don't have success with it. So
25:15
again, most banks, it's better to partner with vendors who've done this for
25:18
a long time, and again, do do really good diligence on your on
25:22
your vendors. Ask for referrals, Uh, make sure you know that that
25:26
they've got a long body of evidence of this being successful. Ask for a
25:32
lot of details around their process and how their technology works. A good vendor,
25:37
who's who's been doing this a while, we'll have no problem sharing that,
25:40
and we'll have good documentation around that. If you feel like you're working
25:44
with someone who's maybe trying to figure this out alongside you, that's okay,
25:48
But just be aware of that that they're trying to figure it out alongside you,
25:51
and it may not. It may not be go as well as you
25:52
as you're hearing in the sales process. So you know, I think there's
25:56
gonna be more regulation around this as well. We were talking about that earlier,
26:00
Matt, kind of you know where where's regulation going around this? The
26:04
state of Colorado just passed a law requiring organizations that use AI and these high
26:11
they call it, these kind of these high risk or high impact areas to
26:17
have a lot better governance around what is the AI doing, what data is
26:21
it touching, what is it doing with that data? How is it making
26:23
decisions? How is it impacting the outcome of the bank, is it discriminating?
26:27
Is it being biased towards certain things? So I think more and more
26:32
regulation is going to come out on that. And again, vendors like ourselves,
26:34
we you know, we we have answers for all that, and so
26:37
making sure that your vendor understands what's going on, I think we'll give a
26:41
lot of confidence. Yeah, the transparency and trust I think is key.
26:45
And bringing back to what you said earlier about you know, maybe the LMS
26:48
and chat GPT bringing more awareness to the space in general. Have you seen
26:53
your partners, you know, asking harder questions, getting into deeper diligence or
26:57
are you still having to do a lot of education on you know what what
27:00
some of the terminology is, the mechanisms you're using in this space to help
27:06
demystify some of that. You know, it's a it's a wide spectrum.
27:11
Most I will say this, most banks are doing a good job that when
27:14
they hear AI kind of their the red flag goes up and they ask more
27:18
questions, like, this is not just a routine. I'm not just buying
27:21
you know more more you know, telephones for my office desks at my you
27:25
know, at my company. This is this is something I need to have
27:29
a better understanding up. And so we do a lot of education around that.
27:33
And again, AI is such a broad term. There's so many different
27:36
techniques that, like, what I tell you about how our stuff works does
27:40
not mean every other AI vendor out there is going to work the same way.
27:41
In fact, most of them won. And so getting a good understanding,
27:45
making sure that your vendor can explain it crisply and simply, because you're
27:51
gonna have to go turn around and explain it to your bosses. What is
27:52
it this we're putting in, How does it work? How do we make
27:55
sure it won't you know, run a foul of any of our fair lind
28:00
regulations? How am I gonna get people to buy into it? And so
28:03
I think if you can understand how it works simply, then you'll have more
28:07
confidence that will work well. Yeah, I agree. And one of the
28:11
things I've found is while it's certainly important, you know, pre contract and
28:15
as a one time event, it's also an ongoing exercise because you know,
28:19
companies like ours, I think, are constantly trying to find the next new
28:25
innovative approach or like you said, AI is so broad that the way we're
28:30
leveraging those types of technology is changing day by day. So it's really a
28:33
part of the ongoing relationship. I think if you're using a third party to
28:37
stay in touch with them and how they're evolving their thinking and kind of it's
28:44
like ongoing diligence of the relationship. Absolutely, I mean it's a it's a
28:48
never ending If you find the right vendor, there's there's typically a good chance
28:53
they can do more than just one thing for you. And if they can
28:56
do one more than one thing, that means they're gonna be sticking around for
28:59
a while and they want to do more than one thing. So I think
29:00
it's important when you're not just looking at the technology or the specific use case,
29:04
but talking about vendor and saying, hey is this a type of partner.
29:07
Have they been around a while, are they going to be around?
29:10
Are they really committed to my industry, MySpace, my use cases? Do
29:14
they want to do more with me? Or am I going to be kind
29:15
of a one trick pony where they just show up, do their thing and
29:19
then leave and they never hear from me and collect a connector collector renewal check.
29:25
I think getting a good sense of that's important because you can probably go
29:27
find, you know, just a handful or less of good vendors that can
29:33
do a lot for you across the bank, and you'll make a lot more
29:36
progress sticking with those lenders and are sticking with those vendors and going deep than
29:41
trying to find thirty different vendors who kind of each do one thing for you.
29:45
Gotcha? Well, maybe it's a good time to transition to a closing
29:48
question for you. As I was thinking about our conversation today, I was
29:52
thinking back to a book I read around this, Stuart Russell's Human Compatible.
29:57
I'm not sure if you've read that, but always trying to find whether it's
30:00
a podcast a thought leader book. I thought that one was really interesting just
30:04
around the idea setting up controls and transparency around human and AI interactions. And
30:11
while it was enlightening around AI and some of the topics we've talked about,
30:15
it helped me think about even you know, human decision making and maybe processes
30:19
we take for granted or that are more automatic. So I highly recommend that
30:23
if you haven't read it, but curious if there's anyone you follow in this
30:26
space to again kind of stay relevant and keep your thoughts sharp in this this
30:32
area, whether it's a podcast or you know, maybe a writer that you
30:36
follow. That's a good question, Matt. I would if I mean very
30:40
transparent. I feel like I'm following twenty or thirty people right now and I'm
30:42
still trying to figure out like who is who is real and legit in this
30:48
because it's and I also think kind of the diversity of thought in this phase
30:52
of AI. It's so early. You have well, I think I'll give
30:57
you an example. I think more Andresen he's a prominent venture capitalist. He
31:02
wrote a really good kind of long form essay on AI and how we should
31:07
be thinking about it, wrote it several months ago, so I do encourage
31:10
people to read that. But he does talk about the concept of the Baptists
31:12
and the bootleggers. Relating this back to the prohibition, that there are people
31:17
who are going to really call for the the regulation of AI because they feel
31:23
like in their hearts that's the right thing for the world. But then there
31:26
are people who are going to call for the regulation of AI because they're the
31:29
ones who are going to benefit the most from it. And so I think
31:32
in this early phase of AI and how we use it, I think diversity
31:36
of thought is really important so you can hear kind of the different sides and
31:41
then feel confident, you know, making your own conclusions on what does AI
31:45
mean for us as a as a world, What does it mean for me
31:48
in my business, what does it mean for you know, my own professional
31:51
development going forward. I did read an interesting quote recently, and I can't
31:57
remember who said it, but it is essentially, you shouldn't worry about AI
32:01
replacing your job. You should worry about a person who knows how to use
32:07
AI replacing your job. And so I think, I think there's sometimes this
32:10
fear that AI is going to replace everybody, and I think there will be
32:15
pockets where AI comes in and can automate some of the things that people do,
32:20
but I don't think those are the most interesting jobs. And I think
32:22
annially is going to create a lot more opportunity for a lot more people and
32:25
be a net benefit, just as we've seen with every major technological advancement over
32:30
the history of the history of me and kind. So maybe a conversation for
32:34
another day we can have, Matt, but I encourage you read, read
32:37
diverse thoughts and continue to kind of keep an up in mind on where this
32:42
all goes. Yeah. I couldn't agree more on that, will. I
32:45
think that's really it's a really exciting time to be in industry as these tools
32:49
are coming on board, and I think, like you said, the transparency,
32:52
that openness in the dialogue. Let's find ways to test and see what
32:57
works and be open about the pros and cons and see where this goes.
33:02
So yeah, I'd love to follow up with a conversation on that and appreciate
33:06
your time with me today and thanks again, Well, yeah, thanks so
33:10
much for having me. Matt really enjoyed it. Upstart partners with banks and
33:14
credit unions to grow households and expand consumer lending through its leading AI lending platform.
33:19
Upstart powered banks and credit unions leverage AI to offer higher approval rates and
33:23
experience lower loss rates while simultaneously delivering the exceptional digital first lending experience that consumers
33:30
demand. Whether you're looking to grow and enhance your existing personal and auto lending
33:35
programs, or you're just getting started, upstart can help. Upstart offers an
33:39
end to end solution that can help you find more credit worthy borrowers within your
33:44
risk profile with all digital underwriting, verification, loan closing, and servicing,
33:50
It's all possible with Upstart in your corner. Learn more about finding new borrowers,
33:53
enhancing your credit decisioning process, and growing your business by visiting upstart dot
33:59
com slash Lenders. That's upstart dot com slash lenders